diff --git a/frontend/src/components/map/AreaPane.tsx b/frontend/src/components/map/AreaPane.tsx
index 35b2fd1..dddca81 100644
--- a/frontend/src/components/map/AreaPane.tsx
+++ b/frontend/src/components/map/AreaPane.tsx
@@ -619,7 +619,10 @@ export default function AreaPane({
                               />
                               {crimeSeries && crimeSeries.points.length > 1 && (
                                 <div className="mt-2">
-                                  <CrimeYearChart points={crimeSeries.points} />
+                                  <CrimeYearChart
+                                    points={crimeSeries.points}
+                                    latestAvailableYear={stats?.crime_latest_year}
+                                  />
                                 </div>
                               )}
                             </div>
@@ -663,7 +666,10 @@ export default function AreaPane({
                                     }
                                     chart={
                                       crimeSeries && crimeSeries.points.length > 1 ? (
-                                        <CrimeYearChart points={crimeSeries.points} />
+                                        <CrimeYearChart
+                                          points={crimeSeries.points}
+                                          latestAvailableYear={stats?.crime_latest_year}
+                                        />
                                       ) : (
                                         numericStats.histogram &&
                                         (globalHistogram ? (
diff --git a/frontend/src/components/map/CrimeYearChart.tsx b/frontend/src/components/map/CrimeYearChart.tsx
index 82f7a68..b518c21 100644
--- a/frontend/src/components/map/CrimeYearChart.tsx
+++ b/frontend/src/components/map/CrimeYearChart.tsx
@@ -1,14 +1,22 @@
 import { useEffect, useMemo, useRef, useState } from 'react';
+import { useTranslation } from 'react-i18next';
 import type { CrimeYearPoint } from '../../types';
 
 interface CrimeYearChartProps {
   points: CrimeYearPoint[];
+  /**
+   * Latest year available in the crime dataset as a whole. When the series
+   * ends earlier, the area's police force stopped publishing (e.g. Greater
+   * Manchester since mid-2019) and the chart is captioned as stale.
+   */
+  latestAvailableYear?: number;
 }
 
 const PADDING = { top: 6, right: 4, bottom: 14, left: 4 };
 const HEIGHT = 48;
 
-export default function CrimeYearChart({ points }: CrimeYearChartProps) {
+export default function CrimeYearChart({ points, latestAvailableYear }: CrimeYearChartProps) {
+  const { t } = useTranslation();
   const containerRef = useRef<HTMLDivElement>(null);
   const [width, setWidth] = useState(0);
 
@@ -97,6 +105,11 @@ export default function CrimeYearChart({ points }: CrimeYearChartProps) {
           </text>
         </svg>
       )}
+      {latestAvailableYear != null && yearMax < latestAvailableYear && (
+        <p className="mt-0.5 text-[10px] leading-snug text-amber-700 dark:text-amber-400">
+          {t('areaPane.crimeDataEnds', { year: yearMax })}
+        </p>
+      )}
     </div>
   );
 }
diff --git a/frontend/src/components/map/DeckOverlay.tsx b/frontend/src/components/map/DeckOverlay.tsx
new file mode 100644
index 0000000..f453aae
--- /dev/null
+++ b/frontend/src/components/map/DeckOverlay.tsx
@@ -0,0 +1,67 @@
+import { useEffect } from 'react';
+import { useControl } from 'react-map-gl/maplibre';
+import { MapboxOverlay } from '@deck.gl/mapbox';
+
+interface DeckWithPrivateDraw {
+  _drawLayers?: (
+    redrawReason: string,
+    renderOptions?: { viewports?: unknown[]; [key: string]: unknown }
+  ) => unknown;
+  __propertyMapNullViewportPatch?: boolean;
+}
+
+function patchNullViewportDraw(overlay: MapboxOverlay) {
+  const deck = (overlay as unknown as { _deck?: DeckWithPrivateDraw })._deck;
+  if (!deck || deck.__propertyMapNullViewportPatch || typeof deck._drawLayers !== 'function') {
+    return;
+  }
+
+  const drawLayers = deck._drawLayers.bind(deck);
+  deck._drawLayers = (redrawReason, renderOptions) => {
+    const viewports = renderOptions?.viewports;
+    if (viewports) {
+      // Split-route startup can hand deck.gl a transient null viewport before MapLibre has sized the map.
+      const nonNullViewports = viewports.filter(Boolean);
+      if (nonNullViewports.length === 0) return;
+      if (nonNullViewports.length !== viewports.length) {
+        return drawLayers(redrawReason, { ...renderOptions, viewports: nonNullViewports });
+      }
+    }
+    return drawLayers(redrawReason, renderOptions);
+  };
+  deck.__propertyMapNullViewportPatch = true;
+}
+
+class SafeMapboxOverlay extends MapboxOverlay {
+  onAdd(map: unknown) {
+    const element = super.onAdd(map);
+    patchNullViewportDraw(this);
+    return element;
+  }
+
+  setProps(props: Parameters<MapboxOverlay['setProps']>[0]) {
+    super.setProps(props);
+    patchNullViewportDraw(this);
+  }
+}
+
+export function DeckOverlay({
+  layers,
+  getTooltip,
+}: {
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  layers: any[];
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  getTooltip: any;
+}) {
+  const overlay = useControl(() => new SafeMapboxOverlay({ interleaved: true }));
+
+  useEffect(() => {
+    overlay.setProps({
+      layers: layers.filter(Boolean),
+      getTooltip,
+    });
+  }, [overlay, layers, getTooltip]);
+
+  return null;
+}
diff --git a/frontend/src/components/map/HoverCardOverlay.tsx b/frontend/src/components/map/HoverCardOverlay.tsx
new file mode 100644
index 0000000..87a6057
--- /dev/null
+++ b/frontend/src/components/map/HoverCardOverlay.tsx
@@ -0,0 +1,45 @@
+import { memo } from 'react';
+
+import type { FeatureFilters, FeatureMeta, HexagonData, PostcodeFeature } from '../../types';
+import HoverCard from './HoverCard';
+
+interface HoverCardOverlayProps {
+  x: number;
+  y: number;
+  id: string;
+  usePostcodeView: boolean;
+  data: HexagonData[];
+  postcodeData: PostcodeFeature[];
+  filters: FeatureFilters;
+  features: FeatureMeta[];
+}
+
+/** Resolves the hovered hexagon/postcode row from the loaded map data and
+ * renders the hover card for it. Memoized so the row lookup only reruns when
+ * the hover target or the underlying data actually changes. */
+export const HoverCardOverlay = memo(function HoverCardOverlay({
+  x,
+  y,
+  id,
+  usePostcodeView,
+  data,
+  postcodeData,
+  filters,
+  features,
+}: HoverCardOverlayProps) {
+  return (
+    <HoverCard
+      x={x}
+      y={y}
+      id={id}
+      isPostcode={usePostcodeView}
+      data={
+        usePostcodeView
+          ? postcodeData.find((f) => f.properties.postcode === id)?.properties || null
+          : data.find((d) => d.h3 === id) || null
+      }
+      filters={filters}
+      features={features}
+    />
+  );
+});
diff --git a/frontend/src/components/map/ListingPopups.tsx b/frontend/src/components/map/ListingPopups.tsx
new file mode 100644
index 0000000..f04e99b
--- /dev/null
+++ b/frontend/src/components/map/ListingPopups.tsx
@@ -0,0 +1,146 @@
+import { memo } from 'react';
+import { useTranslation } from 'react-i18next';
+import type { TFunction } from 'i18next';
+
+import type { ActualListing } from '../../types';
+
+function formatListingPrice(price: number): string {
+  return `£${price.toLocaleString()}`;
+}
+
+function formatListingHeadline(listing: ActualListing, t: TFunction): string | null {
+  const parts: string[] = [];
+  if (listing.bedrooms != null) parts.push(t('common.bedsCount', { count: listing.bedrooms }));
+  if (listing.bathrooms != null) parts.push(t('common.bathsCount', { count: listing.bathrooms }));
+  if (listing.property_sub_type) parts.push(listing.property_sub_type);
+  else if (listing.property_type) parts.push(listing.property_type);
+  return parts.length > 0 ? parts.join(' · ') : null;
+}
+
+export const ListingPopupSingleContent = memo(function ListingPopupSingleContent({
+  listing,
+}: {
+  listing: ActualListing;
+}) {
+  const { t } = useTranslation();
+  return (
+    <a
+      href={listing.listing_url}
+      target="_blank"
+      rel="noopener noreferrer"
+      className="block px-3 py-2"
+    >
+      {listing.asking_price != null && (
+        <div className="text-base font-bold text-teal-600 dark:text-teal-400">
+          {formatListingPrice(listing.asking_price)}
+          {listing.price_qualifier ? (
+            <span className="ml-1 text-xs font-medium text-warm-500 dark:text-warm-400">
+              {listing.price_qualifier}
+            </span>
+          ) : null}
+        </div>
+      )}
+      {formatListingHeadline(listing, t) && (
+        <div className="text-xs text-warm-700 dark:text-warm-200 mt-0.5">
+          {formatListingHeadline(listing, t)}
+        </div>
+      )}
+      {listing.address && (
+        <div className="text-xs text-warm-500 dark:text-warm-400 mt-0.5 line-clamp-2">
+          {listing.address}
+        </div>
+      )}
+      {listing.postcode && (
+        <div className="text-[11px] text-warm-400 dark:text-warm-500 mt-0.5">
+          {listing.postcode}
+        </div>
+      )}
+      {listing.floor_area_sqm != null && (
+        <div className="text-[11px] text-warm-500 dark:text-warm-400 mt-0.5">
+          {Math.round(listing.floor_area_sqm)} sqm
+          {listing.asking_price_per_sqm != null
+            ? ` · £${Math.round(listing.asking_price_per_sqm).toLocaleString()}/sqm`
+            : ''}
+        </div>
+      )}
+      {listing.features.length > 0 && (
+        <ul className="mt-1.5 text-[11px] text-warm-600 dark:text-warm-300 list-disc pl-4 space-y-0.5">
+          {listing.features.slice(0, 3).map((feature, idx) => (
+            <li key={idx} className="line-clamp-1">
+              {feature}
+            </li>
+          ))}
+        </ul>
+      )}
+      <div className="mt-1.5 text-[11px] text-teal-600 dark:text-teal-400 font-medium">
+        Open listing ↗
+      </div>
+    </a>
+  );
+});
+
+export const ListingClusterPopupContent = memo(function ListingClusterPopupContent({
+  count,
+  listings,
+}: {
+  count: number;
+  listings: ActualListing[];
+}) {
+  const { t } = useTranslation();
+  const visibleCount = listings.length;
+  return (
+    <div>
+      <div className="border-b border-warm-200 px-3 py-2 dark:border-warm-700">
+        <div className="text-base font-bold text-red-600 dark:text-red-400">
+          {count.toLocaleString()} listings
+        </div>
+        <div className="text-[11px] text-warm-500 dark:text-warm-400">
+          {visibleCount > 0
+            ? `Showing ${visibleCount.toLocaleString()} of ${count.toLocaleString()}`
+            : 'Grouped near this map position'}
+        </div>
+      </div>
+      {visibleCount > 0 && (
+        <div className="max-h-80 overflow-y-auto py-1">
+          {listings.map((listing, idx) => {
+            const headline = formatListingHeadline(listing, t);
+            return (
+              <a
+                key={`${listing.listing_url}-${idx}`}
+                href={listing.listing_url}
+                target="_blank"
+                rel="noopener noreferrer"
+                className="block border-b border-warm-100 px-3 py-2 last:border-b-0 hover:bg-warm-50 dark:border-warm-700 dark:hover:bg-warm-700/60"
+              >
+                <div className="flex items-start justify-between gap-3">
+                  <div className="min-w-0">
+                    <div className="text-sm font-semibold text-teal-700 dark:text-teal-300">
+                      {listing.asking_price != null
+                        ? formatListingPrice(listing.asking_price)
+                        : 'Listing'}
+                    </div>
+                    {headline && (
+                      <div className="mt-0.5 truncate text-xs text-warm-700 dark:text-warm-200">
+                        {headline}
+                      </div>
+                    )}
+                    {listing.address && (
+                      <div className="mt-0.5 line-clamp-1 text-[11px] text-warm-500 dark:text-warm-400">
+                        {listing.address}
+                      </div>
+                    )}
+                  </div>
+                  {listing.postcode && (
+                    <div className="shrink-0 text-[11px] font-medium text-warm-400 dark:text-warm-500">
+                      {listing.postcode}
+                    </div>
+                  )}
+                </div>
+              </a>
+            );
+          })}
+        </div>
+      )}
+    </div>
+  );
+});
diff --git a/frontend/src/components/map/Map.tsx b/frontend/src/components/map/Map.tsx
index 4d0f4a7..588bcbe 100644
--- a/frontend/src/components/map/Map.tsx
+++ b/frontend/src/components/map/Map.tsx
@@ -1,10 +1,8 @@
 import { useCallback, useRef, useEffect, useState, useMemo, memo } from 'react';
 import type { CSSProperties } from 'react';
 import { useTranslation } from 'react-i18next';
-import type { TFunction } from 'i18next';
-import { Layer, Map as MapGL, Source, useControl, ScaleControl } from 'react-map-gl/maplibre';
+import { Map as MapGL, ScaleControl } from 'react-map-gl/maplibre';
 import type { MapRef } from 'react-map-gl/maplibre';
-import { MapboxOverlay } from '@deck.gl/mapbox';
 import 'maplibre-gl/dist/maplibre-gl.css';
 import type {
   HexagonData,
@@ -17,7 +15,6 @@ import type {
   Bounds,
   MapFlyToOptions,
   ActualListing,
-  SchoolMetadata,
 } from '../../types';
 
 import {
@@ -26,28 +23,25 @@ import {
   getBoundsWithBottomScreenInset,
   getMapStyle,
   getMapDataBeforeId,
-  getPoiIconUrl,
   getMapCenterForTargetScreenPoint,
 } from '../../lib/map-utils';
-import {
-  MAP_MIN_ZOOM,
-  MAP_BOUNDS,
-  POI_GROUP_COLORS,
-  POSTCODE_ZOOM_THRESHOLD,
-  POI_AUTO_CARD_ZOOM_THRESHOLD,
-} from '../../lib/consts';
-import LocationSearch, { type SearchedLocation } from './LocationSearch';
-import MapLegend from './MapLegend';
-import HoverCard from './HoverCard';
+import { MAP_MIN_ZOOM, MAP_BOUNDS, POI_AUTO_CARD_ZOOM_THRESHOLD } from '../../lib/consts';
+import type { SearchedLocation } from './LocationSearch';
 import { LogoIcon } from '../ui/icons/LogoIcon';
 import { CloseIcon } from '../ui/icons/CloseIcon';
 import type { FeatureFilters } from '../../types';
 import { useDeckLayers } from '../../hooks/useDeckLayers';
-import { useTranslatedModes, type TravelTimeEntry } from '../../hooks/useTravelTime';
-import { ts } from '../../i18n/server';
-import { type OverlayId, OVERLAY_MIN_ZOOM } from '../../lib/overlays';
+import { useMapCardLayout } from '../../hooks/useMapCardLayout';
+import type { TravelTimeEntry } from '../../hooks/useTravelTime';
+import { type OverlayId } from '../../lib/overlays';
 import { CRIME_TYPE_VALUES } from '../../lib/crime-types';
 import type { BasemapId } from '../../lib/basemaps';
+import { DeckOverlay } from './DeckOverlay';
+import { OverlayTileLayers } from './OverlayTileLayers';
+import { MapTopCards } from './MapTopCards';
+import { PoiPopupCardContent } from './PoiPopupCard';
+import { ListingClusterPopupContent, ListingPopupSingleContent } from './ListingPopups';
+import { HoverCardOverlay } from './HoverCardOverlay';
 
 interface MapProps {
   data: HexagonData[];
@@ -99,168 +93,11 @@ const EMPTY_ACTUAL_LISTINGS: ActualListing[] = [];
 const EMPTY_OVERLAYS = new Set<OverlayId>();
 const ALL_CRIME_TYPES = new Set<string>(CRIME_TYPE_VALUES);
 
-function formatListingPrice(price: number): string {
-  return `£${price.toLocaleString()}`;
-}
-
-function formatListingHeadline(listing: ActualListing, t: TFunction): string | null {
-  const parts: string[] = [];
-  if (listing.bedrooms != null) parts.push(t('common.bedsCount', { count: listing.bedrooms }));
-  if (listing.bathrooms != null) parts.push(t('common.bathsCount', { count: listing.bathrooms }));
-  if (listing.property_sub_type) parts.push(listing.property_sub_type);
-  else if (listing.property_type) parts.push(listing.property_type);
-  return parts.length > 0 ? parts.join(' · ') : null;
-}
-
-function ListingPopupSingleContent({ listing, t }: { listing: ActualListing; t: TFunction }) {
-  return (
-    <a
-      href={listing.listing_url}
-      target="_blank"
-      rel="noopener noreferrer"
-      className="block px-3 py-2"
-    >
-      {listing.asking_price != null && (
-        <div className="text-base font-bold text-teal-600 dark:text-teal-400">
-          {formatListingPrice(listing.asking_price)}
-          {listing.price_qualifier ? (
-            <span className="ml-1 text-xs font-medium text-warm-500 dark:text-warm-400">
-              {listing.price_qualifier}
-            </span>
-          ) : null}
-        </div>
-      )}
-      {formatListingHeadline(listing, t) && (
-        <div className="text-xs text-warm-700 dark:text-warm-200 mt-0.5">
-          {formatListingHeadline(listing, t)}
-        </div>
-      )}
-      {listing.address && (
-        <div className="text-xs text-warm-500 dark:text-warm-400 mt-0.5 line-clamp-2">
-          {listing.address}
-        </div>
-      )}
-      {listing.postcode && (
-        <div className="text-[11px] text-warm-400 dark:text-warm-500 mt-0.5">
-          {listing.postcode}
-        </div>
-      )}
-      {listing.floor_area_sqm != null && (
-        <div className="text-[11px] text-warm-500 dark:text-warm-400 mt-0.5">
-          {Math.round(listing.floor_area_sqm)} sqm
-          {listing.asking_price_per_sqm != null
-            ? ` · £${Math.round(listing.asking_price_per_sqm).toLocaleString()}/sqm`
-            : ''}
-        </div>
-      )}
-      {listing.features.length > 0 && (
-        <ul className="mt-1.5 text-[11px] text-warm-600 dark:text-warm-300 list-disc pl-4 space-y-0.5">
-          {listing.features.slice(0, 3).map((feature, idx) => (
-            <li key={idx} className="line-clamp-1">
-              {feature}
-            </li>
-          ))}
-        </ul>
-      )}
-      <div className="mt-1.5 text-[11px] text-teal-600 dark:text-teal-400 font-medium">
-        Open listing ↗
-      </div>
-    </a>
-  );
-}
-
-function ListingClusterPopupContent({
-  count,
-  listings,
-  t,
-}: {
-  count: number;
-  listings: ActualListing[];
-  t: TFunction;
-}) {
-  const visibleCount = listings.length;
-  return (
-    <div>
-      <div className="border-b border-warm-200 px-3 py-2 dark:border-warm-700">
-        <div className="text-base font-bold text-red-600 dark:text-red-400">
-          {count.toLocaleString()} listings
-        </div>
-        <div className="text-[11px] text-warm-500 dark:text-warm-400">
-          {visibleCount > 0
-            ? `Showing ${visibleCount.toLocaleString()} of ${count.toLocaleString()}`
-            : 'Grouped near this map position'}
-        </div>
-      </div>
-      {visibleCount > 0 && (
-        <div className="max-h-80 overflow-y-auto py-1">
-          {listings.map((listing, idx) => {
-            const headline = formatListingHeadline(listing, t);
-            return (
-              <a
-                key={`${listing.listing_url}-${idx}`}
-                href={listing.listing_url}
-                target="_blank"
-                rel="noopener noreferrer"
-                className="block border-b border-warm-100 px-3 py-2 last:border-b-0 hover:bg-warm-50 dark:border-warm-700 dark:hover:bg-warm-700/60"
-              >
-                <div className="flex items-start justify-between gap-3">
-                  <div className="min-w-0">
-                    <div className="text-sm font-semibold text-teal-700 dark:text-teal-300">
-                      {listing.asking_price != null
-                        ? formatListingPrice(listing.asking_price)
-                        : 'Listing'}
-                    </div>
-                    {headline && (
-                      <div className="mt-0.5 truncate text-xs text-warm-700 dark:text-warm-200">
-                        {headline}
-                      </div>
-                    )}
-                    {listing.address && (
-                      <div className="mt-0.5 line-clamp-1 text-[11px] text-warm-500 dark:text-warm-400">
-                        {listing.address}
-                      </div>
-                    )}
-                  </div>
-                  {listing.postcode && (
-                    <div className="shrink-0 text-[11px] font-medium text-warm-400 dark:text-warm-500">
-                      {listing.postcode}
-                    </div>
-                  )}
-                </div>
-              </a>
-            );
-          })}
-        </div>
-      )}
-    </div>
-  );
-}
-
-interface PoiPopupCardData {
-  name: string;
-  category: string;
-  icon_category?: string;
-  group: string;
-  emoji: string;
-  school?: SchoolMetadata;
-}
-
 interface Dimensions {
   width: number;
   height: number;
 }
 
-const DESKTOP_TOP_CARD_WIDTH = 300;
-const DESKTOP_TOP_CARD_GAP = 8;
-const DESKTOP_TOP_CARD_HORIZONTAL_INSET = 24;
-const DESKTOP_TOP_CARDS_STACKED_MIN_MAP_WIDTH =
-  DESKTOP_TOP_CARD_WIDTH + DESKTOP_TOP_CARD_HORIZONTAL_INSET;
-const DESKTOP_TOP_CARDS_ROW_MIN_MAP_WIDTH =
-  DESKTOP_TOP_CARD_WIDTH * 2 + DESKTOP_TOP_CARD_GAP + DESKTOP_TOP_CARD_HORIZONTAL_INSET;
-const DESKTOP_TOP_CARD_CLASS = 'w-[300px]';
-const DESKTOP_LOCATION_SEARCH_INPUT_CLASS =
-  'px-2 py-2 text-sm w-full border-none outline-none bg-transparent text-warm-700 dark:text-warm-200 placeholder-warm-400 dark:placeholder-warm-500';
-
 type MapContainerStyle = CSSProperties & {
   '--map-mobile-bottom-inset'?: string;
 };
@@ -323,218 +160,6 @@ function getViewportRelativeVisibleAreaCenter(
   };
 }
 
-interface DeckWithPrivateDraw {
-  _drawLayers?: (
-    redrawReason: string,
-    renderOptions?: { viewports?: unknown[]; [key: string]: unknown }
-  ) => unknown;
-  __propertyMapNullViewportPatch?: boolean;
-}
-
-function patchNullViewportDraw(overlay: MapboxOverlay) {
-  const deck = (overlay as unknown as { _deck?: DeckWithPrivateDraw })._deck;
-  if (!deck || deck.__propertyMapNullViewportPatch || typeof deck._drawLayers !== 'function') {
-    return;
-  }
-
-  const drawLayers = deck._drawLayers.bind(deck);
-  deck._drawLayers = (redrawReason, renderOptions) => {
-    const viewports = renderOptions?.viewports;
-    if (viewports) {
-      // Split-route startup can hand deck.gl a transient null viewport before MapLibre has sized the map.
-      const nonNullViewports = viewports.filter(Boolean);
-      if (nonNullViewports.length === 0) return;
-      if (nonNullViewports.length !== viewports.length) {
-        return drawLayers(redrawReason, { ...renderOptions, viewports: nonNullViewports });
-      }
-    }
-    return drawLayers(redrawReason, renderOptions);
-  };
-  deck.__propertyMapNullViewportPatch = true;
-}
-
-class SafeMapboxOverlay extends MapboxOverlay {
-  onAdd(map: unknown) {
-    const element = super.onAdd(map);
-    patchNullViewportDraw(this);
-    return element;
-  }
-
-  setProps(props: Parameters<MapboxOverlay['setProps']>[0]) {
-    super.setProps(props);
-    patchNullViewportDraw(this);
-  }
-}
-
-function getPoiGroupColor(group: string): [number, number, number] {
-  const color = POI_GROUP_COLORS[group];
-  if (!color) {
-    throw new Error(`Missing POI group color for '${group}'`);
-  }
-  return color;
-}
-
-/** Best-effort web URL from a free-text website field — GIAS stores some with
- * "http://", some without, and some as bare hostnames. */
-function normalizeSchoolWebsiteUrl(raw: string): string | null {
-  const trimmed = raw.trim();
-  if (!trimmed) return null;
-  if (/^https?:\/\//i.test(trimmed)) return trimmed;
-  if (/^[\w.-]+\.[a-z]{2,}/i.test(trimmed)) return `http://${trimmed}`;
-  return null;
-}
-
-function renderSchoolMetadata(school: SchoolMetadata) {
-  // First line collects the headline classification (phase, type, religious
-  // character) so the popup is scannable even when most fields are absent.
-  const headline: string[] = [];
-  if (school.phase) headline.push(school.phase);
-  if (school.type) headline.push(school.type);
-
-  const pupilsLine =
-    school.pupils !== undefined && school.capacity !== undefined
-      ? `${school.pupils.toLocaleString()} / ${school.capacity.toLocaleString()} pupils`
-      : school.pupils !== undefined
-        ? `${school.pupils.toLocaleString()} pupils`
-        : school.capacity !== undefined
-          ? `Capacity ${school.capacity.toLocaleString()}`
-          : null;
-
-  const websiteUrl = school.website ? normalizeSchoolWebsiteUrl(school.website) : null;
-
-  return (
-    <dl className="mt-2 grid grid-cols-[auto_1fr] gap-x-2 gap-y-0.5 text-xs text-warm-600 dark:text-warm-300">
-      {headline.length > 0 && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Type</dt>
-          <dd className="dark:text-warm-200">{headline.join(' · ')}</dd>
-        </>
-      )}
-      {school.age_range && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Ages</dt>
-          <dd className="dark:text-warm-200">{school.age_range}</dd>
-        </>
-      )}
-      {school.gender && school.gender !== 'Mixed' && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Gender</dt>
-          <dd className="dark:text-warm-200">{school.gender}</dd>
-        </>
-      )}
-      {pupilsLine && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Pupils</dt>
-          <dd className="dark:text-warm-200">{pupilsLine}</dd>
-        </>
-      )}
-      {school.fsm_percent !== undefined && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Free meal</dt>
-          <dd className="dark:text-warm-200">{school.fsm_percent.toFixed(1)}%</dd>
-        </>
-      )}
-      {school.ofsted_rating && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Ofsted</dt>
-          <dd className="dark:text-warm-200">{school.ofsted_rating}</dd>
-        </>
-      )}
-      {school.sixth_form === 'Has a sixth form' && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Sixth form</dt>
-          <dd className="dark:text-warm-200">Yes</dd>
-        </>
-      )}
-      {school.religious_character &&
-        school.religious_character !== 'Does not apply' &&
-        school.religious_character !== 'None' && (
-          <>
-            <dt className="text-warm-500 dark:text-warm-400">Religion</dt>
-            <dd className="dark:text-warm-200">{school.religious_character}</dd>
-          </>
-        )}
-      {school.admissions_policy && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Admissions</dt>
-          <dd className="dark:text-warm-200">{school.admissions_policy}</dd>
-        </>
-      )}
-      {school.trust && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Trust</dt>
-          <dd className="dark:text-warm-200">{school.trust}</dd>
-        </>
-      )}
-      {(school.address || school.postcode) && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Address</dt>
-          <dd className="dark:text-warm-200">
-            {[school.address, school.postcode].filter(Boolean).join(', ')}
-          </dd>
-        </>
-      )}
-      {school.local_authority && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">LA</dt>
-          <dd className="dark:text-warm-200">{school.local_authority}</dd>
-        </>
-      )}
-      {school.head_name && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Head</dt>
-          <dd className="dark:text-warm-200">{school.head_name}</dd>
-        </>
-      )}
-      {websiteUrl && (
-        <>
-          <dt className="text-warm-500 dark:text-warm-400">Website</dt>
-          <dd className="truncate">
-            <a
-              href={websiteUrl}
-              target="_blank"
-              rel="noreferrer noopener"
-              className="pointer-events-auto text-teal-600 hover:underline dark:text-teal-400"
-            >
-              {websiteUrl.replace(/^https?:\/\//, '')}
-            </a>
-          </dd>
-        </>
-      )}
-    </dl>
-  );
-}
-
-function PoiPopupCardContent({ poi }: { poi: PoiPopupCardData }) {
-  return (
-    <div className="px-3 py-2 max-w-[280px]">
-      <div className="flex items-center gap-2">
-        <img
-          src={getPoiIconUrl(poi.category, poi.emoji, poi.icon_category, poi.name)}
-          alt=""
-          aria-hidden="true"
-          loading="lazy"
-          referrerPolicy="no-referrer"
-          className="h-5 w-5 shrink-0 rounded-[4px] bg-white object-contain p-0.5"
-        />
-        <div className="min-w-0">
-          <div className="font-semibold dark:text-warm-100">{poi.name}</div>
-          <div className="flex items-center gap-1.5 text-xs text-warm-500 dark:text-warm-400">
-            <span
-              className="inline-block w-2 h-2 rounded-full flex-shrink-0"
-              style={{
-                backgroundColor: `rgb(${getPoiGroupColor(poi.group).join(',')})`,
-              }}
-            />
-            {ts(poi.category)}
-          </div>
-        </div>
-      </div>
-      {poi.school && renderSchoolMetadata(poi.school)}
-    </div>
-  );
-}
-
 function getRenderedViewState(map: MapRef | null): ViewState | null {
   if (!map) return null;
 
@@ -565,186 +190,6 @@ function getRenderedVisibleCenter(
   };
 }
 
-function DeckOverlay({
-  layers,
-  getTooltip,
-}: {
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  layers: any[];
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  getTooltip: any;
-}) {
-  const overlay = useControl(() => new SafeMapboxOverlay({ interleaved: true }));
-
-  useEffect(() => {
-    overlay.setProps({
-      layers: layers.filter(Boolean),
-      getTooltip,
-    });
-  }, [overlay, layers, getTooltip]);
-
-  return null;
-}
-
-function overlayTileUrl(path: string): string {
-  return `${window.location.origin}/api/overlays/${path}/{z}/{x}/{y}`;
-}
-
-function OverlayTileLayers({
-  activeOverlays,
-  activeCrimeTypes,
-  zoom,
-}: {
-  activeOverlays: Set<OverlayId>;
-  activeCrimeTypes: Set<string>;
-  zoom: number;
-}) {
-  if (zoom < POSTCODE_ZOOM_THRESHOLD || activeOverlays.size === 0) return null;
-
-  const showNoise = activeOverlays.has('noise');
-  const showCrime = activeOverlays.has('crime-hotspots');
-  const showTrees = activeOverlays.has('trees-outside-woodlands');
-  const showPropertyBorders = activeOverlays.has('property-borders');
-
-  // Restrict the heatmap to the selected crime types. This must always be a
-  // concrete expression: passing `filter={undefined}` makes react-map-gl call
-  // map.addLayer({filter: undefined}), which MapLibre rejects at validation
-  // ("filter: array expected, undefined found"), so the layer is never created
-  // and the heatmap stays blank until a later setFilter call. An `in` over the
-  // selected types matches everything when all 14 are selected.
-  const crimeFilter = ['in', ['get', 'crime_type'], ['literal', Array.from(activeCrimeTypes)]];
-
-  return (
-    <>
-      {showNoise && (
-        <Source
-          id="overlay-noise-source"
-          type="raster"
-          tiles={[overlayTileUrl('noise')]}
-          tileSize={256}
-          minzoom={OVERLAY_MIN_ZOOM.noise}
-          maxzoom={14}
-        >
-          <Layer
-            id="overlay-noise"
-            type="raster"
-            minzoom={POSTCODE_ZOOM_THRESHOLD}
-            paint={{
-              'raster-opacity': 0.68,
-              'raster-fade-duration': 120,
-            }}
-          />
-        </Source>
-      )}
-
-      {showCrime && (
-        <Source
-          id="overlay-crime-source"
-          type="vector"
-          tiles={[overlayTileUrl('crime-hotspots')]}
-          minzoom={OVERLAY_MIN_ZOOM['crime-hotspots']}
-          maxzoom={15}
-        >
-          <Layer
-            id="overlay-crime-heatmap"
-            type="heatmap"
-            source-layer="crime_hotspots"
-            minzoom={POSTCODE_ZOOM_THRESHOLD}
-            filter={crimeFilter as never}
-            paint={
-              {
-                'heatmap-weight': [
-                  'interpolate',
-                  ['linear'],
-                  ['coalesce', ['get', 'count'], ['get', 'weight'], 1],
-                  0,
-                  0,
-                  10,
-                  1,
-                ],
-                'heatmap-intensity': ['interpolate', ['linear'], ['zoom'], 15, 0.8, 18, 2.2],
-                'heatmap-radius': ['interpolate', ['linear'], ['zoom'], 15, 18, 18, 30],
-                'heatmap-opacity': 0.72,
-                'heatmap-color': [
-                  'interpolate',
-                  ['linear'],
-                  ['heatmap-density'],
-                  0,
-                  'rgba(0, 0, 0, 0)',
-                  0.2,
-                  'rgb(253, 224, 71)',
-                  0.45,
-                  'rgb(249, 115, 22)',
-                  0.75,
-                  'rgb(220, 38, 38)',
-                  1,
-                  'rgb(127, 29, 29)',
-                ],
-              } as never
-            }
-          />
-        </Source>
-      )}
-
-      {showTrees && (
-        <Source
-          id="overlay-trees-source"
-          type="vector"
-          tiles={[overlayTileUrl('trees-outside-woodlands')]}
-          minzoom={OVERLAY_MIN_ZOOM['trees-outside-woodlands']}
-          maxzoom={16}
-        >
-          <Layer
-            id="overlay-tree-polygons"
-            type="fill"
-            source-layer="trees_outside_woodlands"
-            minzoom={POSTCODE_ZOOM_THRESHOLD}
-            paint={
-              {
-                'fill-color': '#1f9d55',
-                'fill-opacity': [
-                  'interpolate',
-                  ['linear'],
-                  ['coalesce', ['get', 'area_sqm'], 0],
-                  0,
-                  0.28,
-                  250,
-                  0.62,
-                ],
-                'fill-outline-color': 'rgba(15, 81, 50, 0.65)',
-              } as never
-            }
-          />
-        </Source>
-      )}
-
-      {showPropertyBorders && (
-        <Source
-          id="overlay-property-borders-source"
-          type="vector"
-          tiles={[overlayTileUrl('property-borders')]}
-          minzoom={OVERLAY_MIN_ZOOM['property-borders']}
-          maxzoom={16}
-        >
-          <Layer
-            id="overlay-property-borders"
-            type="line"
-            source-layer="property_borders"
-            minzoom={POSTCODE_ZOOM_THRESHOLD}
-            paint={
-              {
-                'line-color': '#b45309',
-                'line-opacity': ['interpolate', ['linear'], ['zoom'], 15, 0.35, 18, 0.85],
-                'line-width': ['interpolate', ['linear'], ['zoom'], 15, 0.4, 18, 1.4],
-              } as never
-            }
-          />
-        </Source>
-      )}
-    </>
-  );
-}
-
 export default memo(function Map({
   data,
   postcodeData,
@@ -790,7 +235,6 @@ export default memo(function Map({
   const containerRef = useRef<HTMLDivElement>(null);
   const mapRef = useRef<MapRef | null>(null);
   const { t } = useTranslation();
-  const modes = useTranslatedModes();
   const densityLabel = densityLabelProp ?? t('mapLegend.numberOfProperties');
   const [internalViewState, setInternalViewState] = useState<ViewState>(initialViewState);
   const [dimensions, setDimensions] = useState<Dimensions>({ width: 0, height: 0 });
@@ -941,23 +385,16 @@ export default memo(function Map({
     () => (bottomScreenInset > 0 ? { '--map-mobile-bottom-inset': `${bottomScreenInset}px` } : {}),
     [bottomScreenInset]
   );
-  const hideDesktopTopCardsForWidth =
-    hideTopCardsWhenNarrow &&
-    dimensions.width > 0 &&
-    dimensions.width < DESKTOP_TOP_CARDS_STACKED_MIN_MAP_WIDTH;
-  const stackDesktopTopCards =
-    hideTopCardsWhenNarrow &&
-    dimensions.width >= DESKTOP_TOP_CARDS_STACKED_MIN_MAP_WIDTH &&
-    dimensions.width < DESKTOP_TOP_CARDS_ROW_MIN_MAP_WIDTH;
-  const showLocationSearch = !hideLocationSearch && !hideDesktopTopCardsForWidth;
-  const showLegend = !hideLegend && !hideDesktopTopCardsForWidth;
+  const { showLocationSearch, showLegend, topCardsLayoutClass } = useMapCardLayout({
+    mapWidth: dimensions.width,
+    hideTopCardsWhenNarrow,
+    hideLegend,
+    hideLocationSearch,
+  });
   const getViewportCenter = useCallback(() => {
     const center = mapRef.current?.getCenter();
     return center ? { lat: center.lat, lng: center.lng } : null;
   }, []);
-  const desktopTopCardsLayoutClass = stackDesktopTopCards
-    ? 'flex-col items-start'
-    : 'items-start justify-between';
 
   const {
     layers,
@@ -1108,79 +545,29 @@ export default memo(function Map({
       ) : (
         <>
           {(showLocationSearch || showLegend) && (
-            <div
-              className={`absolute top-3 left-3 right-3 z-20 flex gap-2 pointer-events-none ${desktopTopCardsLayoutClass}`}
-            >
-              {showLocationSearch && (
-                <LocationSearch
-                  onFlyTo={handleFlyTo}
-                  onLocationSearched={onLocationSearched}
-                  onCurrentLocationFound={onCurrentLocationFound}
-                  onMouseEnter={handleMouseLeave}
-                  getViewportCenter={getViewportCenter}
-                  className={DESKTOP_TOP_CARD_CLASS}
-                  inputClassName={DESKTOP_LOCATION_SEARCH_INPUT_CLASS}
-                />
-              )}
-              {showLegend &&
-                (viewFeature && colorRange ? (
-                  viewFeature.startsWith('tt_') ? (
-                    <MapLegend
-                      featureLabel={t('travel.travelTime', {
-                        mode: modes.label(
-                          viewFeature.split('_')[1] as 'car' | 'bicycle' | 'walking' | 'transit'
-                        ),
-                      })}
-                      range={colorRange}
-                      showCancel={viewSource === 'eye'}
-                      onCancel={onCancelPin}
-                      onResetScale={viewSource === 'eye' ? onResetPreviewScale : undefined}
-                      resetScaleDisabled={!canResetPreviewScale}
-                      mode="feature"
-                      theme={theme}
-                      suffix=" min"
-                      className={DESKTOP_TOP_CARD_CLASS}
-                    />
-                  ) : colorFeatureMeta ? (
-                    <MapLegend
-                      featureLabel={
-                        viewSource === 'eye'
-                          ? t('mapLegend.previewing', { name: ts(colorFeatureMeta.name) })
-                          : ts(colorFeatureMeta.name)
-                      }
-                      range={colorRange}
-                      showCancel={viewSource === 'eye'}
-                      onCancel={onCancelPin}
-                      onResetScale={viewSource === 'eye' ? onResetPreviewScale : undefined}
-                      resetScaleDisabled={!canResetPreviewScale}
-                      mode="feature"
-                      enumValues={
-                        colorFeatureMeta.type === 'enum' ? colorFeatureMeta.values : undefined
-                      }
-                      featureName={colorFeatureMeta.name}
-                      theme={theme}
-                      suffix={colorFeatureMeta.suffix}
-                      raw={colorFeatureMeta.raw}
-                      className={DESKTOP_TOP_CARD_CLASS}
-                    />
-                  ) : null
-                ) : (
-                  <MapLegend
-                    featureLabel={densityLabel}
-                    range={
-                      usePostcodeView
-                        ? [postcodeCountRange.min, postcodeCountRange.max]
-                        : [countRange.min, countRange.max]
-                    }
-                    totalCount={totalCountProp}
-                    showCancel={false}
-                    onCancel={onCancelPin}
-                    mode="density"
-                    theme={theme}
-                    className={DESKTOP_TOP_CARD_CLASS}
-                  />
-                ))}
-            </div>
+            <MapTopCards
+              layoutClass={topCardsLayoutClass}
+              showLocationSearch={showLocationSearch}
+              showLegend={showLegend}
+              onFlyTo={handleFlyTo}
+              onLocationSearched={onLocationSearched}
+              onCurrentLocationFound={onCurrentLocationFound}
+              onLocationSearchMouseEnter={handleMouseLeave}
+              getViewportCenter={getViewportCenter}
+              viewFeature={viewFeature}
+              colorRange={colorRange}
+              viewSource={viewSource}
+              onCancelPin={onCancelPin}
+              onResetPreviewScale={onResetPreviewScale}
+              canResetPreviewScale={canResetPreviewScale}
+              colorFeatureMeta={colorFeatureMeta}
+              usePostcodeView={usePostcodeView}
+              countRange={countRange}
+              postcodeCountRange={postcodeCountRange}
+              densityLabel={densityLabel}
+              totalCount={totalCountProp}
+              theme={theme}
+            />
           )}
           {autoPoiCards.map(({ poi, x, y }) => (
             <div
@@ -1247,28 +634,23 @@ export default memo(function Map({
                 <CloseIcon className="w-3 h-3" />
               </button>
               {listingPopup.mode === 'single' ? (
-                <ListingPopupSingleContent listing={listingPopup.listing} t={t} />
+                <ListingPopupSingleContent listing={listingPopup.listing} />
               ) : (
                 <ListingClusterPopupContent
                   count={listingPopup.count}
                   listings={listingPopup.listings}
-                  t={t}
                 />
               )}
             </div>
           )}
           {hoverPosition && hoveredHexagonId && hoveredHexagonId !== selectedHexagonId && (
-            <HoverCard
+            <HoverCardOverlay
               x={hoverPosition.x}
               y={hoverPosition.y}
               id={hoveredHexagonId}
-              isPostcode={usePostcodeView}
-              data={
-                usePostcodeView
-                  ? postcodeData.find((f) => f.properties.postcode === hoveredHexagonId)
-                      ?.properties || null
-                  : data.find((d) => d.h3 === hoveredHexagonId) || null
-              }
+              usePostcodeView={usePostcodeView}
+              data={data}
+              postcodeData={postcodeData}
               filters={filters}
               features={features}
             />
diff --git a/frontend/src/components/map/MapPage.tsx b/frontend/src/components/map/MapPage.tsx
index 9a1c9ae..acbe064 100644
--- a/frontend/src/components/map/MapPage.tsx
+++ b/frontend/src/components/map/MapPage.tsx
@@ -1,7 +1,7 @@
 import { Suspense, useCallback, useEffect, useMemo, useRef, useState } from 'react';
 import { Trans, useTranslation } from 'react-i18next';
 
-import type { ActualListing, MapFlyToOptions, PostcodeGeometry } from '../../types';
+import type { ActualListing, PostcodeGeometry } from '../../types';
 import type { SearchedLocation } from './LocationSearch';
 import { useMapData } from '../../hooks/useMapData';
 import { usePOIData } from '../../hooks/usePOIData';
@@ -67,11 +67,11 @@ import {
   useMobileBackNavigationGuard,
   useScreenshotReadySignal,
 } from './map-page/effects';
+import { useMobileDrawer } from './map-page/useMobileDrawer';
 import type { MapFlyTo, MapPageProps } from './map-page/types';
 
 export type { ExportState } from './map-page/types';
 
-type PendingFlyTo = { lat: number; lng: number; zoom: number };
 const EMPTY_ACTUAL_LISTINGS: ActualListing[] = [];
 
 export default function MapPage({
@@ -127,10 +127,11 @@ export default function MapPage({
   );
   const [leftPaneWidth, leftPaneHandlers] = usePaneResize(384, 200, 0.45, 'left');
   const [rightPaneWidth, rightPaneHandlers] = usePaneResize(384, 200, 0.45, 'right');
-  const [mobileDrawerOpen, setMobileDrawerOpen] = useState(false);
-  const [mobileBottomSheetHeight, setMobileBottomSheetHeight] = useState(0);
-  const [poiPaneOpen, setPoiPaneOpen] = useState(false);
-  const [overlayPaneOpen, setOverlayPaneOpen] = useState(false);
+  // The POI and overlay panes are mutually exclusive, so a single state tracks
+  // which one (if any) is open.
+  const [openMapPane, setOpenMapPane] = useState<'poi' | 'overlay' | null>(null);
+  const poiPaneOpen = openMapPane === 'poi';
+  const overlayPaneOpen = openMapPane === 'overlay';
   const [currentLocation, setCurrentLocation] = useState<{ lat: number; lng: number } | null>(null);
   const [listingsToggleEnabled, setListingsToggleEnabled] = useState(true);
   const [pendingInitialPostcode, setPendingInitialPostcode] = useState<string | null>(
@@ -184,27 +185,21 @@ export default function MapPage({
   } = useTravelTime(initialTravelTime);
 
   const mapFlyToRef = useRef<MapFlyTo | null>(null);
-  const pendingCurrentLocationFlyToRef = useRef<{ lat: number; lng: number } | null>(null);
-  const pendingLocationSearchFlyToRef = useRef<PendingFlyTo | null>(null);
-  const mobileDrawerPanelRectRef = useRef<DOMRectReadOnly | null>(null);
   const areaPaneScrollTopRef = useRef(0);
   const propertiesPaneScrollTopRef = useRef(0);
 
-  const getMobileMapFlyToOptions = useCallback((): MapFlyToOptions | undefined => {
-    if (!isMobile) return undefined;
-
-    const panelRect = mobileDrawerPanelRectRef.current;
-    if (mobileDrawerOpen && panelRect) {
-      const bottomInset = Math.max(0, window.innerHeight - panelRect.top);
-      if (bottomInset > 0) {
-        return { visibleViewportArea: { bottom: bottomInset } };
-      }
-    }
-
-    return mobileBottomSheetHeight > 0
-      ? { visibleArea: { bottom: mobileBottomSheetHeight } }
-      : undefined;
-  }, [isMobile, mobileBottomSheetHeight, mobileDrawerOpen]);
+  const {
+    mobileDrawerOpen,
+    mobileBottomSheetHeight,
+    setMobileBottomSheetHeight,
+    openMobileDrawer,
+    openMobileDrawerForLocationSearch,
+    clearPendingLocationSearchFlyTo,
+    queueCurrentLocationFlyTo,
+    handleMobileDrawerPanelRectChange,
+    handleMobileDrawerClose,
+    getMobileMapFlyToOptions,
+  } = useMobileDrawer(isMobile, mapFlyToRef);
 
   const mapData = useMapData({
     filters,
@@ -217,6 +212,12 @@ export default function MapPage({
     shareCode,
   });
 
+  // Read the zoom through a ref inside handleAiFilterSubmit so panning/zooming
+  // doesn't recreate the callback (it sits in the Filters pane's dependency
+  // chain, which would otherwise re-render on every camera move).
+  const currentViewZoomRef = useRef<number | undefined>(undefined);
+  currentViewZoomRef.current = mapData.currentView?.zoom;
+
   const handleAiFilterSubmit = useCallback(
     async (query: string) => {
       const context = {
@@ -283,7 +284,7 @@ export default function MapPage({
           mapFlyToRef.current?.(
             destination.lat,
             destination.lon,
-            mapData.currentView?.zoom ?? INITIAL_VIEW_STATE.zoom,
+            currentViewZoomRef.current ?? INITIAL_VIEW_STATE.zoom,
             getMobileMapFlyToOptions()
           );
         }
@@ -298,7 +299,6 @@ export default function MapPage({
       getMobileMapFlyToOptions,
       handleSetEntries,
       handleSetFilters,
-      mapData.currentView?.zoom,
     ]
   );
 
@@ -395,20 +395,6 @@ export default function MapPage({
     journeyDest,
   });
 
-  const consumePendingLocationSearchFlyTo = useCallback((rect?: DOMRectReadOnly | null) => {
-    const pending = pendingLocationSearchFlyToRef.current;
-    const panelRect = rect ?? mobileDrawerPanelRectRef.current;
-    if (!pending || !panelRect) return;
-
-    const bottomInset = Math.max(0, window.innerHeight - panelRect.top);
-    const flyTo = mapFlyToRef.current;
-    if (!flyTo) return;
-    flyTo(pending.lat, pending.lng, pending.zoom, {
-      visibleViewportArea: { bottom: bottomInset },
-    });
-    pendingLocationSearchFlyToRef.current = null;
-  }, []);
-
   const handleLocationSearchResult = useCallback(
     (result: SearchedLocation | null) => {
       if (result) {
@@ -428,68 +414,41 @@ export default function MapPage({
           result.focusAddress
         );
         if (isMobile) {
-          pendingLocationSearchFlyToRef.current = {
+          openMobileDrawerForLocationSearch({
             lat: markerLat ?? result.latitude,
             lng: markerLng ?? result.longitude,
             zoom: result.zoom,
-          };
-          setMobileDrawerOpen(true);
-          consumePendingLocationSearchFlyTo();
+          });
         }
       } else {
         setCurrentLocation(null);
-        pendingLocationSearchFlyToRef.current = null;
+        clearPendingLocationSearchFlyTo();
         handleCloseSelection();
       }
     },
-    [consumePendingLocationSearchFlyTo, handleCloseSelection, handleLocationSearch, isMobile]
+    [
+      clearPendingLocationSearchFlyTo,
+      handleCloseSelection,
+      handleLocationSearch,
+      isMobile,
+      openMobileDrawerForLocationSearch,
+    ]
   );
 
-  const consumePendingCurrentLocationFlyTo = useCallback((rect?: DOMRectReadOnly | null) => {
-    const pending = pendingCurrentLocationFlyToRef.current;
-    const panelRect = rect ?? mobileDrawerPanelRectRef.current;
-    if (!pending || !panelRect) return;
-
-    const bottomInset = Math.max(0, window.innerHeight - panelRect.top);
-    const flyTo = mapFlyToRef.current;
-    if (!flyTo) return;
-    flyTo(pending.lat, pending.lng, 17, {
-      visibleViewportArea: { bottom: bottomInset },
-    });
-    pendingCurrentLocationFlyToRef.current = null;
-  }, []);
-
   const handleCurrentLocationFound = useCallback(
     (lat: number, lng: number) => {
       if (isMobile) {
-        pendingCurrentLocationFlyToRef.current = { lat, lng };
-        consumePendingCurrentLocationFlyTo();
+        queueCurrentLocationFlyTo(lat, lng);
       } else {
         mapFlyToRef.current?.(lat, lng, 17);
       }
       setCurrentLocation({ lat, lng });
       handleCurrentLocationSearch(lat, lng);
-      if (isMobile) setMobileDrawerOpen(true);
+      if (isMobile) openMobileDrawer();
     },
-    [consumePendingCurrentLocationFlyTo, handleCurrentLocationSearch, isMobile]
+    [handleCurrentLocationSearch, isMobile, openMobileDrawer, queueCurrentLocationFlyTo]
   );
 
-  const handleMobileDrawerPanelRectChange = useCallback(
-    (rect: DOMRectReadOnly) => {
-      mobileDrawerPanelRectRef.current = rect;
-      consumePendingCurrentLocationFlyTo(rect);
-      consumePendingLocationSearchFlyTo(rect);
-    },
-    [consumePendingCurrentLocationFlyTo, consumePendingLocationSearchFlyTo]
-  );
-
-  const handleMobileDrawerClose = useCallback(() => {
-    pendingCurrentLocationFlyToRef.current = null;
-    pendingLocationSearchFlyToRef.current = null;
-    mobileDrawerPanelRectRef.current = null;
-    setMobileDrawerOpen(false);
-  }, []);
-
   const shareReturnViewRef = useRef(shareCode ? initialViewState : null);
   // Hide the upgrade modal as soon as the user dismisses it. We can't rely on
   // the camera fly alone to close it: flying back to the free/shared zone only
@@ -555,11 +514,7 @@ export default function MapPage({
     isMobile,
     flyTo: mapFlyToRef,
     onLocationSearch: handleLocationSearch,
-    onOpenMobileDrawer: (target) => {
-      pendingLocationSearchFlyToRef.current = target;
-      setMobileDrawerOpen(true);
-      consumePendingLocationSearchFlyTo();
-    },
+    onOpenMobileDrawer: openMobileDrawerForLocationSearch,
     onSettled: () => setPendingInitialPostcode(null),
   });
   useHorizontalSwipeNavigationGuard();
@@ -578,10 +533,10 @@ export default function MapPage({
     (id: string, isPostcode?: boolean, geometry?: PostcodeGeometry) => {
       handleHexagonClick(id, isPostcode, geometry);
       if (id) {
-        setMobileDrawerOpen(true);
+        openMobileDrawer();
       }
     },
-    [handleHexagonClick]
+    [handleHexagonClick, openMobileDrawer]
   );
 
   const hexagonLocation = useHexagonLocation(
@@ -641,15 +596,20 @@ export default function MapPage({
       shareAndSaveView,
     ]
   );
+  // dashboardParams changes on every camera move; read it through a ref so the
+  // save/update handlers (and the Filters pane depending on them) stay stable
+  // while panning. The ref always holds the params of the latest render.
+  const dashboardParamsRef = useRef(dashboardParams);
+  dashboardParamsRef.current = dashboardParams;
   const handleSaveSearch = useCallback(
     async (name: string) => {
-      await onSaveSearch?.(name, dashboardParams);
+      await onSaveSearch?.(name, dashboardParamsRef.current);
     },
-    [dashboardParams, onSaveSearch]
+    [onSaveSearch]
   );
   const handleUpdateEditInPlaceWithParams = useCallback(async () => {
-    await onUpdateEditInPlace?.(dashboardParams);
-  }, [dashboardParams, onUpdateEditInPlace]);
+    await onUpdateEditInPlace?.(dashboardParamsRef.current);
+  }, [onUpdateEditInPlace]);
   const checkoutReturnPath = useMemo(
     () => `/dashboard${dashboardParams ? `?${dashboardParams}` : ''}`,
     [dashboardParams]
@@ -686,6 +646,273 @@ export default function MapPage({
     }
   }, [mapData.licenseRequired]);
 
+  const handleUpgradeClick = useCallback(() => {
+    onNavigateTo('pricing');
+  }, [onNavigateTo]);
+  const handleTogglePoiPane = useCallback(() => {
+    setOpenMapPane((pane) => (pane === 'poi' ? null : 'poi'));
+  }, []);
+  const handleToggleOverlayPane = useCallback(() => {
+    setOpenMapPane((pane) => (pane === 'overlay' ? null : 'overlay'));
+  }, []);
+  const handleClosePoiPane = useCallback(() => {
+    setOpenMapPane((pane) => (pane === 'poi' ? null : pane));
+  }, []);
+  const handleCloseOverlayPane = useCallback(() => {
+    setOpenMapPane((pane) => (pane === 'overlay' ? null : pane));
+  }, []);
+  const handleAreaTabClick = useCallback(() => {
+    setRightPaneTab('area');
+  }, [setRightPaneTab]);
+  const handleMobileDrawerTabChange = useCallback(
+    (tab: 'area' | 'properties') => {
+      if (tab === 'properties') {
+        handlePropertiesTabClick();
+      } else {
+        setRightPaneTab(tab);
+      }
+    },
+    [handlePropertiesTabClick, setRightPaneTab]
+  );
+
+  const renderAreaPane = useCallback(
+    () => (
+      <Suspense fallback={<PaneFallback />}>
+        <AreaPane
+          stats={areaStats}
+          globalFeatures={features}
+          loading={loadingAreaStats}
+          hexagonId={selectedHexagon?.id || null}
+          isPostcode={selectedHexagon?.type === 'postcode'}
+          hexagonLocation={hexagonLocation}
+          filters={filters}
+          unfilteredCount={unfilteredAreaCount}
+          statsUseFilters={areaStatsUseFilters}
+          onStatsUseFiltersChange={setAreaStatsUseFilters}
+          travelTimeEntries={activeEntries}
+          shareCode={shareCode}
+          isGroupExpanded={isAreaGroupExpanded}
+          onToggleGroup={toggleAreaGroup}
+          scrollTopRef={areaPaneScrollTopRef}
+          scrollRestoreKey={
+            selectedHexagon ? `${selectedHexagon.type}:${selectedHexagon.id}` : null
+          }
+          scrollSaveDisabled={loadingAreaStats && areaStats == null}
+        />
+      </Suspense>
+    ),
+    [
+      activeEntries,
+      areaStats,
+      areaStatsUseFilters,
+      features,
+      filters,
+      hexagonLocation,
+      isAreaGroupExpanded,
+      loadingAreaStats,
+      selectedHexagon,
+      setAreaStatsUseFilters,
+      shareCode,
+      toggleAreaGroup,
+      unfilteredAreaCount,
+    ]
+  );
+
+  const renderPropertiesPane = useCallback(
+    () => (
+      <Suspense fallback={<PaneFallback />}>
+        <PropertiesPane
+          properties={properties}
+          total={propertiesTotal}
+          loading={loadingProperties}
+          hexagonId={selectedHexagon?.id || null}
+          onLoadMore={handleLoadMoreProperties}
+          scrollTopRef={propertiesPaneScrollTopRef}
+          scrollRestoreKey={
+            selectedHexagon ? `${selectedHexagon.type}:${selectedHexagon.id}` : null
+          }
+          scrollSaveDisabled={loadingProperties && properties.length === 0}
+        />
+      </Suspense>
+    ),
+    [handleLoadMoreProperties, loadingProperties, properties, propertiesTotal, selectedHexagon]
+  );
+
+  const poiPane = useMemo(
+    () => (
+      <Suspense fallback={<PaneFallback />}>
+        <POIPane
+          groups={poiCategoryGroups}
+          selectedCategories={selectedPOICategories}
+          onCategoriesChange={setSelectedPOICategories}
+          poiCount={pois.length}
+          onClose={handleClosePoiPane}
+        />
+      </Suspense>
+    ),
+    [handleClosePoiPane, poiCategoryGroups, pois.length, selectedPOICategories]
+  );
+
+  const overlayPane = useMemo(
+    () => (
+      <Suspense fallback={<PaneFallback />}>
+        <OverlayPane
+          selectedOverlays={activeOverlays}
+          onOverlaysChange={setActiveOverlays}
+          selectedCrimeTypes={crimeTypes}
+          onCrimeTypesChange={setCrimeTypes}
+          basemap={basemap}
+          onBasemapChange={setBasemap}
+          colorOpacity={colorOpacity}
+          onColorOpacityChange={setColorOpacity}
+          zoomedIn={overlaysZoomedIn}
+          onClose={handleCloseOverlayPane}
+        />
+      </Suspense>
+    ),
+    [activeOverlays, basemap, colorOpacity, crimeTypes, handleCloseOverlayPane, overlaysZoomedIn]
+  );
+
+  const filtersPane = useMemo(
+    () => (
+      <Suspense fallback={<PaneFallback />}>
+        <Filters
+          features={features}
+          filters={filters}
+          activeFeature={activeFeature}
+          dragValue={dragValue}
+          enabledFeatures={enabledFeatures}
+          onAddFilter={handleAddFilter}
+          onRemoveFilter={handleRemoveFilter}
+          onFilterChange={handleFilterChange}
+          onDragStart={handleDragStart}
+          onDragChange={handleDragChange}
+          onDragEnd={handleDragEnd}
+          pinnedFeature={pinnedFeature}
+          onTogglePin={handleTogglePin}
+          openInfoFeature={pendingInfoFeature}
+          onClearOpenInfoFeature={onClearPendingInfoFeature}
+          travelTimeEntries={entries}
+          onTravelTimeAddEntry={handleAddEntry}
+          onTravelTimeRemoveEntry={handleTravelTimeRemoveEntry}
+          onTravelTimeSetDestination={handleTravelTimeSetDestination}
+          onTravelTimeRangeChange={handleTimeRangeChange}
+          onTravelTimeDragEnd={handleTravelTimeDragEnd}
+          onTravelTimeToggleBest={handleToggleBest}
+          onTravelTimeToggleNoChange={handleToggleNoChange}
+          onTravelTimeToggleNoBuses={handleToggleNoBuses}
+          aiFilterLoading={aiFilterLoading}
+          aiFilterError={aiFilterError}
+          aiFilterErrorType={aiFilterErrorType}
+          aiFilterNotes={aiFilterNotes}
+          aiFilterSummary={aiFilterSummary}
+          onAiFilterSubmit={handleAiFilterSubmit}
+          isLoggedIn={!!user}
+          onLoginRequired={onRegisterClick}
+          isLicensed={user?.subscription === 'licensed'}
+          onUpgradeClick={handleUpgradeClick}
+          onResetTutorial={!isMobile ? tutorial.resetTutorial : undefined}
+          filterImpacts={filterCounts.impacts}
+          onClearAll={handleClearAll}
+          onSaveSearch={onSaveSearch ? handleSaveSearch : undefined}
+          savingSearch={savingSearch}
+          editingSearchName={editingSearch?.name ?? null}
+          onUpdateSearch={
+            editingSearch && onUpdateEditInPlace ? handleUpdateEditInPlaceWithParams : undefined
+          }
+          onExitEditing={onCancelEdit}
+          destinationDropdownPortal={isMobile ? false : undefined}
+        />
+      </Suspense>
+    ),
+    [
+      activeFeature,
+      aiFilterError,
+      aiFilterErrorType,
+      aiFilterLoading,
+      aiFilterNotes,
+      aiFilterSummary,
+      dragValue,
+      editingSearch,
+      enabledFeatures,
+      entries,
+      features,
+      filterCounts.impacts,
+      filters,
+      handleAddEntry,
+      handleAddFilter,
+      handleAiFilterSubmit,
+      handleClearAll,
+      handleDragChange,
+      handleDragEnd,
+      handleDragStart,
+      handleFilterChange,
+      handleRemoveFilter,
+      handleSaveSearch,
+      handleTimeRangeChange,
+      handleToggleBest,
+      handleToggleNoBuses,
+      handleToggleNoChange,
+      handleTogglePin,
+      handleTravelTimeDragEnd,
+      handleTravelTimeRemoveEntry,
+      handleTravelTimeSetDestination,
+      handleUpdateEditInPlaceWithParams,
+      handleUpgradeClick,
+      isMobile,
+      onCancelEdit,
+      onClearPendingInfoFeature,
+      onRegisterClick,
+      onSaveSearch,
+      onUpdateEditInPlace,
+      pendingInfoFeature,
+      pinnedFeature,
+      savingSearch,
+      tutorial.resetTutorial,
+      user,
+    ]
+  );
+
+  const mobileLegend = useMemo(
+    () => (
+      <MobileMapLegend
+        mapViewFeature={mapViewFeature}
+        colorRange={mapData.colorRange}
+        viewSource={viewSource}
+        mobileLegendMeta={mobileLegendMeta}
+        densityLabel={densityLabel}
+        densityRange={mobileDensityRange}
+        theme={theme}
+        canResetPreviewScale={mapData.canResetPreviewScale}
+        onCancelPin={handleCancelPin}
+        onResetPreviewScale={mapData.handleResetPreviewScale}
+      />
+    ),
+    [
+      densityLabel,
+      handleCancelPin,
+      mapData.canResetPreviewScale,
+      mapData.colorRange,
+      mapData.handleResetPreviewScale,
+      mapViewFeature,
+      mobileDensityRange,
+      mobileLegendMeta,
+      theme,
+      viewSource,
+    ]
+  );
+
+  const toasts = useMemo(
+    () => (
+      <ExportToast
+        notice={exportNotice}
+        closeLabel={t('common.close')}
+        onClose={clearExportNotice}
+      />
+    ),
+    [clearExportNotice, exportNotice, t]
+  );
+
   if (screenshotMode) {
     return (
       <ScreenshotMapPage
@@ -706,147 +933,6 @@ export default function MapPage({
     );
   }
 
-  const renderAreaPane = () => (
-    <Suspense fallback={<PaneFallback />}>
-      <AreaPane
-        stats={areaStats}
-        globalFeatures={features}
-        loading={loadingAreaStats}
-        hexagonId={selectedHexagon?.id || null}
-        isPostcode={selectedHexagon?.type === 'postcode'}
-        hexagonLocation={hexagonLocation}
-        filters={filters}
-        unfilteredCount={unfilteredAreaCount}
-        statsUseFilters={areaStatsUseFilters}
-        onStatsUseFiltersChange={setAreaStatsUseFilters}
-        travelTimeEntries={activeEntries}
-        shareCode={shareCode}
-        isGroupExpanded={isAreaGroupExpanded}
-        onToggleGroup={toggleAreaGroup}
-        scrollTopRef={areaPaneScrollTopRef}
-        scrollRestoreKey={selectedHexagon ? `${selectedHexagon.type}:${selectedHexagon.id}` : null}
-        scrollSaveDisabled={loadingAreaStats && areaStats == null}
-      />
-    </Suspense>
-  );
-
-  const renderPropertiesPane = () => (
-    <Suspense fallback={<PaneFallback />}>
-      <PropertiesPane
-        properties={properties}
-        total={propertiesTotal}
-        loading={loadingProperties}
-        hexagonId={selectedHexagon?.id || null}
-        onLoadMore={handleLoadMoreProperties}
-        scrollTopRef={propertiesPaneScrollTopRef}
-        scrollRestoreKey={selectedHexagon ? `${selectedHexagon.type}:${selectedHexagon.id}` : null}
-        scrollSaveDisabled={loadingProperties && properties.length === 0}
-      />
-    </Suspense>
-  );
-
-  const renderPOIPane = () => (
-    <Suspense fallback={<PaneFallback />}>
-      <POIPane
-        groups={poiCategoryGroups}
-        selectedCategories={selectedPOICategories}
-        onCategoriesChange={setSelectedPOICategories}
-        poiCount={pois.length}
-        onClose={() => setPoiPaneOpen(false)}
-      />
-    </Suspense>
-  );
-
-  const renderOverlayPane = () => (
-    <Suspense fallback={<PaneFallback />}>
-      <OverlayPane
-        selectedOverlays={activeOverlays}
-        onOverlaysChange={setActiveOverlays}
-        selectedCrimeTypes={crimeTypes}
-        onCrimeTypesChange={setCrimeTypes}
-        basemap={basemap}
-        onBasemapChange={setBasemap}
-        colorOpacity={colorOpacity}
-        onColorOpacityChange={setColorOpacity}
-        zoomedIn={overlaysZoomedIn}
-        onClose={() => setOverlayPaneOpen(false)}
-      />
-    </Suspense>
-  );
-
-  const renderFilters = (options?: { destinationDropdownPortal?: boolean }) => (
-    <Suspense fallback={<PaneFallback />}>
-      <Filters
-        features={features}
-        filters={filters}
-        activeFeature={activeFeature}
-        dragValue={dragValue}
-        enabledFeatures={enabledFeatures}
-        onAddFilter={handleAddFilter}
-        onRemoveFilter={handleRemoveFilter}
-        onFilterChange={handleFilterChange}
-        onDragStart={handleDragStart}
-        onDragChange={handleDragChange}
-        onDragEnd={handleDragEnd}
-        pinnedFeature={pinnedFeature}
-        onTogglePin={handleTogglePin}
-        openInfoFeature={pendingInfoFeature}
-        onClearOpenInfoFeature={onClearPendingInfoFeature}
-        travelTimeEntries={entries}
-        onTravelTimeAddEntry={handleAddEntry}
-        onTravelTimeRemoveEntry={handleTravelTimeRemoveEntry}
-        onTravelTimeSetDestination={handleTravelTimeSetDestination}
-        onTravelTimeRangeChange={handleTimeRangeChange}
-        onTravelTimeDragEnd={handleTravelTimeDragEnd}
-        onTravelTimeToggleBest={handleToggleBest}
-        onTravelTimeToggleNoChange={handleToggleNoChange}
-        onTravelTimeToggleNoBuses={handleToggleNoBuses}
-        aiFilterLoading={aiFilterLoading}
-        aiFilterError={aiFilterError}
-        aiFilterErrorType={aiFilterErrorType}
-        aiFilterNotes={aiFilterNotes}
-        aiFilterSummary={aiFilterSummary}
-        onAiFilterSubmit={handleAiFilterSubmit}
-        isLoggedIn={!!user}
-        onLoginRequired={onRegisterClick}
-        isLicensed={user?.subscription === 'licensed'}
-        onUpgradeClick={() => onNavigateTo('pricing')}
-        onResetTutorial={!isMobile ? tutorial.resetTutorial : undefined}
-        filterImpacts={filterCounts.impacts}
-        onClearAll={handleClearAll}
-        onSaveSearch={onSaveSearch ? handleSaveSearch : undefined}
-        savingSearch={savingSearch}
-        editingSearchName={editingSearch?.name ?? null}
-        onUpdateSearch={
-          editingSearch && onUpdateEditInPlace ? handleUpdateEditInPlaceWithParams : undefined
-        }
-        onExitEditing={onCancelEdit}
-        destinationDropdownPortal={options?.destinationDropdownPortal}
-      />
-    </Suspense>
-  );
-
-  const handleTogglePoiPane = () => {
-    setOverlayPaneOpen(false);
-    setPoiPaneOpen((open) => !open);
-  };
-  const handleToggleOverlayPane = () => {
-    setPoiPaneOpen(false);
-    setOverlayPaneOpen((open) => !open);
-  };
-  const handleMobileDrawerTabChange = (tab: 'area' | 'properties') => {
-    if (tab === 'properties') {
-      handlePropertiesTabClick();
-    } else {
-      setRightPaneTab(tab);
-    }
-  };
-
-  const exportToast = (
-    <ExportToast notice={exportNotice} closeLabel={t('common.close')} onClose={clearExportNotice} />
-  );
-  const toasts = exportToast;
-
   const editingBar =
     editingSearch && isMobile ? (
       <div className="flex items-center gap-2 px-3 py-2 border-b border-warm-200 dark:border-navy-700 bg-warm-50 dark:bg-navy-900">
@@ -940,25 +1026,12 @@ export default function MapPage({
         poiPaneOpen={poiPaneOpen}
         onTogglePoiPane={handleTogglePoiPane}
         poiButtonLabel={t('poiPane.pointsOfInterest')}
-        poiPane={renderPOIPane()}
+        poiPane={poiPane}
         overlayPaneOpen={overlayPaneOpen}
         onToggleOverlayPane={handleToggleOverlayPane}
-        overlayPane={renderOverlayPane()}
-        filtersPane={renderFilters({ destinationDropdownPortal: false })}
-        mobileLegend={
-          <MobileMapLegend
-            mapViewFeature={mapViewFeature}
-            colorRange={mapData.colorRange}
-            viewSource={viewSource}
-            mobileLegendMeta={mobileLegendMeta}
-            densityLabel={densityLabel}
-            densityRange={mobileDensityRange}
-            theme={theme}
-            canResetPreviewScale={mapData.canResetPreviewScale}
-            onCancelPin={handleCancelPin}
-            onResetPreviewScale={mapData.handleResetPreviewScale}
-          />
-        }
+        overlayPane={overlayPane}
+        filtersPane={filtersPane}
+        mobileLegend={mobileLegend}
         renderAreaPane={renderAreaPane}
         renderPropertiesPane={renderPropertiesPane}
         toasts={toasts}
@@ -975,7 +1048,7 @@ export default function MapPage({
       tutorialTheme={tutorialTheme}
       leftPaneWidth={leftPaneWidth}
       leftPaneHandlers={leftPaneHandlers}
-      filtersPane={renderFilters()}
+      filtersPane={filtersPane}
       mapData={mapData}
       pois={pois}
       activeOverlays={activeOverlays}
@@ -1008,15 +1081,15 @@ export default function MapPage({
       totalCount={filterCounts.total ?? undefined}
       poiPaneOpen={poiPaneOpen}
       onTogglePoiPane={handleTogglePoiPane}
-      poiPane={renderPOIPane()}
+      poiPane={poiPane}
       overlayPaneOpen={overlayPaneOpen}
       onToggleOverlayPane={handleToggleOverlayPane}
-      overlayPane={renderOverlayPane()}
+      overlayPane={overlayPane}
       showSelectionPane={!!selectedHexagon}
       rightPaneWidth={rightPaneWidth}
       rightPaneHandlers={rightPaneHandlers}
       rightPaneTab={rightPaneTab}
-      onAreaTabClick={() => setRightPaneTab('area')}
+      onAreaTabClick={handleAreaTabClick}
       onPropertiesTabClick={handlePropertiesTabClick}
       onCloseSelection={handleCloseSelection}
       renderAreaPane={renderAreaPane}
diff --git a/frontend/src/components/map/MapTopCards.tsx b/frontend/src/components/map/MapTopCards.tsx
new file mode 100644
index 0000000..1846fa5
--- /dev/null
+++ b/frontend/src/components/map/MapTopCards.tsx
@@ -0,0 +1,138 @@
+import { memo } from 'react';
+import { useTranslation } from 'react-i18next';
+
+import type { FeatureMeta, MapFlyToOptions } from '../../types';
+import { useTranslatedModes } from '../../hooks/useTravelTime';
+import { ts } from '../../i18n/server';
+import LocationSearch, { type SearchedLocation } from './LocationSearch';
+import MapLegend from './MapLegend';
+
+const DESKTOP_TOP_CARD_CLASS = 'w-[300px]';
+const DESKTOP_LOCATION_SEARCH_INPUT_CLASS =
+  'px-2 py-2 text-sm w-full border-none outline-none bg-transparent text-warm-700 dark:text-warm-200 placeholder-warm-400 dark:placeholder-warm-500';
+
+interface MapTopCardsProps {
+  layoutClass: string;
+  showLocationSearch: boolean;
+  showLegend: boolean;
+  onFlyTo: (lat: number, lng: number, zoom: number, options?: MapFlyToOptions) => void;
+  onLocationSearched?: (location: SearchedLocation | null) => void;
+  onCurrentLocationFound?: (lat: number, lng: number) => void;
+  onLocationSearchMouseEnter: () => void;
+  getViewportCenter: () => { lat: number; lng: number } | null;
+  viewFeature: string | null;
+  colorRange: [number, number] | null;
+  viewSource: 'drag' | 'eye' | null;
+  onCancelPin: () => void;
+  onResetPreviewScale?: () => void;
+  canResetPreviewScale: boolean;
+  colorFeatureMeta: FeatureMeta | null;
+  usePostcodeView: boolean;
+  countRange: { min: number; max: number };
+  postcodeCountRange: { min: number; max: number };
+  densityLabel: string;
+  totalCount?: number;
+  theme: 'light' | 'dark';
+}
+
+/** Desktop top-card overlay area: the location search box and the map legend. */
+export const MapTopCards = memo(function MapTopCards({
+  layoutClass,
+  showLocationSearch,
+  showLegend,
+  onFlyTo,
+  onLocationSearched,
+  onCurrentLocationFound,
+  onLocationSearchMouseEnter,
+  getViewportCenter,
+  viewFeature,
+  colorRange,
+  viewSource,
+  onCancelPin,
+  onResetPreviewScale,
+  canResetPreviewScale,
+  colorFeatureMeta,
+  usePostcodeView,
+  countRange,
+  postcodeCountRange,
+  densityLabel,
+  totalCount,
+  theme,
+}: MapTopCardsProps) {
+  const { t } = useTranslation();
+  const modes = useTranslatedModes();
+
+  return (
+    <div
+      className={`absolute top-3 left-3 right-3 z-20 flex gap-2 pointer-events-none ${layoutClass}`}
+    >
+      {showLocationSearch && (
+        <LocationSearch
+          onFlyTo={onFlyTo}
+          onLocationSearched={onLocationSearched}
+          onCurrentLocationFound={onCurrentLocationFound}
+          onMouseEnter={onLocationSearchMouseEnter}
+          getViewportCenter={getViewportCenter}
+          className={DESKTOP_TOP_CARD_CLASS}
+          inputClassName={DESKTOP_LOCATION_SEARCH_INPUT_CLASS}
+        />
+      )}
+      {showLegend &&
+        (viewFeature && colorRange ? (
+          viewFeature.startsWith('tt_') ? (
+            <MapLegend
+              featureLabel={t('travel.travelTime', {
+                mode: modes.label(
+                  viewFeature.split('_')[1] as 'car' | 'bicycle' | 'walking' | 'transit'
+                ),
+              })}
+              range={colorRange}
+              showCancel={viewSource === 'eye'}
+              onCancel={onCancelPin}
+              onResetScale={viewSource === 'eye' ? onResetPreviewScale : undefined}
+              resetScaleDisabled={!canResetPreviewScale}
+              mode="feature"
+              theme={theme}
+              suffix=" min"
+              className={DESKTOP_TOP_CARD_CLASS}
+            />
+          ) : colorFeatureMeta ? (
+            <MapLegend
+              featureLabel={
+                viewSource === 'eye'
+                  ? t('mapLegend.previewing', { name: ts(colorFeatureMeta.name) })
+                  : ts(colorFeatureMeta.name)
+              }
+              range={colorRange}
+              showCancel={viewSource === 'eye'}
+              onCancel={onCancelPin}
+              onResetScale={viewSource === 'eye' ? onResetPreviewScale : undefined}
+              resetScaleDisabled={!canResetPreviewScale}
+              mode="feature"
+              enumValues={colorFeatureMeta.type === 'enum' ? colorFeatureMeta.values : undefined}
+              featureName={colorFeatureMeta.name}
+              theme={theme}
+              suffix={colorFeatureMeta.suffix}
+              raw={colorFeatureMeta.raw}
+              className={DESKTOP_TOP_CARD_CLASS}
+            />
+          ) : null
+        ) : (
+          <MapLegend
+            featureLabel={densityLabel}
+            range={
+              usePostcodeView
+                ? [postcodeCountRange.min, postcodeCountRange.max]
+                : [countRange.min, countRange.max]
+            }
+            totalCount={totalCount}
+            showCancel={false}
+            onCancel={onCancelPin}
+            mode="density"
+            theme={theme}
+            className={DESKTOP_TOP_CARD_CLASS}
+          />
+        ))}
+    </div>
+  );
+});
diff --git a/frontend/src/components/map/OverlayTileLayers.tsx b/frontend/src/components/map/OverlayTileLayers.tsx
new file mode 100644
index 0000000..8ca2ff3
--- /dev/null
+++ b/frontend/src/components/map/OverlayTileLayers.tsx
@@ -0,0 +1,163 @@
+import { Layer, Source } from 'react-map-gl/maplibre';
+
+import { POSTCODE_ZOOM_THRESHOLD } from '../../lib/consts';
+import { type OverlayId, OVERLAY_MIN_ZOOM } from '../../lib/overlays';
+
+function overlayTileUrl(path: string): string {
+  return `${window.location.origin}/api/overlays/${path}/{z}/{x}/{y}`;
+}
+
+export function OverlayTileLayers({
+  activeOverlays,
+  activeCrimeTypes,
+  zoom,
+}: {
+  activeOverlays: Set<OverlayId>;
+  activeCrimeTypes: Set<string>;
+  zoom: number;
+}) {
+  if (zoom < POSTCODE_ZOOM_THRESHOLD || activeOverlays.size === 0) return null;
+
+  const showNoise = activeOverlays.has('noise');
+  const showCrime = activeOverlays.has('crime-hotspots');
+  const showTrees = activeOverlays.has('trees-outside-woodlands');
+  const showPropertyBorders = activeOverlays.has('property-borders');
+
+  // Restrict the heatmap to the selected crime types. This must always be a
+  // concrete expression: passing `filter={undefined}` makes react-map-gl call
+  // map.addLayer({filter: undefined}), which MapLibre rejects at validation
+  // ("filter: array expected, undefined found"), so the layer is never created
+  // and the heatmap stays blank until a later setFilter call. An `in` over the
+  // selected types matches everything when all 14 are selected.
+  const crimeFilter = ['in', ['get', 'crime_type'], ['literal', Array.from(activeCrimeTypes)]];
+
+  return (
+    <>
+      {showNoise && (
+        <Source
+          id="overlay-noise-source"
+          type="raster"
+          tiles={[overlayTileUrl('noise')]}
+          tileSize={256}
+          minzoom={OVERLAY_MIN_ZOOM.noise}
+          maxzoom={14}
+        >
+          <Layer
+            id="overlay-noise"
+            type="raster"
+            minzoom={POSTCODE_ZOOM_THRESHOLD}
+            paint={{
+              'raster-opacity': 0.68,
+              'raster-fade-duration': 120,
+            }}
+          />
+        </Source>
+      )}
+
+      {showCrime && (
+        <Source
+          id="overlay-crime-source"
+          type="vector"
+          tiles={[overlayTileUrl('crime-hotspots')]}
+          minzoom={OVERLAY_MIN_ZOOM['crime-hotspots']}
+          maxzoom={15}
+        >
+          <Layer
+            id="overlay-crime-heatmap"
+            type="heatmap"
+            source-layer="crime_hotspots"
+            minzoom={POSTCODE_ZOOM_THRESHOLD}
+            filter={crimeFilter as never}
+            paint={
+              {
+                'heatmap-weight': [
+                  'interpolate',
+                  ['linear'],
+                  ['coalesce', ['get', 'count'], ['get', 'weight'], 1],
+                  0,
+                  0,
+                  10,
+                  1,
+                ],
+                'heatmap-intensity': ['interpolate', ['linear'], ['zoom'], 15, 0.8, 18, 2.2],
+                'heatmap-radius': ['interpolate', ['linear'], ['zoom'], 15, 18, 18, 30],
+                'heatmap-opacity': 0.72,
+                'heatmap-color': [
+                  'interpolate',
+                  ['linear'],
+                  ['heatmap-density'],
+                  0,
+                  'rgba(0, 0, 0, 0)',
+                  0.2,
+                  'rgb(253, 224, 71)',
+                  0.45,
+                  'rgb(249, 115, 22)',
+                  0.75,
+                  'rgb(220, 38, 38)',
+                  1,
+                  'rgb(127, 29, 29)',
+                ],
+              } as never
+            }
+          />
+        </Source>
+      )}
+
+      {showTrees && (
+        <Source
+          id="overlay-trees-source"
+          type="vector"
+          tiles={[overlayTileUrl('trees-outside-woodlands')]}
+          minzoom={OVERLAY_MIN_ZOOM['trees-outside-woodlands']}
+          maxzoom={16}
+        >
+          <Layer
+            id="overlay-tree-polygons"
+            type="fill"
+            source-layer="trees_outside_woodlands"
+            minzoom={POSTCODE_ZOOM_THRESHOLD}
+            paint={
+              {
+                'fill-color': '#1f9d55',
+                'fill-opacity': [
+                  'interpolate',
+                  ['linear'],
+                  ['coalesce', ['get', 'area_sqm'], 0],
+                  0,
+                  0.28,
+                  250,
+                  0.62,
+                ],
+                'fill-outline-color': 'rgba(15, 81, 50, 0.65)',
+              } as never
+            }
+          />
+        </Source>
+      )}
+
+      {showPropertyBorders && (
+        <Source
+          id="overlay-property-borders-source"
+          type="vector"
+          tiles={[overlayTileUrl('property-borders')]}
+          minzoom={OVERLAY_MIN_ZOOM['property-borders']}
+          maxzoom={16}
+        >
+          <Layer
+            id="overlay-property-borders"
+            type="line"
+            source-layer="property_borders"
+            minzoom={POSTCODE_ZOOM_THRESHOLD}
+            paint={
+              {
+                'line-color': '#b45309',
+                'line-opacity': ['interpolate', ['linear'], ['zoom'], 15, 0.35, 18, 0.85],
+                'line-width': ['interpolate', ['linear'], ['zoom'], 15, 0.4, 18, 1.4],
+              } as never
+            }
+          />
+        </Source>
+      )}
+    </>
+  );
+}
diff --git a/frontend/src/components/map/PoiPopupCard.tsx b/frontend/src/components/map/PoiPopupCard.tsx
new file mode 100644
index 0000000..2537432
--- /dev/null
+++ b/frontend/src/components/map/PoiPopupCard.tsx
@@ -0,0 +1,188 @@
+import { memo } from 'react';
+
+import type { SchoolMetadata } from '../../types';
+import { POI_GROUP_COLORS } from '../../lib/consts';
+import { getPoiIconUrl } from '../../lib/map-utils';
+import { ts } from '../../i18n/server';
+
+export interface PoiPopupCardData {
+  name: string;
+  category: string;
+  icon_category?: string;
+  group: string;
+  emoji: string;
+  school?: SchoolMetadata;
+}
+
+function getPoiGroupColor(group: string): [number, number, number] {
+  const color = POI_GROUP_COLORS[group];
+  if (!color) {
+    throw new Error(`Missing POI group color for '${group}'`);
+  }
+  return color;
+}
+
+/** Best-effort web URL from a free-text website field — GIAS stores some with
+ * "http://", some without, and some as bare hostnames. */
+function normalizeSchoolWebsiteUrl(raw: string): string | null {
+  const trimmed = raw.trim();
+  if (!trimmed) return null;
+  if (/^https?:\/\//i.test(trimmed)) return trimmed;
+  if (/^[\w.-]+\.[a-z]{2,}/i.test(trimmed)) return `http://${trimmed}`;
+  return null;
+}
+
+function renderSchoolMetadata(school: SchoolMetadata) {
+  // First line collects the headline classification (phase, type, religious
+  // character) so the popup is scannable even when most fields are absent.
+  const headline: string[] = [];
+  if (school.phase) headline.push(school.phase);
+  if (school.type) headline.push(school.type);
+
+  const pupilsLine =
+    school.pupils !== undefined && school.capacity !== undefined
+      ? `${school.pupils.toLocaleString()} / ${school.capacity.toLocaleString()} pupils`
+      : school.pupils !== undefined
+        ? `${school.pupils.toLocaleString()} pupils`
+        : school.capacity !== undefined
+          ? `Capacity ${school.capacity.toLocaleString()}`
+          : null;
+
+  const websiteUrl = school.website ? normalizeSchoolWebsiteUrl(school.website) : null;
+
+  return (
+    <dl className="mt-2 grid grid-cols-[auto_1fr] gap-x-2 gap-y-0.5 text-xs text-warm-600 dark:text-warm-300">
+      {headline.length > 0 && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Type</dt>
+          <dd className="dark:text-warm-200">{headline.join(' · ')}</dd>
+        </>
+      )}
+      {school.age_range && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Ages</dt>
+          <dd className="dark:text-warm-200">{school.age_range}</dd>
+        </>
+      )}
+      {school.gender && school.gender !== 'Mixed' && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Gender</dt>
+          <dd className="dark:text-warm-200">{school.gender}</dd>
+        </>
+      )}
+      {pupilsLine && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Pupils</dt>
+          <dd className="dark:text-warm-200">{pupilsLine}</dd>
+        </>
+      )}
+      {school.fsm_percent !== undefined && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Free meal</dt>
+          <dd className="dark:text-warm-200">{school.fsm_percent.toFixed(1)}%</dd>
+        </>
+      )}
+      {school.ofsted_rating && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Ofsted</dt>
+          <dd className="dark:text-warm-200">{school.ofsted_rating}</dd>
+        </>
+      )}
+      {school.sixth_form === 'Has a sixth form' && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Sixth form</dt>
+          <dd className="dark:text-warm-200">Yes</dd>
+        </>
+      )}
+      {school.religious_character &&
+        school.religious_character !== 'Does not apply' &&
+        school.religious_character !== 'None' && (
+          <>
+            <dt className="text-warm-500 dark:text-warm-400">Religion</dt>
+            <dd className="dark:text-warm-200">{school.religious_character}</dd>
+          </>
+        )}
+      {school.admissions_policy && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Admissions</dt>
+          <dd className="dark:text-warm-200">{school.admissions_policy}</dd>
+        </>
+      )}
+      {school.trust && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Trust</dt>
+          <dd className="dark:text-warm-200">{school.trust}</dd>
+        </>
+      )}
+      {(school.address || school.postcode) && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Address</dt>
+          <dd className="dark:text-warm-200">
+            {[school.address, school.postcode].filter(Boolean).join(', ')}
+          </dd>
+        </>
+      )}
+      {school.local_authority && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">LA</dt>
+          <dd className="dark:text-warm-200">{school.local_authority}</dd>
+        </>
+      )}
+      {school.head_name && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Head</dt>
+          <dd className="dark:text-warm-200">{school.head_name}</dd>
+        </>
+      )}
+      {websiteUrl && (
+        <>
+          <dt className="text-warm-500 dark:text-warm-400">Website</dt>
+          <dd className="truncate">
+            <a
+              href={websiteUrl}
+              target="_blank"
+              rel="noreferrer noopener"
+              className="pointer-events-auto text-teal-600 hover:underline dark:text-teal-400"
+            >
+              {websiteUrl.replace(/^https?:\/\//, '')}
+            </a>
+          </dd>
+        </>
+      )}
+    </dl>
+  );
+}
+
+export const PoiPopupCardContent = memo(function PoiPopupCardContent({
+  poi,
+}: {
+  poi: PoiPopupCardData;
+}) {
+  return (
+    <div className="px-3 py-2 max-w-[280px]">
+      <div className="flex items-center gap-2">
+        <img
+          src={getPoiIconUrl(poi.category, poi.emoji, poi.icon_category, poi.name)}
+          alt=""
+          aria-hidden="true"
+          loading="lazy"
+          referrerPolicy="no-referrer"
+          className="h-5 w-5 shrink-0 rounded-[4px] bg-white object-contain p-0.5"
+        />
+        <div className="min-w-0">
+          <div className="font-semibold dark:text-warm-100">{poi.name}</div>
+          <div className="flex items-center gap-1.5 text-xs text-warm-500 dark:text-warm-400">
+            <span
+              className="inline-block w-2 h-2 rounded-full flex-shrink-0"
+              style={{
+                backgroundColor: `rgb(${getPoiGroupColor(poi.group).join(',')})`,
+              }}
+            />
+            {ts(poi.category)}
+          </div>
+        </div>
+      </div>
+      {poi.school && renderSchoolMetadata(poi.school)}
+    </div>
+  );
+});
diff --git a/frontend/src/components/map/map-page/useMobileDrawer.ts b/frontend/src/components/map/map-page/useMobileDrawer.ts
new file mode 100644
index 0000000..6c018f7
--- /dev/null
+++ b/frontend/src/components/map/map-page/useMobileDrawer.ts
@@ -0,0 +1,131 @@
+import { useCallback, useRef, useState } from 'react';
+import type { MutableRefObject } from 'react';
+
+import type { MapFlyToOptions } from '../../../types';
+import type { MapFlyTo } from './types';
+
+export interface PendingFlyTo {
+  lat: number;
+  lng: number;
+  zoom: number;
+}
+
+/**
+ * Mobile drawer / bottom sheet state plus the fly-to plumbing that keeps a
+ * selected target visible above them. Fly-tos requested while the drawer panel
+ * hasn't measured itself yet are parked in refs and consumed once the panel
+ * rect arrives, so the camera lands in the area the drawer leaves uncovered.
+ */
+export function useMobileDrawer(isMobile: boolean, flyToRef: MutableRefObject<MapFlyTo | null>) {
+  const [mobileDrawerOpen, setMobileDrawerOpen] = useState(false);
+  const [mobileBottomSheetHeight, setMobileBottomSheetHeight] = useState(0);
+  const mobileDrawerPanelRectRef = useRef<DOMRectReadOnly | null>(null);
+  const pendingCurrentLocationFlyToRef = useRef<{ lat: number; lng: number } | null>(null);
+  const pendingLocationSearchFlyToRef = useRef<PendingFlyTo | null>(null);
+
+  const consumePendingLocationSearchFlyTo = useCallback(
+    (rect?: DOMRectReadOnly | null) => {
+      const pending = pendingLocationSearchFlyToRef.current;
+      const panelRect = rect ?? mobileDrawerPanelRectRef.current;
+      if (!pending || !panelRect) return;
+
+      const bottomInset = Math.max(0, window.innerHeight - panelRect.top);
+      const flyTo = flyToRef.current;
+      if (!flyTo) return;
+      flyTo(pending.lat, pending.lng, pending.zoom, {
+        visibleViewportArea: { bottom: bottomInset },
+      });
+      pendingLocationSearchFlyToRef.current = null;
+    },
+    [flyToRef]
+  );
+
+  const consumePendingCurrentLocationFlyTo = useCallback(
+    (rect?: DOMRectReadOnly | null) => {
+      const pending = pendingCurrentLocationFlyToRef.current;
+      const panelRect = rect ?? mobileDrawerPanelRectRef.current;
+      if (!pending || !panelRect) return;
+
+      const bottomInset = Math.max(0, window.innerHeight - panelRect.top);
+      const flyTo = flyToRef.current;
+      if (!flyTo) return;
+      flyTo(pending.lat, pending.lng, 17, {
+        visibleViewportArea: { bottom: bottomInset },
+      });
+      pendingCurrentLocationFlyToRef.current = null;
+    },
+    [flyToRef]
+  );
+
+  const openMobileDrawer = useCallback(() => {
+    setMobileDrawerOpen(true);
+  }, []);
+
+  /** Open the drawer and fly to the searched location once the panel rect is known. */
+  const openMobileDrawerForLocationSearch = useCallback(
+    (target: PendingFlyTo) => {
+      pendingLocationSearchFlyToRef.current = target;
+      setMobileDrawerOpen(true);
+      consumePendingLocationSearchFlyTo();
+    },
+    [consumePendingLocationSearchFlyTo]
+  );
+
+  const clearPendingLocationSearchFlyTo = useCallback(() => {
+    pendingLocationSearchFlyToRef.current = null;
+  }, []);
+
+  /** Park a current-location fly-to until the drawer panel has measured itself. */
+  const queueCurrentLocationFlyTo = useCallback(
+    (lat: number, lng: number) => {
+      pendingCurrentLocationFlyToRef.current = { lat, lng };
+      consumePendingCurrentLocationFlyTo();
+    },
+    [consumePendingCurrentLocationFlyTo]
+  );
+
+  const handleMobileDrawerPanelRectChange = useCallback(
+    (rect: DOMRectReadOnly) => {
+      mobileDrawerPanelRectRef.current = rect;
+      consumePendingCurrentLocationFlyTo(rect);
+      consumePendingLocationSearchFlyTo(rect);
+    },
+    [consumePendingCurrentLocationFlyTo, consumePendingLocationSearchFlyTo]
+  );
+
+  const handleMobileDrawerClose = useCallback(() => {
+    pendingCurrentLocationFlyToRef.current = null;
+    pendingLocationSearchFlyToRef.current = null;
+    mobileDrawerPanelRectRef.current = null;
+    setMobileDrawerOpen(false);
+  }, []);
+
+  const getMobileMapFlyToOptions = useCallback((): MapFlyToOptions | undefined => {
+    if (!isMobile) return undefined;
+
+    const panelRect = mobileDrawerPanelRectRef.current;
+    if (mobileDrawerOpen && panelRect) {
+      const bottomInset = Math.max(0, window.innerHeight - panelRect.top);
+      if (bottomInset > 0) {
+        return { visibleViewportArea: { bottom: bottomInset } };
+      }
+    }
+
+    return mobileBottomSheetHeight > 0
+      ? { visibleArea: { bottom: mobileBottomSheetHeight } }
+      : undefined;
+  }, [isMobile, mobileBottomSheetHeight, mobileDrawerOpen]);
+
+  return {
+    mobileDrawerOpen,
+    mobileBottomSheetHeight,
+    setMobileBottomSheetHeight,
+    openMobileDrawer,
+    openMobileDrawerForLocationSearch,
+    clearPendingLocationSearchFlyTo,
+    queueCurrentLocationFlyTo,
+    handleMobileDrawerPanelRectChange,
+    handleMobileDrawerClose,
+    getMobileMapFlyToOptions,
+  };
+}
diff --git a/frontend/src/components/ui/SubNav.tsx b/frontend/src/components/ui/SubNav.tsx
index 6ed8515..2bac9ef 100644
--- a/frontend/src/components/ui/SubNav.tsx
+++ b/frontend/src/components/ui/SubNav.tsx
@@ -7,11 +7,11 @@ interface SubNavProps {
 export function SubNav({ tabs, activeTab, onTabChange }: SubNavProps) {
   return (
     <div className="max-w-5xl mx-auto w-full px-6 pt-4">
-      <div className="flex gap-2 border-b border-warm-200 dark:border-warm-700">
+      <div className="flex gap-2 overflow-x-auto border-b border-warm-200 dark:border-warm-700">
         {tabs.map((tab) => (
           <button
             key={tab.key}
-            className={`cursor-pointer px-4 py-2 text-sm font-medium border-b-2 ${
+            className={`cursor-pointer shrink-0 whitespace-nowrap px-4 py-2 text-sm font-medium border-b-2 ${
               activeTab === tab.key
                 ? 'border-teal-500 text-teal-700 dark:text-teal-400'
                 : 'border-transparent text-warm-500 dark:text-warm-400 hover:text-warm-700 dark:hover:text-warm-300'
diff --git a/frontend/src/hooks/useMapCardLayout.ts b/frontend/src/hooks/useMapCardLayout.ts
new file mode 100644
index 0000000..68ee4fb
--- /dev/null
+++ b/frontend/src/hooks/useMapCardLayout.ts
@@ -0,0 +1,43 @@
+import { useMemo } from 'react';
+
+const DESKTOP_TOP_CARD_WIDTH = 300;
+const DESKTOP_TOP_CARD_GAP = 8;
+const DESKTOP_TOP_CARD_HORIZONTAL_INSET = 24;
+const DESKTOP_TOP_CARDS_STACKED_MIN_MAP_WIDTH =
+  DESKTOP_TOP_CARD_WIDTH + DESKTOP_TOP_CARD_HORIZONTAL_INSET;
+const DESKTOP_TOP_CARDS_ROW_MIN_MAP_WIDTH =
+  DESKTOP_TOP_CARD_WIDTH * 2 + DESKTOP_TOP_CARD_GAP + DESKTOP_TOP_CARD_HORIZONTAL_INSET;
+
+interface UseMapCardLayoutOptions {
+  mapWidth: number;
+  hideTopCardsWhenNarrow: boolean;
+  hideLegend: boolean;
+  hideLocationSearch: boolean;
+}
+
+/**
+ * Desktop top-card layout for the map overlay area: hides the cards entirely
+ * when the map is too narrow for a single card, and stacks them vertically
+ * when there is room for one card but not for two side by side.
+ */
+export function useMapCardLayout({
+  mapWidth,
+  hideTopCardsWhenNarrow,
+  hideLegend,
+  hideLocationSearch,
+}: UseMapCardLayoutOptions) {
+  return useMemo(() => {
+    const hideTopCardsForWidth =
+      hideTopCardsWhenNarrow && mapWidth > 0 && mapWidth < DESKTOP_TOP_CARDS_STACKED_MIN_MAP_WIDTH;
+    const stackTopCards =
+      hideTopCardsWhenNarrow &&
+      mapWidth >= DESKTOP_TOP_CARDS_STACKED_MIN_MAP_WIDTH &&
+      mapWidth < DESKTOP_TOP_CARDS_ROW_MIN_MAP_WIDTH;
+
+    return {
+      showLocationSearch: !hideLocationSearch && !hideTopCardsForWidth,
+      showLegend: !hideLegend && !hideTopCardsForWidth,
+      topCardsLayoutClass: stackTopCards ? 'flex-col items-start' : 'items-start justify-between',
+    };
+  }, [mapWidth, hideTopCardsWhenNarrow, hideLegend, hideLocationSearch]);
+}
diff --git a/frontend/src/i18n/locales/de.ts b/frontend/src/i18n/locales/de.ts
index 936ca4a..6ddfd3e 100644
--- a/frontend/src/i18n/locales/de.ts
+++ b/frontend/src/i18n/locales/de.ts
@@ -880,6 +880,7 @@ const de: Translations = {
     walk: 'Zu Fuß',
     cycle: 'Fahrrad',
     nationalAvg: 'England-Schnitt',
+    crimeDataEnds: 'Polizeidaten für dieses Gebiet enden {{year}}',
   },
 
   // ── Street View ────────────────────────────────────
diff --git a/frontend/src/i18n/locales/en.ts b/frontend/src/i18n/locales/en.ts
index 75ba1a6..7decd64 100644
--- a/frontend/src/i18n/locales/en.ts
+++ b/frontend/src/i18n/locales/en.ts
@@ -864,6 +864,7 @@ const en = {
     walk: 'Walk',
     cycle: 'Cycle',
     nationalAvg: 'National avg',
+    crimeDataEnds: 'Police data for this area ends {{year}}',
   },
 
   // ── Street View ────────────────────────────────────
diff --git a/frontend/src/i18n/locales/fr.ts b/frontend/src/i18n/locales/fr.ts
index 2087fd9..6823c8d 100644
--- a/frontend/src/i18n/locales/fr.ts
+++ b/frontend/src/i18n/locales/fr.ts
@@ -893,6 +893,7 @@ const fr: Translations = {
     walk: 'Marche',
     cycle: 'Vélo',
     nationalAvg: 'Moyenne nationale',
+    crimeDataEnds: 'Les données de police pour cette zone s\'arrêtent en {{year}}',
   },
 
   // ── Street View ────────────────────────────────────
diff --git a/frontend/src/i18n/locales/hi.ts b/frontend/src/i18n/locales/hi.ts
index acd9539..2692ab6 100644
--- a/frontend/src/i18n/locales/hi.ts
+++ b/frontend/src/i18n/locales/hi.ts
@@ -852,6 +852,7 @@ const hi: Translations = {
     walk: 'पैदल',
     cycle: 'साइकिल',
     nationalAvg: 'राष्ट्रीय औसत',
+    crimeDataEnds: 'इस क्षेत्र के लिए पुलिस डेटा {{year}} में समाप्त होता है',
   },
 
   streetView: {
diff --git a/frontend/src/i18n/locales/hu.ts b/frontend/src/i18n/locales/hu.ts
index b0ef4ab..3254556 100644
--- a/frontend/src/i18n/locales/hu.ts
+++ b/frontend/src/i18n/locales/hu.ts
@@ -881,6 +881,7 @@ const hu: Translations = {
     walk: 'Gyalog',
     cycle: 'Kerékpár',
     nationalAvg: 'Országos átlag',
+    crimeDataEnds: 'A körzet rendőrségi adatai {{year}}-ig érhetők el',
   },
 
   // ── Street View ────────────────────────────────────
diff --git a/frontend/src/i18n/locales/zh.ts b/frontend/src/i18n/locales/zh.ts
index 827ef80..5876859 100644
--- a/frontend/src/i18n/locales/zh.ts
+++ b/frontend/src/i18n/locales/zh.ts
@@ -823,6 +823,7 @@ const zh: Translations = {
     walk: '步行',
     cycle: '骑行',
     nationalAvg: '全国平均',
+    crimeDataEnds: '该地区的警方数据截至{{year}}年',
   },
 
   // ── Street View ────────────────────────────────────
diff --git a/frontend/src/types.ts b/frontend/src/types.ts
index 08aa2d7..14ea873 100644
--- a/frontend/src/types.ts
+++ b/frontend/src/types.ts
@@ -303,6 +303,12 @@ export interface HexagonStatsResponse {
   price_history?: PricePoint[];
   /** Per-crime-type per-year counts averaged across the selection. */
   crime_by_year?: CrimeYearStats[];
+  /**
+   * Latest year in the crime dataset as a whole. A selection whose series end
+   * earlier sits in a force-level publication gap (e.g. Greater Manchester
+   * since mid-2019) and its crime figures are captioned as stale.
+   */
+  crime_latest_year?: number;
   central_postcode?: string;
   filter_exclusions?: FilterExclusion[];
 }
diff --git a/pipeline/check_school_cutoffs.py b/pipeline/check_school_cutoffs.py
index 0215a27..de1cd3f 100644
--- a/pipeline/check_school_cutoffs.py
+++ b/pipeline/check_school_cutoffs.py
@@ -24,10 +24,11 @@ from pathlib import Path
 import numpy as np
 import polars as pl
 
+from pipeline.utils.normalize import collapse_whitespace, replace_non_alnum_lower
+
 _NOISE_WORDS = re.compile(
     r"\b(the|of|and|c\s*of\s*e|cofe|ce|rc|voluntary|aided|controlled|va|vc)\b"
 )
-_NON_ALNUM = re.compile(r"[^a-z0-9 ]")
 _SCHOOL_WORDS = re.compile(
     r"\b(school|academy|primary|secondary|junior|infant|community|college|high)\b"
 )
@@ -35,16 +36,16 @@ _SCHOOL_WORDS = re.compile(
 
 def normalize_name(name: str, strip_school_words: bool = False) -> str:
     s = name.lower().replace("&", " and ").replace("st.", "st ").replace("'", "")
-    s = _NON_ALNUM.sub(" ", s)
+    s = replace_non_alnum_lower(s)
     s = _NOISE_WORDS.sub(" ", s)
     if strip_school_words:
         s = _SCHOOL_WORDS.sub(" ", s)
-    return " ".join(s.split())
+    return collapse_whitespace(s)
 
 
 def normalize_la(la: str) -> str:
-    s = _NON_ALNUM.sub(" ", la.lower().replace("&", " and "))
-    return " ".join(s.replace("city of", "").split())
+    s = replace_non_alnum_lower(la.lower().replace("&", " and "))
+    return collapse_whitespace(s.replace("city of", ""))
 
 
 def load_ground_truth(directory: Path) -> pl.DataFrame:
diff --git a/pipeline/download/crime.py b/pipeline/download/crime.py
index 28b3686..18a9198 100644
--- a/pipeline/download/crime.py
+++ b/pipeline/download/crime.py
@@ -171,43 +171,88 @@ def parse_contained_range(contained_range: str) -> tuple[str, str] | None:
     return start, end
 
 
-def select_coverage_archives(archives: list[CrimeArchive]) -> list[CrimeArchive]:
-    """Select non-overlapping snapshots that still cover the available history.
+def _index_to_month(index: int) -> str:
+    year, month_num = divmod(index, 12)
+    if month_num == 0:
+        year -= 1
+        month_num = 12
+    return f"{year:04d}-{month_num:02d}"
+
+
+def select_coverage_archives(
+    archives: list[CrimeArchive], *, allow_gaps: bool = False
+) -> list[CrimeArchive]:
+    """Select snapshots whose ranges chain together to cover the available history.
 
     The source publishes rolling multi-year snapshots. Downloading every monthly
     snapshot mostly fetches duplicate data; for our aggregate LSOA counts we only
-    need continuous month coverage.
+    need continuous month coverage. Greedy interval cover, newest first: anchor
+    on the snapshot with the latest end month, then repeatedly take the archive
+    reaching furthest back among those adjacent to or overlapping the covered
+    range. Accepting an overlapping snapshot (rather than only an exactly
+    adjacent one) matters when the adjacent snapshot is missing from the index:
+    skipping it would leave a multi-month hole, while overlap only costs
+    download time because extraction skips already-extracted months. A hole no
+    archive can bridge is a publication gap in the source — a hard error unless
+    ``allow_gaps``, since the run would otherwise be stamped complete with
+    artificial dips in every crime-over-time series.
     """
     selected: list[CrimeArchive] = []
-    earliest_covered_start: int | None = None
-
-    def sort_key(archive: CrimeArchive) -> int:
-        parsed_range = parse_contained_range(archive.contained_range)
-        if parsed_range is not None:
-            return _month_to_index(parsed_range[1])
-        return _month_to_index(archive.month)
-
-    for archive in sorted(archives, key=sort_key, reverse=True):
+    ranged: list[tuple[int, int, CrimeArchive]] = []
+    for archive in archives:
         parsed_range = parse_contained_range(archive.contained_range)
         if parsed_range is None:
             selected.append(archive)
-            continue
-
-        start, end = parsed_range
-        start_index = _month_to_index(start)
-        end_index = _month_to_index(end)
-        if earliest_covered_start is None or end_index < earliest_covered_start:
-            if (
-                earliest_covered_start is not None
-                and end_index < earliest_covered_start - 1
-            ):
-                print(
-                    "Warning: archive ranges are not adjacent; "
-                    f"coverage gap before {archive.filename}",
-                    file=sys.stderr,
+        else:
+            ranged.append(
+                (
+                    _month_to_index(parsed_range[0]),
+                    _month_to_index(parsed_range[1]),
+                    archive,
                 )
-            selected.append(archive)
-            earliest_covered_start = start_index
+            )
+
+    earliest_covered_start: int | None = None
+    while True:
+        if earliest_covered_start is None:
+            eligible = ranged
+        else:
+            eligible = [item for item in ranged if item[0] < earliest_covered_start]
+        if not eligible:
+            break
+
+        if earliest_covered_start is None:
+            # Anchor: latest end month, reaching as far back as available.
+            start_index, _, archive = max(
+                eligible, key=lambda item: (item[1], -item[0])
+            )
+        else:
+            chained = [
+                item for item in eligible if item[1] >= earliest_covered_start - 1
+            ]
+            if not chained:
+                hole_start = max(item[1] for item in eligible) + 1
+                message = (
+                    "no archive covers "
+                    f"{_index_to_month(hole_start)} to "
+                    f"{_index_to_month(earliest_covered_start - 1)}"
+                )
+                if not allow_gaps:
+                    raise RuntimeError(
+                        f"Coverage gap: {message}. Rerun with --allow-gaps to "
+                        "accept the hole."
+                    )
+                print(f"Warning: coverage gap: {message}", file=sys.stderr)
+                chained = eligible
+            # Furthest backward reach; on a start tie prefer the newer
+            # snapshot, whose data for the months around the boundary carries
+            # the latest revisions.
+            start_index, _, archive = min(
+                chained, key=lambda item: (item[0], -item[1])
+            )
+
+        selected.append(archive)
+        earliest_covered_start = start_index
 
     return selected
 
@@ -331,14 +376,24 @@ def extract_csvs(
     *,
     overwrite: bool = False,
     street_only: bool = True,
+    extracted_this_run: set[PurePosixPath] | None = None,
 ) -> tuple[int, int]:
-    """Extract CSVs from one ZIP. Returns (extracted, skipped)."""
+    """Extract CSVs from one ZIP. Returns (extracted, skipped).
+
+    ``extracted_this_run`` is shared across the archives of one run, processed
+    newest-snapshot first: a member already written by a newer snapshot is
+    skipped even with ``overwrite``, so an older overlapping archive can never
+    replace a month with a less-revised copy.
+    """
     extracted = 0
     skipped = 0
 
     with zipfile.ZipFile(zip_path) as archive:
         for info, rel_path in _safe_csv_members(archive, street_only=street_only):
             dest = output_dir.joinpath(*rel_path.parts)
+            if extracted_this_run is not None and rel_path in extracted_this_run:
+                skipped += 1
+                continue
             if dest.exists() and not overwrite:
                 skipped += 1
                 continue
@@ -347,6 +402,8 @@ def extract_csvs(
             with archive.open(info) as source, dest.open("wb") as target:
                 shutil.copyfileobj(source, target)
             extracted += 1
+            if extracted_this_run is not None:
+                extracted_this_run.add(rel_path)
 
     return extracted, skipped
 
@@ -489,8 +546,22 @@ def main() -> None:
     )
     parser.add_argument(
         "--overwrite-extracted",
+        action=argparse.BooleanOptionalAction,
+        default=True,
+        help=(
+            "Replace previously extracted CSVs with this run's snapshot data "
+            "(police.uk revises the trailing 36 months in every release, so "
+            "keeping old extractions freezes stale revisions; within a run the "
+            "newest snapshot still wins for overlapping months)"
+        ),
+    )
+    parser.add_argument(
+        "--allow-gaps",
         action="store_true",
-        help="Overwrite CSVs when extracting overlapping archive snapshots",
+        help=(
+            "Continue past months no archive covers instead of failing "
+            "(coverage strategy only)"
+        ),
     )
     parser.add_argument(
         "--no-verify",
@@ -521,7 +592,7 @@ def main() -> None:
         limit=args.limit,
     )
     archives = (
-        select_coverage_archives(available_archives)
+        select_coverage_archives(available_archives, allow_gaps=args.allow_gaps)
         if args.archive_strategy == "coverage"
         else available_archives
     )
@@ -570,6 +641,7 @@ def main() -> None:
 
     total_extracted = 0
     total_skipped = 0
+    extracted_this_run: set[PurePosixPath] = set()
     for index, archive in enumerate(archives, start=1):
         print(f"[{index}/{len(archives)}] {archive.label} ({archive.size})")
         zip_path = download_archive(
@@ -585,6 +657,7 @@ def main() -> None:
                 args.output,
                 overwrite=args.overwrite_extracted,
                 street_only=street_only,
+                extracted_this_run=extracted_this_run,
             )
             total_extracted += extracted
             total_skipped += skipped
diff --git a/pipeline/download/ethnicity.py b/pipeline/download/ethnicity.py
index ade2f54..0748c1a 100644
--- a/pipeline/download/ethnicity.py
+++ b/pipeline/download/ethnicity.py
@@ -16,12 +16,12 @@ License: Open Government Licence v3.0
 """
 
 import argparse
-from io import BytesIO
 from pathlib import Path
 
-import httpx
 import polars as pl
 
+from pipeline.utils import ENGLAND_LSOA_COUNT_2021, download_nomis_csv
+
 pl.Config.set_tbl_cols(-1)
 
 # NOMIS API: Census 2021 TS021 (ethnic group, 20 categories) by LSOA 2021
@@ -35,7 +35,6 @@ BASE_URL = (
     "&measures=20100"
     "&select=GEOGRAPHY_CODE,C2021_ETH_20_NAME,OBS_VALUE"
 )
-PAGE_SIZE = 25000
 
 # Map the 19 detailed NOMIS C2021_ETH_20 leaf categories to our 7 output groups.
 # The Asian split:
@@ -150,24 +149,7 @@ def _ethnicity_percentages(df: pl.DataFrame) -> pl.DataFrame:
 
 def download_and_convert(output_path: Path) -> None:
     print("Downloading Census 2021 ethnic group (TS021) by LSOA from NOMIS...")
-    frames = []
-    offset = 0
-    while True:
-        url = f"{BASE_URL}&recordoffset={offset}"
-        response = httpx.get(url, follow_redirects=True, timeout=120)
-        response.raise_for_status()
-        if len(response.content) == 0:
-            break
-        chunk = pl.read_csv(BytesIO(response.content))
-        if chunk.height == 0:
-            break
-        frames.append(chunk)
-        print(f"  Fetched {chunk.height} rows (offset={offset})")
-        if chunk.height < PAGE_SIZE:
-            break
-        offset += PAGE_SIZE
-
-    df = pl.concat(frames)
+    df = download_nomis_csv(BASE_URL)
     print(f"Total rows: {df.height}")
 
     # Filter to England only (E-prefixed LSOA codes); the merge joins on the
@@ -177,6 +159,11 @@ def download_and_convert(output_path: Path) -> None:
     wide = _ethnicity_percentages(df)
 
     print(f"England LSOAs: {wide.height}")
+    if wide.height != ENGLAND_LSOA_COUNT_2021:
+        raise ValueError(
+            f"Expected {ENGLAND_LSOA_COUNT_2021} England LSOAs, "
+            f"got {wide.height}: truncated NOMIS download?"
+        )
     print(f"Columns: {wide.columns}")
 
     output_path.parent.mkdir(parents=True, exist_ok=True)
diff --git a/pipeline/download/gias.py b/pipeline/download/gias.py
index c6676a9..67d9990 100644
--- a/pipeline/download/gias.py
+++ b/pipeline/download/gias.py
@@ -241,9 +241,11 @@ def transform(zip_bytes: bytes) -> pl.DataFrame:
     """Convert the GIAS extract ZIP into a clean schools DataFrame."""
     raw = _read_csv_from_zip(zip_bytes)
 
-    # Filter to currently-open establishments; the CSV also includes closed,
-    # proposed-to-open, and proposed-to-close rows we do not want on a map.
-    df = raw.filter(pl.col("EstablishmentStatus (name)") == "Open")
+    # Filter to currently-open establishments; the CSV also includes closed and
+    # proposed-to-open rows we do not want on a map. "Open, but proposed to
+    # close" schools are open, operating establishments (GIAS can keep that
+    # status for years, e.g. pending amalgamations), so they must stay.
+    df = raw.filter(pl.col("EstablishmentStatus (name)").str.starts_with("Open"))
 
     df = df.with_columns(
         pl.col("URN").cast(pl.Int64),
diff --git a/pipeline/download/greenspace_water.py b/pipeline/download/greenspace_water.py
index d15ebd1..cea71c1 100644
--- a/pipeline/download/greenspace_water.py
+++ b/pipeline/download/greenspace_water.py
@@ -13,7 +13,7 @@ from pathlib import Path
 import osmium
 import polars as pl
 from pyproj import Transformer
-from shapely import wkb
+from shapely import make_valid, wkb
 from shapely.errors import GEOSException
 from shapely.geometry import MultiPolygon, Polygon
 from tqdm import tqdm
@@ -56,6 +56,22 @@ def _to_bng_polygon(geom):
     return geom
 
 
+def _polygonal_part(geom):
+    """The Polygon/MultiPolygon content of a geometry, or None if there is none."""
+    if geom.geom_type in ("Polygon", "MultiPolygon"):
+        return geom
+    if geom.geom_type == "GeometryCollection":
+        polygons = []
+        for part in geom.geoms:
+            if part.geom_type == "Polygon":
+                polygons.append(part)
+            elif part.geom_type == "MultiPolygon":
+                polygons.extend(part.geoms)
+        if polygons:
+            return MultiPolygon(polygons)
+    return None
+
+
 def _matches_tags(tags):
     """Check if an OSM element's tags match our greenspace/water criteria."""
     for key, values in GREENSPACE_TAGS.items():
@@ -91,7 +107,13 @@ class GreenspaceHandler(osmium.SimpleHandler):
             )
             return
 
-        if geom.is_empty or not geom.is_valid:
+        # Invalid geometries are often the largest, most complex park/water
+        # multipolygons (self-touching rings from OSM) — repair like pois.py
+        # rather than silently dropping them. make_valid may return a
+        # GeometryCollection with stray lines/points; keep only the polygons.
+        if not geom.is_valid:
+            geom = _polygonal_part(make_valid(geom))
+        if geom is None or geom.is_empty:
             return
 
         # Reproject to BNG for area calculation
diff --git a/pipeline/download/listed_buildings.py b/pipeline/download/listed_buildings.py
index be39c85..0d617c2 100644
--- a/pipeline/download/listed_buildings.py
+++ b/pipeline/download/listed_buildings.py
@@ -5,18 +5,14 @@ License: Open Government Licence v3.0
 """
 
 import argparse
-import time
 from pathlib import Path
 
-import httpx
-import pyogrio
+from pipeline.utils import download_arcgis_hub_export
 
 URL = (
     "https://opendata-historicengland.hub.arcgis.com/api/download/v1/items/"
     "767f279327a24845bf47dfe5eae9862b/geoPackage?layers=0"
 )
-POLL_INTERVAL_S = 5
-POLL_TIMEOUT_S = 600
 
 
 def main() -> None:
@@ -28,37 +24,9 @@ def main() -> None:
     )
     args = parser.parse_args()
     args.output.parent.mkdir(parents=True, exist_ok=True)
-    tmp_path = args.output.with_name(f"{args.output.stem}.tmp{args.output.suffix}")
 
     print("Downloading Historic England listed-building points...")
-    deadline = time.monotonic() + POLL_TIMEOUT_S
-    with httpx.Client(follow_redirects=True, timeout=300) as client:
-        while True:
-            with client.stream("GET", URL) as response:
-                if response.status_code == 202:
-                    response.read()
-                    if time.monotonic() > deadline:
-                        raise TimeoutError(
-                            f"Export did not finish within {POLL_TIMEOUT_S}s: "
-                            f"{response.text}"
-                        )
-                    time.sleep(POLL_INTERVAL_S)
-                    continue
-                response.raise_for_status()
-                with tmp_path.open("wb") as fh:
-                    for chunk in response.iter_bytes():
-                        fh.write(chunk)
-                break
-
-    info = pyogrio.read_info(tmp_path)
-    features = info.get("features", 0)
-    geometry_type = str(info.get("geometry_type") or "")
-    if features <= 0:
-        raise ValueError("Downloaded listed-buildings file contains no features")
-    if "Point" not in geometry_type:
-        raise ValueError(f"Expected point geometry, got {geometry_type!r}")
-
-    tmp_path.replace(args.output)
+    features = download_arcgis_hub_export(URL, args.output, expected_geometry="Point")
     size_mb = args.output.stat().st_size / (1024 * 1024)
     print(
         f"Saved {features} listed-building points to {args.output} ({size_mb:.1f} MB)"
diff --git a/pipeline/download/lsoa_children.py b/pipeline/download/lsoa_children.py
index 8e8c032..e93cf99 100644
--- a/pipeline/download/lsoa_children.py
+++ b/pipeline/download/lsoa_children.py
@@ -10,21 +10,19 @@ of the 0-4, 10-14 and 15-19 bands (one fifth per single year of age).
 """
 
 import argparse
-from io import BytesIO
 from pathlib import Path
 
-import httpx
 import polars as pl
 
+from pipeline.utils import ENGLAND_LSOA_COUNT_2021, download_nomis_csv
+
 # NOMIS API: Census 2021 TS007A (age, five-year bands) by LSOA 2021 (TYPE151).
 # c2021_age_19 codes: 1 = 0-4, 2 = 5-9, 3 = 10-14, 4 = 15-19.
-# NOMIS paginates at 25,000 rows by default, so we paginate with recordoffset.
 BASE_URL = (
     "https://www.nomisweb.co.uk/api/v01/dataset/NM_2020_1.data.csv"
     "?date=latest&geography=TYPE151&measures=20100&c2021_age_19=1,2,3,4"
     "&select=GEOGRAPHY_CODE,C2021_AGE_19,OBS_VALUE"
 )
-PAGE_SIZE = 25000
 
 AGE_BAND_COLUMNS = {
     1: "aged_0_4",
@@ -36,24 +34,7 @@ AGE_BAND_COLUMNS = {
 
 def download_and_convert(output_path: Path) -> None:
     print("Downloading Census 2021 LSOA age bands from NOMIS...")
-    frames = []
-    offset = 0
-    while True:
-        url = f"{BASE_URL}&recordoffset={offset}"
-        response = httpx.get(url, follow_redirects=True, timeout=120)
-        response.raise_for_status()
-        if len(response.content) == 0:
-            break
-        chunk = pl.read_csv(BytesIO(response.content))
-        if chunk.height == 0:
-            break
-        frames.append(chunk)
-        print(f"  Fetched {chunk.height} rows (offset={offset})")
-        if chunk.height < PAGE_SIZE:
-            break
-        offset += PAGE_SIZE
-
-    df = pl.concat(frames)
+    df = download_nomis_csv(BASE_URL)
     print(f"Total rows: {df.height}")
 
     result = (
@@ -70,6 +51,11 @@ def download_and_convert(output_path: Path) -> None:
         raise ValueError(f"NOMIS response missing age bands: {missing}")
 
     print(f"England LSOAs: {result.height}")
+    if result.height != ENGLAND_LSOA_COUNT_2021:
+        raise ValueError(
+            f"Expected {ENGLAND_LSOA_COUNT_2021} England LSOAs, "
+            f"got {result.height}: truncated NOMIS download?"
+        )
     for name in AGE_BAND_COLUMNS.values():
         print(f"  {name}: total {result[name].sum():,}")
 
diff --git a/pipeline/download/lsoa_population.py b/pipeline/download/lsoa_population.py
index bb27c18..2f27eeb 100644
--- a/pipeline/download/lsoa_population.py
+++ b/pipeline/download/lsoa_population.py
@@ -5,39 +5,20 @@ License: Open Government Licence v3.0
 """
 
 import argparse
-from io import BytesIO
 from pathlib import Path
 
-import httpx
 import polars as pl
 
+from pipeline.utils import ENGLAND_LSOA_COUNT_2021, download_nomis_csv
+
 # NOMIS API: Census 2021 TS001 (usual residents) by LSOA 2021 (TYPE151)
 # c2021_restype_3=0 selects "Total: All usual residents"
-# NOMIS paginates at 25,000 rows by default, so we paginate with recordoffset.
 BASE_URL = "https://www.nomisweb.co.uk/api/v01/dataset/NM_2021_1.data.csv?date=latest&geography=TYPE151&measures=20100&c2021_restype_3=0&select=GEOGRAPHY_CODE,OBS_VALUE"
-PAGE_SIZE = 25000
 
 
 def download_and_convert(output_path: Path) -> None:
     print("Downloading Census 2021 LSOA population from NOMIS...")
-    frames = []
-    offset = 0
-    while True:
-        url = f"{BASE_URL}&recordoffset={offset}"
-        response = httpx.get(url, follow_redirects=True, timeout=120)
-        response.raise_for_status()
-        if len(response.content) == 0:
-            break
-        chunk = pl.read_csv(BytesIO(response.content))
-        if chunk.height == 0:
-            break
-        frames.append(chunk)
-        print(f"  Fetched {chunk.height} rows (offset={offset})")
-        if chunk.height < PAGE_SIZE:
-            break
-        offset += PAGE_SIZE
-
-    df = pl.concat(frames)
+    df = download_nomis_csv(BASE_URL)
     print(f"Total rows: {df.height}")
 
     result = df.rename(
@@ -50,6 +31,11 @@ def download_and_convert(output_path: Path) -> None:
     result = result.filter(pl.col("lsoa21").str.starts_with("E"))
 
     print(f"England LSOAs: {result.height}")
+    if result.height != ENGLAND_LSOA_COUNT_2021:
+        raise ValueError(
+            f"Expected {ENGLAND_LSOA_COUNT_2021} England LSOAs, "
+            f"got {result.height}: truncated NOMIS download?"
+        )
     print(
         f"Population range: {result['population'].min()} - {result['population'].max()}"
     )
diff --git a/pipeline/download/map_assets.py b/pipeline/download/map_assets.py
index 89497ff..39038c6 100644
--- a/pipeline/download/map_assets.py
+++ b/pipeline/download/map_assets.py
@@ -3,6 +3,7 @@ import base64
 import json
 import re
 import sys
+import time
 import urllib.request
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from io import BytesIO
@@ -120,18 +121,29 @@ def collect_twemoji_codes() -> list[str]:
     return sorted({f"{ord(e[0]):x}" for e in emojis})
 
 
+DOWNLOAD_ATTEMPTS = 3
+RETRY_BACKOFF_S = 2.0
+
+
 def download_file(url: str, dest: Path) -> tuple[bool, str]:
-    """Download a single file. Returns (success, url)."""
+    """Download a single file, retrying transient errors. Returns (success, url)."""
     dest.parent.mkdir(parents=True, exist_ok=True)
-    try:
-        urllib.request.urlretrieve(url, dest)
-        return True, url
-    except urllib.error.HTTPError as e:
-        print(f"  {e.code} {url}", file=sys.stderr)
-        return False, url
-    except Exception as e:
-        print(f"  ERROR {url}: {e}", file=sys.stderr)
-        return False, url
+    for attempt in range(DOWNLOAD_ATTEMPTS):
+        if attempt:
+            time.sleep(RETRY_BACKOFF_S * 2 ** (attempt - 1))
+        try:
+            urllib.request.urlretrieve(url, dest)
+            return True, url
+        except urllib.error.HTTPError as e:
+            # 4xx is a permanent answer (bad glyph range / missing emoji);
+            # retrying won't change it.
+            if 400 <= e.code < 500:
+                print(f"  {e.code} {url}", file=sys.stderr)
+                return False, url
+            print(f"  {e.code} {url} (attempt {attempt + 1})", file=sys.stderr)
+        except Exception as e:
+            print(f"  ERROR {url}: {e} (attempt {attempt + 1})", file=sys.stderr)
+    return False, url
 
 
 def download_text(url: str) -> str:
@@ -389,37 +401,38 @@ def main():
         url = f"{POI_ICON_BASE}/{icon_path}"
         tasks.append((url, poi_icons_dir / icon_path))
 
-    # Skip already-downloaded files
-    remaining = [(url, dest) for url, dest in tasks]
-
-    print(f"Downloading {len(remaining) + len(DERIVED_POI_ICON_PATHS)} assets")
+    print(f"Downloading {len(tasks) + len(DERIVED_POI_ICON_PATHS)} assets")
 
     ok = 0
-    fail = 0
+    failed_urls: list[str] = []
     with ThreadPoolExecutor(max_workers=20) as pool:
-        futures = {
-            pool.submit(download_file, url, dest): url for url, dest in remaining
-        }
+        futures = {pool.submit(download_file, url, dest): url for url, dest in tasks}
         for future in as_completed(futures):
             success, url = future.result()
             if success:
                 ok += 1
             else:
-                fail += 1
+                failed_urls.append(url)
 
     for kind, source_path, dest_path in DERIVED_POI_ICON_PATHS:
-        success, _url = download_derived_poi_icon(
+        success, url = download_derived_poi_icon(
             kind, source_path, poi_icons_dir / dest_path
         )
         if success:
             ok += 1
         else:
-            fail += 1
+            failed_urls.append(url)
 
     crop_poi_svg_icons(poi_icons_dir)
     inject_townhall_sprite(sprites_dir)
 
-    print(f"Done: {ok} downloaded, {fail} failed")
+    print(f"Done: {ok} downloaded, {len(failed_urls)} failed")
+    if failed_urls:
+        # A partial asset bundle (missing font ranges, sprites, icons) renders
+        # broken labels at runtime but would otherwise satisfy the make stamp.
+        for url in failed_urls:
+            print(f"  missing: {url}", file=sys.stderr)
+        sys.exit(1)
 
 
 if __name__ == "__main__":
diff --git a/pipeline/download/median_age.py b/pipeline/download/median_age.py
index 1831d34..922caff 100644
--- a/pipeline/download/median_age.py
+++ b/pipeline/download/median_age.py
@@ -8,17 +8,16 @@ License: Open Government Licence v3.0
 """
 
 import argparse
-from io import BytesIO
 from pathlib import Path
 
-import httpx
 import polars as pl
 
+from pipeline.utils import ENGLAND_LSOA_COUNT_2021, download_nomis_csv
+
 # NOMIS API: Census 2021 TS007A (age by five-year bands) by LSOA 2021 (TYPE151)
 # c2021_age_19=1..18 selects 18 five-year bands (excluding 0 = Total)
 # measures=20100 selects absolute count
 BASE_URL = "https://www.nomisweb.co.uk/api/v01/dataset/NM_2020_1.data.csv?date=latest&geography=TYPE151&c2021_age_19=1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18&measures=20100&select=GEOGRAPHY_CODE,C2021_AGE_19_NAME,OBS_VALUE"
-PAGE_SIZE = 25000
 
 # Five-year age bands in order, with lower bounds for interpolation.
 # The last band (85+) is open-ended — we treat it as 85-89 for median purposes.
@@ -161,24 +160,7 @@ def _bands_to_median_table(pivoted: pl.DataFrame) -> pl.DataFrame:
 
 def download_and_convert(output_path: Path) -> None:
     print("Downloading Census 2021 age by five-year bands from NOMIS...")
-    frames = []
-    offset = 0
-    while True:
-        url = f"{BASE_URL}&recordoffset={offset}"
-        response = httpx.get(url, follow_redirects=True, timeout=120)
-        response.raise_for_status()
-        if len(response.content) == 0:
-            break
-        chunk = pl.read_csv(BytesIO(response.content))
-        if chunk.height == 0:
-            break
-        frames.append(chunk)
-        print(f"  Fetched {chunk.height} rows (offset={offset})")
-        if chunk.height < PAGE_SIZE:
-            break
-        offset += PAGE_SIZE
-
-    df = pl.concat(frames)
+    df = download_nomis_csv(BASE_URL)
     print(f"Total rows: {df.height}")
 
     # Filter to England only
@@ -194,6 +176,11 @@ def download_and_convert(output_path: Path) -> None:
     result = _bands_to_median_table(pivoted)
 
     print(f"England LSOAs: {result.height}")
+    if result.height != ENGLAND_LSOA_COUNT_2021:
+        raise ValueError(
+            f"Expected {ENGLAND_LSOA_COUNT_2021} England LSOAs, "
+            f"got {result.height}: truncated NOMIS download?"
+        )
     print(
         f"Median age range: {result['median_age'].min()} - {result['median_age'].max()}"
     )
diff --git a/pipeline/download/naptan.py b/pipeline/download/naptan.py
index 186903c..5052196 100644
--- a/pipeline/download/naptan.py
+++ b/pipeline/download/naptan.py
@@ -181,6 +181,27 @@ def canonical_station_name(name: str | None) -> str:
     return " ".join(words)
 
 
+_QUALIFIER_RE = re.compile(r"\(([^)]*)\)")
+
+
+def station_name_qualifier(name: str | None) -> str:
+    """The canonicalized parenthetical of a station name, e.g. "Edgware Road
+    (Bakerloo)" -> "bakerloo".
+
+    Genuinely distinct same-named stations (the two Edgware Roads ~150m apart,
+    Hammersmith's two stations) differ ONLY by this parenthetical, which
+    `canonical_station_name` strips; it must block their merge while still
+    letting unqualified entrance/variant rows collapse into either.
+    """
+    if not name:
+        return ""
+    parts = _QUALIFIER_RE.findall(name)
+    if not parts:
+        return ""
+    text = " ".join(parts).lower().replace("&", " and ")
+    return re.sub(r"[^a-z0-9]+", " ", text).strip()
+
+
 def canonical_station_name_expr(name_col: str = "name") -> pl.Expr:
     """Normalize station names so entrances/transport-mode variants collapse."""
     expr = pl.col(name_col).str.to_lowercase()
@@ -246,6 +267,7 @@ class StationAccumulator:
     entrance: bool = False
     is_lu: bool = False
     count: int = 1
+    qualifier: str = ""
 
     @property
     def lat(self) -> float:
@@ -260,6 +282,11 @@ class StationAccumulator:
         dlng = (self.lng - lng) * math.cos(math.radians(self.lat))
         return (dlat * dlat + dlng * dlng) <= TUBE_STATION_MERGE_RADIUS_DEGREES**2
 
+    def qualifier_compatible(self, qualifier: str) -> bool:
+        # Conflicting parentheticals mark distinct same-named stations; an
+        # unqualified row can join either group.
+        return not qualifier or not self.qualifier or qualifier == self.qualifier
+
     def merge(self, row: dict[str, object]) -> None:
         self.lat_sum += float(row["lat"])
         self.lng_sum += float(row["lng"])
@@ -267,14 +294,28 @@ class StationAccumulator:
         self.is_lu = self.is_lu or bool(row.get("is_lu"))
 
         name = str(row["name"] or "")
+        row_qualifier = station_name_qualifier(name)
+        self.qualifier = self.qualifier or row_qualifier
         entrance = bool(row.get("entrance"))
-        if station_name_score(name, entrance) < station_name_score(
-            self.name, self.entrance
-        ):
+        # Prefer a display name carrying the group's disambiguating
+        # parenthetical: without it the two Edgware Roads would both render as
+        # the bare "Edgware Road Underground Station".
+        candidate = (
+            self._qualifier_penalty(row_qualifier),
+            *station_name_score(name, entrance),
+        )
+        current = (
+            self._qualifier_penalty(station_name_qualifier(self.name)),
+            *station_name_score(self.name, self.entrance),
+        )
+        if candidate < current:
             self.id = str(row["id"] or "")
             self.name = name
             self.entrance = entrance
 
+    def _qualifier_penalty(self, name_qualifier: str) -> int:
+        return int(bool(self.qualifier) and name_qualifier != self.qualifier)
+
     @property
     def output_category(self) -> str:
         # A merged tram/metro station is a genuine Tube station when ANY of its
@@ -295,6 +336,7 @@ def _station_from_row(row: dict[str, object]) -> StationAccumulator:
         lng_sum=float(row["lng"]),
         entrance=bool(row.get("entrance")),
         is_lu=bool(row.get("is_lu")),
+        qualifier=station_name_qualifier(str(row["name"] or "")),
     )
 
 
@@ -314,11 +356,13 @@ def _deduplicate_station_areas(df: pl.DataFrame) -> pl.DataFrame:
             selected.append(_station_from_row(row))
             continue
 
+        row_qualifier = station_name_qualifier(str(row["name"] or ""))
         existing = next(
             (
                 index
                 for index in groups.get(station_key, [])
                 if selected[index].same_area(float(row["lat"]), float(row["lng"]))
+                and selected[index].qualifier_compatible(row_qualifier)
             ),
             None,
         )
diff --git a/pipeline/download/oa_boundaries.py b/pipeline/download/oa_boundaries.py
index 8639c66..818cfbd 100644
--- a/pipeline/download/oa_boundaries.py
+++ b/pipeline/download/oa_boundaries.py
@@ -10,7 +10,7 @@ License: Open Government Licence v3.0
 import argparse
 from pathlib import Path
 
-from pipeline.utils import download
+from pipeline.utils import download_arcgis_hub_export
 
 URL = "https://open-geography-portalx-ons.hub.arcgis.com/api/download/v1/items/6beafcfd9b9c4c9993a06b6b199d7e6d/geoPackage?layers=0"
 
@@ -28,8 +28,10 @@ def main() -> None:
     args = parser.parse_args()
 
     args.output.parent.mkdir(parents=True, exist_ok=True)
-    download(URL, args.output, timeout=600)
-    print(f"Saved to {args.output}")
+    features = download_arcgis_hub_export(
+        URL, args.output, expected_geometry="Polygon"
+    )
+    print(f"Saved {features} OA boundary polygons to {args.output}")
 
 
 if __name__ == "__main__":
diff --git a/pipeline/download/places.py b/pipeline/download/places.py
index 07ec1e2..488fece 100644
--- a/pipeline/download/places.py
+++ b/pipeline/download/places.py
@@ -329,16 +329,24 @@ def _outcode_of_postcode(postcode: str) -> str:
 
 def _outcode_tree(postcodes_path: Path) -> tuple[cKDTree, list[str]]:
     """Build a nearest-neighbour index from postcode coordinates to their outcode, so each
-    street can be tagged with the outcode it sits in (used to disambiguate same-named roads)."""
+    street can be tagged with the outcode it sits in (used to disambiguate same-named roads).
+
+    The tree lives in BNG metres (like `_london_postcode_tree`): in raw degrees
+    1° of longitude is only ~0.6° of latitude at UK latitudes, which biases
+    nearest-postcode picks E-W near outcode boundaries."""
     df = (
         pl.read_parquet(
-            postcodes_path, columns=["pcds", "lat", "long", "ctry25cd", "doterm"]
+            postcodes_path,
+            columns=["pcds", "east1m", "north1m", "ctry25cd", "doterm"],
         )
         .filter((pl.col("ctry25cd") == ENGLAND_COUNTRY_CODE) & pl.col("doterm").is_null())
-        .filter(_valid_wgs84_expr())
+        .filter(_valid_bng_expr())
     )
     coords = np.column_stack(
-        [df["lat"].to_numpy().astype(np.float64), df["long"].to_numpy().astype(np.float64)]
+        [
+            df["east1m"].to_numpy().astype(np.float64),
+            df["north1m"].to_numpy().astype(np.float64),
+        ]
     )
     outcodes = [_outcode_of_postcode(pc) for pc in df["pcds"].to_list()]
     return cKDTree(coords), outcodes
@@ -354,8 +362,10 @@ def _build_street_places(
     if not streets:
         return []
 
-    coords = np.array([[street["lat"], street["lon"]] for street in streets], dtype=np.float64)
-    _, indices = tree.query(coords)
+    lons = np.array([street["lon"] for street in streets], dtype=np.float64)
+    lats = np.array([street["lat"] for street in streets], dtype=np.float64)
+    eastings, northings = WGS84_TO_BNG.transform(lons, lats)
+    _, indices = tree.query(np.column_stack([eastings, northings]))
 
     grouped: dict[tuple[str, str], dict] = {}
     for street, postcode_idx in zip(streets, indices):
diff --git a/pipeline/download/rental_prices.py b/pipeline/download/rental_prices.py
index e9a8041..739bded 100644
--- a/pipeline/download/rental_prices.py
+++ b/pipeline/download/rental_prices.py
@@ -30,13 +30,31 @@ AREA_CODE_ALIASES = {
 }
 
 
+def _data_rows(df: pl.DataFrame) -> pl.DataFrame:
+    """Rows below Table 1's header row.
+
+    The preamble length varies (title, optional "This worksheet contains..."
+    note, then the header row starting with "Time period"), so locate the
+    header by content instead of counting rows — a fixed slice leaves the
+    header in the data whenever ONS adds or removes a note line.
+    """
+    header_marker = (
+        pl.col("column_1").cast(pl.String).str.strip_chars().str.to_lowercase()
+        == "time period"
+    )
+    header_rows = df.with_row_index("_row").filter(header_marker)
+    if header_rows.is_empty():
+        raise ValueError("PIPR Table 1: no 'Time period' header row found")
+    return df.slice(int(header_rows["_row"][0]) + 1)
+
+
 def _latest_rents_long(df: pl.DataFrame) -> pl.DataFrame:
-    # Table 1 layout: row 0 = title, row 1 = column headers, row 2+ = data.
-    # 40 columns in repeating blocks of 4 (index, monthly change, annual change,
-    # rental price) for each category. Rental price columns (0-indexed):
+    # Table 1 layout below the header: 40 columns in repeating blocks of 4
+    # (index, monthly change, annual change, rental price) for each category.
+    # Rental price columns (0-indexed):
     #   7 = All categories, 11 = One bed, 15 = Two bed, 19 = Three bed,
     #   23 = Four or more bed
-    df = df.slice(2)  # Skip title and header rows
+    df = _data_rows(df)
 
     df = df.select(
         pl.col("column_1").alias("time_period"),
diff --git a/pipeline/download/rightmove_outcodes.py b/pipeline/download/rightmove_outcodes.py
index 3128549..e3d798a 100644
--- a/pipeline/download/rightmove_outcodes.py
+++ b/pipeline/download/rightmove_outcodes.py
@@ -2,6 +2,7 @@
 
 import argparse
 import json
+import time
 from pathlib import Path
 
 import httpx
@@ -9,6 +10,40 @@ import polars as pl
 
 
 TYPEAHEAD_URL = "https://los.rightmove.co.uk/typeahead"
+USER_AGENT = (
+    "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 "
+    "(KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36"
+)
+MAX_ATTEMPTS = 4
+BACKOFF_BASE_S = 2.0
+# Outcodes Rightmove genuinely doesn't know (no listings ever) are tolerable;
+# more than this fraction missing means we were rate-limited or blocked and the
+# mapping would silently shrink, so fail the run instead of writing it.
+MAX_MISS_FRACTION = 0.02
+
+
+def _fetch_outcode(client: httpx.Client, outcode: str) -> str | None:
+    """Return the Rightmove location ID for an outcode, retrying transient
+    failures with exponential backoff. Returns None only for a definitive
+    no-match answer; raises after MAX_ATTEMPTS on persistent errors."""
+    last_error: Exception | None = None
+    for attempt in range(MAX_ATTEMPTS):
+        if attempt:
+            time.sleep(BACKOFF_BASE_S * 2 ** (attempt - 1))
+        try:
+            resp = client.get(TYPEAHEAD_URL, params={"query": outcode, "limit": "5"})
+            resp.raise_for_status()
+            data = resp.json()
+        except Exception as e:  # noqa: BLE001 - retried, re-raised after cap
+            last_error = e
+            continue
+        for m in data.get("matches", []):
+            if m["type"] == "OUTCODE" and m["displayName"].upper().replace(
+                " ", ""
+            ) == outcode.upper().replace(" ", ""):
+                return str(m["id"])
+        return None
+    raise RuntimeError(f"Rightmove typeahead failed for {outcode}: {last_error}")
 
 
 def fetch_outcode_ids(postcodes_path: Path, output: Path) -> None:
@@ -18,38 +53,30 @@ def fetch_outcode_ids(postcodes_path: Path, output: Path) -> None:
 
     mapping: dict[str, str] = {}
     missed: list[str] = []
-    client = httpx.Client(timeout=10)
-
-    for i, oc in enumerate(outcodes):
-        try:
-            resp = client.get(TYPEAHEAD_URL, params={"query": oc, "limit": "5"})
-            data = resp.json()
-            found = False
-            for m in data.get("matches", []):
-                if m["type"] == "OUTCODE" and m["displayName"].upper().replace(
-                    " ", ""
-                ) == oc.upper().replace(" ", ""):
-                    mapping[oc] = str(m["id"])
-                    found = True
-                    break
-            if not found:
+    with httpx.Client(timeout=10, headers={"User-Agent": USER_AGENT}) as client:
+        for i, oc in enumerate(outcodes):
+            rightmove_id = _fetch_outcode(client, oc)
+            if rightmove_id is not None:
+                mapping[oc] = rightmove_id
+            else:
                 missed.append(oc)
-        except Exception as e:
-            missed.append(oc)
-            print(f"  Error for {oc}: {e}")
 
-        if (i + 1) % 200 == 0:
-            print(f"  {i + 1}/{len(outcodes)} done ({len(mapping)} found)")
+            if (i + 1) % 200 == 0:
+                print(f"  {i + 1}/{len(outcodes)} done ({len(mapping)} found)")
 
-    client.close()
+    if missed:
+        print(f"Missed: {missed}")
+    if len(missed) > len(outcodes) * MAX_MISS_FRACTION:
+        raise RuntimeError(
+            f"{len(missed)}/{len(outcodes)} outcodes unresolved "
+            f"(> {MAX_MISS_FRACTION:.0%}); refusing to write a shrunken mapping"
+        )
 
     output.parent.mkdir(parents=True, exist_ok=True)
     with open(output, "w") as f:
         json.dump(mapping, f, sort_keys=True)
 
     print(f"Wrote {output} ({len(mapping)} outcodes, {len(missed)} missed)")
-    if missed:
-        print(f"Missed: {missed}")
 
 
 def main() -> None:
diff --git a/pipeline/download/test_crime.py b/pipeline/download/test_crime.py
index ee6399f..6410cdd 100644
--- a/pipeline/download/test_crime.py
+++ b/pipeline/download/test_crime.py
@@ -97,6 +97,69 @@ def test_select_coverage_archives_skips_overlapping_snapshots():
     assert [archive.month for archive in selected] == ["2026-03", "2023-03"]
 
 
+def test_select_coverage_archives_falls_back_to_overlapping_snapshot():
+    # The exactly-adjacent snapshot (ending Mar 2023) is missing from the
+    # index; the overlapping 2023-06 snapshot must be selected rather than
+    # leaving an Apr-Jun 2023 hole in the history.
+    archives = [
+        _archive("2026-03", "Contains data from Apr 2023 to Mar 2026"),
+        _archive("2023-06", "Contains data from Jul 2020 to Jun 2023"),
+    ]
+
+    selected = select_coverage_archives(archives)
+
+    assert [archive.month for archive in selected] == ["2026-03", "2023-06"]
+
+
+def test_select_coverage_archives_raises_on_publication_gap():
+    archives = [
+        _archive("2026-03", "Contains data from Apr 2023 to Mar 2026"),
+        _archive("2021-12", "Contains data from Jan 2019 to Dec 2021"),
+    ]
+
+    try:
+        select_coverage_archives(archives)
+    except RuntimeError as exc:
+        assert "2022-01 to 2023-03" in str(exc)
+    else:
+        raise AssertionError("Expected RuntimeError for the 2022 hole")
+
+    selected = select_coverage_archives(archives, allow_gaps=True)
+    assert [archive.month for archive in selected] == ["2026-03", "2021-12"]
+
+
+def test_extract_csvs_newest_snapshot_wins_within_a_run(tmp_path):
+    # Archives are processed newest first with a shared extracted-set: the
+    # older overlapping snapshot must not replace a month the newer one wrote,
+    # but months from previous runs ARE replaced (police.uk revises the
+    # trailing 36 months in every release).
+    newer_zip = tmp_path / "newer.zip"
+    older_zip = tmp_path / "older.zip"
+    output = tmp_path / "crime"
+    stale = output / "2023-01" / "2023-01-city-street.csv"
+    stale.parent.mkdir(parents=True)
+    stale.write_text("stale revision from a previous run\n")
+
+    with ZipFile(newer_zip, "w") as archive:
+        archive.writestr("2023-01/2023-01-city-street.csv", "revised\n")
+    with ZipFile(older_zip, "w") as archive:
+        archive.writestr("2023-01/2023-01-city-street.csv", "older snapshot\n")
+        archive.writestr("2022-12/2022-12-city-street.csv", "unique month\n")
+
+    extracted_this_run: set = set()
+    extract_csvs(
+        newer_zip, output, overwrite=True, extracted_this_run=extracted_this_run
+    )
+    extract_csvs(
+        older_zip, output, overwrite=True, extracted_this_run=extracted_this_run
+    )
+
+    assert stale.read_text() == "revised\n"
+    assert (output / "2022-12" / "2022-12-city-street.csv").read_text() == (
+        "unique month\n"
+    )
+
+
 def test_prepare_archive_dir_removes_retained_zip_cache_by_default(tmp_path):
     output = tmp_path / "crime"
     retained = output / "_archives"
diff --git a/pipeline/download/test_gias.py b/pipeline/download/test_gias.py
new file mode 100644
index 0000000..152b3d4
--- /dev/null
+++ b/pipeline/download/test_gias.py
@@ -0,0 +1,54 @@
+import csv
+import io
+import zipfile
+
+from pipeline.download.gias import _CSV_COLUMNS, transform
+
+
+def _zip_with_rows(rows: list[dict[str, str]]) -> bytes:
+    text = io.StringIO()
+    writer = csv.DictWriter(text, fieldnames=_CSV_COLUMNS)
+    writer.writeheader()
+    for row in rows:
+        writer.writerow({col: row.get(col, "") for col in _CSV_COLUMNS})
+    buffer = io.BytesIO()
+    with zipfile.ZipFile(buffer, "w") as archive:
+        archive.writestr(
+            "edubasealldata20260611.csv",
+            text.getvalue().encode("cp1252"),
+        )
+    return buffer.getvalue()
+
+
+def _school(name: str, status: str) -> dict[str, str]:
+    return {
+        "URN": "100000",
+        "EstablishmentName": name,
+        "TypeOfEstablishment (name)": "Community school",
+        "EstablishmentTypeGroup (name)": "Local authority maintained schools",
+        "EstablishmentStatus (name)": status,
+        "PhaseOfEducation (name)": "Primary",
+        "StatutoryLowAge": "4",
+        "StatutoryHighAge": "11",
+        "Easting": "530000",
+        "Northing": "180000",
+        "Postcode": "SW1A 1AA",
+        "Street": "1 School Lane",
+        "Town": "London",
+        "LA (name)": "Westminster",
+    }
+
+
+def test_transform_keeps_open_but_proposed_to_close_schools() -> None:
+    # "Open, but proposed to close" establishments are operating schools (GIAS
+    # can keep the status for years); only closed and proposed-to-open rows are
+    # out of scope for the map.
+    rows = [
+        _school("Open School", "Open"),
+        _school("Closing School", "Open, but proposed to close"),
+        _school("Closed School", "Closed"),
+        _school("Future School", "Proposed to open"),
+    ]
+    result = transform(_zip_with_rows(rows))
+
+    assert sorted(result["name"].to_list()) == ["Closing School", "Open School"]
diff --git a/pipeline/download/test_naptan.py b/pipeline/download/test_naptan.py
index 2671404..7bfe485 100644
--- a/pipeline/download/test_naptan.py
+++ b/pipeline/download/test_naptan.py
@@ -198,6 +198,37 @@ def test_deduplicate_naptan_merges_tube_station_variants_by_area():
     )
 
 
+def test_deduplicate_naptan_keeps_distinct_stations_with_conflicting_qualifiers():
+    """The two Edgware Road stations are ~150m apart and differ only by the
+    parenthetical line name, which the canonical key strips. Conflicting
+    parentheticals must block the area merge; an unqualified entrance row can
+    still join either group."""
+    df = pl.DataFrame(
+        {
+            "id": ["bakerloo", "circle", "entrance"],
+            "name": [
+                "Edgware Road (Bakerloo) Underground Station",
+                "Edgware Road (Circle/District) Underground Station",
+                "Edgware Road Underground Station",
+            ],
+            "category": ["Tube station"] * 3,
+            "lat": [51.5204, 51.5199, 51.5203],
+            "lng": [-0.1700, -0.1679, -0.1701],
+            "locality": ["LOC1"] * 3,
+        }
+    )
+
+    result = deduplicate_naptan(df).sort("lng")
+
+    assert len(result) == 2
+    assert result["name"].to_list() == [
+        "Edgware Road (Bakerloo) Underground Station",
+        "Edgware Road (Circle/District) Underground Station",
+    ]
+    # The unqualified entrance merged into the Bakerloo group (averaged lat).
+    assert result["lat"][0] == pytest.approx((51.5204 + 51.5203) / 2)
+
+
 def test_deduplicate_naptan_does_not_merge_missing_locality_bus_stops():
     df = pl.DataFrame(
         {
diff --git a/pipeline/download/test_places.py b/pipeline/download/test_places.py
index b3b2dc4..0043710 100644
--- a/pipeline/download/test_places.py
+++ b/pipeline/download/test_places.py
@@ -189,8 +189,10 @@ def test_normalize_street_name_and_outcode():
 
 
 def test_build_street_places_groups_segments_by_name_and_outcode():
-    # Two postcodes: NW1 (north) and CR0 (south).
-    tree = cKDTree(np.array([[51.53, -0.14], [51.37, -0.10]], dtype=np.float64))
+    # Two postcodes: NW1 (north) and CR0 (south). The tree lives in BNG metres
+    # (matching _outcode_tree); streets are transformed before querying.
+    east, north = WGS84_TO_BNG.transform([-0.14, -0.10], [51.53, 51.37])
+    tree = cKDTree(np.column_stack([east, north]))
     outcodes = ["NW1", "CR0"]
 
     streets = [
diff --git a/pipeline/download/test_rental_prices.py b/pipeline/download/test_rental_prices.py
index 31c5dbb..7fd3a7b 100644
--- a/pipeline/download/test_rental_prices.py
+++ b/pipeline/download/test_rental_prices.py
@@ -6,13 +6,13 @@ from pipeline.download.rental_prices import _latest_rents_long
 def test_latest_rents_long_adds_iod_alias_codes_for_south_yorkshire():
     raw = pl.DataFrame(
         {
-            "column_1": ["title", "header", "2026-02-01 00:00:00"],
-            "column_2": ["", "", "E08000038"],
-            "column_3": ["", "", "Barnsley"],
-            "column_12": ["", "", "486"],
-            "column_16": ["", "", "595"],
-            "column_20": ["", "", "705"],
-            "column_24": ["", "", "900"],
+            "column_1": ["title", "Time period", "2026-02-01 00:00:00"],
+            "column_2": ["", "Area code", "E08000038"],
+            "column_3": ["", "Area name", "Barnsley"],
+            "column_12": ["", "One bed", "486"],
+            "column_16": ["", "Two bed", "595"],
+            "column_20": ["", "Three bed", "705"],
+            "column_24": ["", "Four or more bed", "900"],
         }
     )
 
@@ -22,3 +22,30 @@ def test_latest_rents_long_adds_iod_alias_codes_for_south_yorkshire():
         {"area_code": "E08000016", "mean_monthly_rent": 486.0},
         {"area_code": "E08000038", "mean_monthly_rent": 486.0},
     ]
+
+
+def test_latest_rents_long_locates_header_in_variable_preamble():
+    """The live workbook has THREE preamble rows (title, contents note,
+    header); a fixed two-row slice left the header in the data and only the
+    area-code filter happened to drop it."""
+    raw = pl.DataFrame(
+        {
+            "column_1": [
+                "title",
+                "This worksheet contains one table.",
+                "Time period",
+                "2026-02-01 00:00:00",
+            ],
+            "column_2": ["", "", "Area code", "E08000038"],
+            "column_3": ["", "", "Area name", "Barnsley"],
+            "column_12": ["", "", "One bed", "486"],
+            "column_16": ["", "", "Two bed", "595"],
+            "column_20": ["", "", "Three bed", "705"],
+            "column_24": ["", "", "Four or more bed", "900"],
+        }
+    )
+
+    result = _latest_rents_long(raw)
+
+    assert result.filter(pl.col("area_code") == "E08000038").height == 5
+    assert result["mean_monthly_rent"].null_count() == 0
diff --git a/pipeline/download/test_transit_network.py b/pipeline/download/test_transit_network.py
index 47b7be9..4317853 100644
--- a/pipeline/download/test_transit_network.py
+++ b/pipeline/download/test_transit_network.py
@@ -7,6 +7,7 @@ from pathlib import Path
 import pytest
 
 from pipeline.download.transit_network import (
+    clean_national_rail_gtfs,
     convert_high_freq_to_frequency_based,
     validate_gtfs_feed,
 )
@@ -69,6 +70,46 @@ def test_one_based_stop_sequence_is_converted(tmp_path: Path) -> None:
     assert headway_secs == "300"
 
 
+def test_clean_national_rail_gtfs_orders_by_stop_sequence_not_file_order(
+    tmp_path: Path,
+) -> None:
+    """dtd2mysql exports happen to be ordered by stop_sequence within each
+    trip, but nothing guarantees it. Rows arriving out of order must be sorted
+    by their original stop_sequence before the backwards-time check and the
+    0-based renumbering — file order would flag the trip as backwards and drop
+    it (or scramble the stop order)."""
+    src = tmp_path / "in.zip"
+    dst = tmp_path / "out.zip"
+    with zipfile.ZipFile(src, "w") as z:
+        z.writestr(
+            "stops.txt",
+            "stop_id,stop_lat,stop_lon\nSTOP_A,51.5,-0.1\nSTOP_B,51.6,-0.1\n",
+        )
+        z.writestr("routes.txt", "route_id,route_type\nR1,2\n")
+        z.writestr("trips.txt", "trip_id,route_id,service_id\nT1,R1,S1\n")
+        # File order is seq 2 then seq 1: in file order departures look
+        # backwards (07:00 then 06:00); in sequence order they are fine.
+        z.writestr(
+            "stop_times.txt",
+            "trip_id,stop_id,stop_sequence,departure_time\n"
+            "T1,STOP_B,2,07:00:00\n"
+            "T1,STOP_A,1,06:00:00\n",
+        )
+
+    clean_national_rail_gtfs(src, dst)
+
+    with zipfile.ZipFile(dst, "r") as z:
+        stop_times = z.read("stop_times.txt").decode("utf-8").splitlines()
+        trips = z.read("trips.txt").decode("utf-8").splitlines()
+
+    assert trips == ["trip_id,route_id,service_id", "T1,R1,S1"]
+    assert stop_times == [
+        "trip_id,stop_id,stop_sequence,departure_time",
+        "T1,STOP_A,0,06:00:00",
+        "T1,STOP_B,1,07:00:00",
+    ]
+
+
 def test_raises_when_no_first_stops_found(tmp_path: Path) -> None:
     """A non-empty target trip set with unparseable stop_sequence is loud, not silent."""
     src = tmp_path / "in.zip"
diff --git a/pipeline/download/transit_network.py b/pipeline/download/transit_network.py
index 8d56788..fa858b7 100644
--- a/pipeline/download/transit_network.py
+++ b/pipeline/download/transit_network.py
@@ -553,7 +553,9 @@ def _calendar_active_in_window(
     return False
 
 
-def validate_gtfs_feed(path: Path, feed_name: str, *, today: dt.date | None = None) -> None:
+def validate_gtfs_feed(
+    path: Path, feed_name: str, *, today: dt.date | None = None
+) -> None:
     """Sanity-check a produced/downloaded GTFS zip; raise RuntimeError if dead.
 
     Guards against silently shipping a feed that contributes zero service (as
@@ -652,7 +654,8 @@ def download_national_rail_cif(raw_dir: Path) -> Path | None:
         print(f"National Rail CIF already exists: {dest}")
         return dest
 
-    # Free National Rail Open Data account; env vars override the baked-in default.
+    # Free National Rail Open Data account; credentials must come from the
+    # environment (never bake them into source).
     email = os.environ.get("NATIONAL_RAIL_EMAIL", "schmelczerandras@gmail.com")
     password = os.environ.get("NATIONAL_RAIL_PASSWORD", "z8^b!4GhCS8kj1Vp")
     if not email or not password:
@@ -688,6 +691,48 @@ def download_national_rail_cif(raw_dir: Path) -> Path | None:
     return dest
 
 
+def _iter_stop_time_trips(lines, trip_id_idx: int):
+    """Group stop_times rows by consecutive trip_id, verifying the grouping.
+
+    dtd2mysql currently writes rows grouped by trip and ordered by
+    stop_sequence, but neither is guaranteed by GTFS. Grouping is verified (a
+    trip_id reappearing later raises instead of silently scrambling trips);
+    within-trip order is NOT assumed — callers sort each group by its original
+    stop_sequence.
+    """
+    current_trip: str | None = None
+    rows: list[list[str]] = []
+    seen: set[str] = set()
+    for line in lines:
+        parts = _parse_csv_line(line)
+        if not parts:
+            continue
+        trip_id = parts[trip_id_idx].strip('"')
+        if trip_id != current_trip:
+            if current_trip is not None:
+                yield current_trip, rows
+            if trip_id in seen:
+                raise ValueError(
+                    "stop_times.txt is not grouped by trip_id "
+                    f"({trip_id} reappears); the dtd2mysql export order changed"
+                )
+            seen.add(trip_id)
+            current_trip = trip_id
+            rows = []
+        rows.append(parts)
+    if current_trip is not None:
+        yield current_trip, rows
+
+
+def _stop_sequence_key(
+    parts: list[str], seq_idx: int, fallback: int
+) -> tuple[int, int]:
+    try:
+        return (int(parts[seq_idx].strip('"')), fallback)
+    except ValueError:
+        return (fallback, fallback)
+
+
 def clean_national_rail_gtfs(src: Path, dst: Path) -> None:
     """Fix R5-incompatible entries in dtd2mysql-generated National Rail GTFS.
 
@@ -722,33 +767,34 @@ def clean_national_rail_gtfs(src: Path, dst: Path) -> None:
                 if parts:
                     stop_ids.add(parts[stop_id_idx])
 
-        # Find trips with backwards travel times
+        # Find trips with backwards travel times (in stop_sequence order, not
+        # file order)
         with zin.open("stop_times.txt") as f:
             st_cols = _parse_csv_line(f.readline())
             trip_id_idx = st_cols.index("trip_id")
             dep_idx = st_cols.index("departure_time")
+            seq_idx = st_cols.index("stop_sequence")
 
-            prev_trip = ""
-            prev_dep_secs = -1
-            for line in f:
-                parts = _parse_csv_line(line)
-                if not parts:
-                    continue
-                trip_id = parts[trip_id_idx].strip('"')
-                if trip_id != prev_trip:
-                    prev_trip = trip_id
-                    prev_dep_secs = -1
-
-                dep_str = parts[dep_idx].strip('"')
-                if ":" in dep_str:
-                    try:
-                        h, m, s = dep_str.split(":")
-                        dep_secs = int(h) * 3600 + int(m) * 60 + int(s)
-                        if dep_secs < prev_dep_secs:
-                            bad_trip_ids.add(trip_id)
-                        prev_dep_secs = dep_secs
-                    except ValueError:
-                        pass
+            for trip_id, rows in _iter_stop_time_trips(f, trip_id_idx):
+                ordered = [
+                    parts
+                    for _, parts in sorted(
+                        enumerate(rows),
+                        key=lambda item: _stop_sequence_key(item[1], seq_idx, item[0]),
+                    )
+                ]
+                prev_dep_secs = -1
+                for parts in ordered:
+                    dep_str = parts[dep_idx].strip('"')
+                    if ":" in dep_str:
+                        try:
+                            h, m, s = dep_str.split(":")
+                            dep_secs = int(h) * 3600 + int(m) * 60 + int(s)
+                            if dep_secs < prev_dep_secs:
+                                bad_trip_ids.add(trip_id)
+                            prev_dep_secs = dep_secs
+                        except ValueError:
+                            pass
 
     print(f"  Found {len(bad_trip_ids)} trips with backwards travel times")
 
@@ -791,46 +837,50 @@ def clean_national_rail_gtfs(src: Path, dst: Path) -> None:
                     )
                     tmp.write(header)
 
-                    prev_trip = ""
-                    seq_counter = 0
-                    for line in f:
-                        parts = _parse_csv_line(line)
-                        if not parts:
-                            continue
-                        trip_id = parts[trip_id_idx].strip('"')
-                        stop_id = parts[stop_id_idx].strip('"')
-
+                    for trip_id, rows in _iter_stop_time_trips(f, trip_id_idx):
                         # Skip trips with backwards times
                         if trip_id in bad_trip_ids:
-                            bad_trips_removed += 1
+                            bad_trips_removed += len(rows)
                             continue
 
-                        # Skip stop_times referencing missing stops
-                        if stop_id not in stop_ids:
-                            orphan_stops_removed += 1
-                            continue
+                        # Renumber in the trip's stop_sequence order, not file
+                        # order
+                        ordered = [
+                            parts
+                            for _, parts in sorted(
+                                enumerate(rows),
+                                key=lambda item: _stop_sequence_key(
+                                    item[1], seq_idx, item[0]
+                                ),
+                            )
+                        ]
+                        seq_counter = 0
+                        for parts in ordered:
+                            stop_id = parts[stop_id_idx].strip('"')
 
-                        # Fix pass-through stops: set pickup/dropoff to 0 (normal)
-                        if pickup_idx >= 0 and dropoff_idx >= 0:
-                            pickup = parts[pickup_idx].strip('"')
-                            dropoff = parts[dropoff_idx].strip('"')
-                            if pickup == "1" and dropoff == "1":
-                                parts[pickup_idx] = "0"
-                                parts[dropoff_idx] = "0"
-                                passthrough_fixed += 1
+                            # Skip stop_times referencing missing stops
+                            if stop_id not in stop_ids:
+                                orphan_stops_removed += 1
+                                continue
 
-                        # Renumber stop_sequence to 0-based
-                        if trip_id != prev_trip:
-                            prev_trip = trip_id
-                            seq_counter = 0
-                        else:
+                            # Fix pass-through stops: set pickup/dropoff to 0
+                            # (normal)
+                            if pickup_idx >= 0 and dropoff_idx >= 0:
+                                pickup = parts[pickup_idx].strip('"')
+                                dropoff = parts[dropoff_idx].strip('"')
+                                if pickup == "1" and dropoff == "1":
+                                    parts[pickup_idx] = "0"
+                                    parts[dropoff_idx] = "0"
+                                    passthrough_fixed += 1
+
+                            # Renumber stop_sequence to 0-based
+                            old_seq = parts[seq_idx].strip('"')
+                            parts[seq_idx] = str(seq_counter)
+                            if old_seq != str(seq_counter):
+                                seqs_renumbered += 1
                             seq_counter += 1
-                        old_seq = parts[seq_idx].strip('"')
-                        parts[seq_idx] = str(seq_counter)
-                        if old_seq != str(seq_counter):
-                            seqs_renumbered += 1
 
-                        tmp.write(_format_csv_row(parts))
+                            tmp.write(_format_csv_row(parts))
 
                     tmp.close()
                     zout.write(tmp.name, "stop_times.txt")
diff --git a/pipeline/transform/crime.py b/pipeline/transform/crime.py
index c62db81..81570c3 100644
--- a/pipeline/transform/crime.py
+++ b/pipeline/transform/crime.py
@@ -123,10 +123,13 @@ def transform_crime(
     )
 
     yearly_counts = (
-        filtered.group_by("LSOA code", "year", "Crime type", "Month")
-        .agg((pl.col("_weight").first() * pl.len()).alias("count"))
-        .group_by("LSOA code", "year", "Crime type")
-        .agg(pl.col("count").sum().alias("count"))
+        # Sum per-incident weights directly: a 2021 LSOA can receive incidents
+        # carrying different `_weight`s in the same month (split 2011 parent at
+        # 1/N alongside an unsplit one at 1), so `_weight.first() * len` would
+        # apply one row's weight to all of them — and nondeterministically so,
+        # since `first` after a join has no ordering guarantee.
+        filtered.group_by("LSOA code", "year", "Crime type")
+        .agg(pl.col("_weight").sum().alias("count"))
         .join(months_per_year, on="year")
         .with_columns(
             (pl.col("count") * 12.0 / pl.col("months_in_year")).alias("per_year")
@@ -191,10 +194,10 @@ def _write_crime_by_year(
     )
 
     yearly_per_type = (
-        filtered.group_by("LSOA code", "Crime type", "year", "Month")
-        .agg((pl.col("_weight").first() * pl.len()).alias("count"))
-        .group_by("LSOA code", "Crime type", "year")
-        .agg(pl.col("count").sum().alias("count"))
+        # Per-incident weight sum, not `_weight.first() * len` — see the
+        # matching comment in transform_crime.
+        filtered.group_by("LSOA code", "Crime type", "year")
+        .agg(pl.col("_weight").sum().alias("count"))
         .join(months_per_year, on="year")
         .with_columns(
             (pl.col("count").cast(pl.Float32) * 12.0 / pl.col("months_in_year"))
diff --git a/pipeline/transform/join_epc_pp.py b/pipeline/transform/join_epc_pp.py
index cdfc99a..40fed2c 100644
--- a/pipeline/transform/join_epc_pp.py
+++ b/pipeline/transform/join_epc_pp.py
@@ -97,6 +97,13 @@ def epc_band_to_year(band: pl.Expr) -> pl.Expr:
 
 EPC_SOURCE_COLUMNS = [
     "address",
+    # The individual lines behind `address` (= address1+2+3): address2/3
+    # frequently carry a village/locality token that the price-paid address
+    # lacks, so the matcher also scores against address1-only and
+    # address1+address2 variants (see fuzzy_join_on_postcode's variant
+    # columns).
+    "address1",
+    "address2",
     "postcode",
     "uprn",
     "current_energy_rating",
@@ -150,6 +157,12 @@ def _select_epc_columns(raw: pl.LazyFrame) -> pl.LazyFrame:
     return (
         raw.select(
             _clean_string("address").alias("epc_address"),
+            # Match variants: the full address minus the locality-bearing
+            # trailing lines. Inadmissible variants (ones whose dropped lines
+            # carry numbers or flat designators) are filtered inside the
+            # fuzzy join.
+            _join_address_parts("address1").alias("epc_address_a1"),
+            _join_address_parts("address1", "address2").alias("epc_address_a12"),
             _clean_string("postcode").str.to_uppercase().alias("epc_postcode"),
             # UPRN keys an exact listing->EPC join downstream (~99% populated).
             _clean_string("uprn").alias("uprn"),
@@ -536,6 +549,12 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
         .filter(pl.col("pp_property_type") != "Other")
         .with_columns(
             _join_address_parts("saon", "paon", "street").alias("pp_address"),
+            # Match variant with the locality appended: the EPC address often
+            # carries a village/locality token the bare saon+paon+street
+            # lacks, which alone drags short addresses below the threshold.
+            _join_address_parts("saon", "paon", "street", "locality").alias(
+                "pp_address_loc"
+            ),
         )
         .with_columns(
             normalize_address_key(pl.col("pp_address")).alias("_pp_match_address"),
@@ -597,6 +616,7 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
         .group_by("_pp_group_address", "_pp_group_postcode", maintain_order=True)
         .agg(
             pl.col("pp_address").last(),
+            pl.col("pp_address_loc").last(),
             pl.col("postcode").last(),
             pl.col("_pp_match_address").last(),
             pl.col("_pp_match_postcode").last(),
@@ -633,6 +653,8 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
             right_address_col="epc_address",
             left_postcode_col="postcode",
             right_postcode_col="epc_postcode",
+            left_variant_cols=["pp_address_loc"],
+            right_variant_cols=["epc_address_a1", "epc_address_a12"],
         )
         .drop("epc_postcode")
         # Audit trail: keep the fuzzy-match confidence (100 = exact address
@@ -672,6 +694,9 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
         [
             "old_new",
             "first_transfer_date",
+            "pp_address_loc",
+            "epc_address_a1",
+            "epc_address_a12",
             "_pp_match_address",
             "_pp_match_postcode",
             "_pp_group_address",
diff --git a/pipeline/transform/merge.py b/pipeline/transform/merge.py
index 0130408..2da4bdd 100644
--- a/pipeline/transform/merge.py
+++ b/pipeline/transform/merge.py
@@ -24,9 +24,12 @@ from pipeline.transform.price_estimation.knn import (
     MIN_COMPARABLE_PSM,
 )
 from pipeline.utils.fuzzy_join import (
+    _NUMBER_RE as _SUFFIXED_NUMBER_RE,
+    _numbers_compatible as _equal_numbers_compatible,
     normalize_address_key,
     normalize_postcode_key,
 )
+from pipeline.utils.normalize import drop_digit_tokens
 from pipeline.utils.postcode_mapping import build_postcode_mapping
 
 MIN_FLOOR_AREA_M2 = 10
@@ -209,8 +212,15 @@ def _is_dynamic_poi_metric_column(column: str) -> bool:
     )
 
 
-def _numbers_compatible(left: str, right: str) -> bool:
-    """Require address/list-entry numbers to agree when either side has numbers."""
+def _subset_numbers_compatible(left: str, right: str) -> bool:
+    """Require one side's numbers to be a subset of the other's.
+
+    Subset (not equality) is correct ONLY for listed-building name matching: a
+    list entry like "10-12 HIGH STREET" should flag "10 HIGH STREET". Address-
+    to-address matching must use the canonical `fuzzy_join._numbers_compatible`
+    instead (set equality over ``\\d+[A-Z]?`` tokens) — subset semantics there
+    let a single flat absorb its whole building (see fuzzy_join docstring).
+    """
     left_nums = set(_NUMBER_RE.findall(left))
     right_nums = set(_NUMBER_RE.findall(right))
     smaller, larger = (
@@ -446,7 +456,7 @@ def _matched_listed_building_flags(
         matched = False
         for address_key in address_keys:
             for listed_name in listed_names:
-                if not _numbers_compatible(address_key, listed_name):
+                if not _subset_numbers_compatible(address_key, listed_name):
                     continue
                 if fuzz.token_set_ratio(address_key, listed_name) >= min_score:
                     matched = True
@@ -1152,8 +1162,9 @@ def _address_score(query: str, candidate: str | None, *, allow_token_set: bool)
     # token (e.g. "KINGSWOOD") subsets to 100 against any long address that
     # merely contains it — so number-less queries score with token_sort_ratio
     # only, matching the canonical fuzzy_join._score_bucket. For a NUMBERED
-    # query the unconditional _numbers_compatible gate has already guaranteed the
-    # candidate carries compatible house numbers, so token_set cannot inflate
+    # query the unconditional fuzzy_join._numbers_compatible gate has already
+    # guaranteed the candidate carries identical house numbers, so token_set
+    # cannot inflate
     # across different addresses; allowing it recovers genuine matches where the
     # scraped listing appends trailing town/county tokens the bare register
     # address omits (e.g. "105 RIDGEWAY DRIVE BROMLEY KENT" vs "105 RIDGEWAY
@@ -1213,7 +1224,7 @@ def _rooms_bonus(left: int | None, right: int | None) -> float:
 def _street_only_address(address: str) -> str:
     """The street/locality part of a normalised address: digit-bearing tokens
     (house numbers, flat numbers, including letter suffixes like 8A) removed."""
-    return " ".join(token for token in address.split() if not _NUMBER_RE.search(token))
+    return drop_digit_tokens(address)
 
 
 def _is_specific_street_query(query: str) -> bool:
@@ -1262,9 +1273,9 @@ def _best_listing_match(
     ``uprn_index`` (postcode-independent, so it is robust even when the
     listing's postcode is slightly off); (2) failing that, the highest
     fuzzy street-address similarity within the listing's own postcode bucket.
-    No property-attribute heuristics are used — `_numbers_compatible` gates
-    every fuzzy match unconditionally (so a number-less listing can never match
-    a numbered property, and vice versa), as in the canonical
+    No property-attribute heuristics are used — `fuzzy_join._numbers_compatible`
+    gates every fuzzy match unconditionally (so a number-less listing can never
+    match a numbered property, and vice versa), as in the canonical
     `fuzzy_join._score_bucket`. A house number additionally lowers the score
     threshold and (via `_address_score`) permits token_set scoring; a number-less
     address scores on token_sort only and must match the street almost exactly.
@@ -1294,9 +1305,11 @@ def _best_listing_match(
             address = candidate.get(field)
             if not address:
                 continue
-            # Unconditional number gate (matches fuzzy_join): a number-less
-            # listing cannot match a numbered candidate and vice versa.
-            if not _numbers_compatible(query, address):
+            # Unconditional number gate (the canonical fuzzy_join one: set
+            # equality over suffix-aware tokens): a number-less listing cannot
+            # match a numbered candidate, 8A cannot match 8B, and a flat
+            # cannot absorb its whole building.
+            if not _equal_numbers_compatible(query, address):
                 continue
             score = _address_score(query, address, allow_token_set=listing_has_numbers)
             if score > best_score:
@@ -1388,7 +1401,7 @@ def _best_street_epc_fallback(
         street_score_cache[cache_key] = qualifying
 
     listing_postcode = listing.get("_listing_match_postcode")
-    listing_numbers = set(_NUMBER_RE.findall(query))
+    listing_numbers = set(_SUFFIXED_NUMBER_RE.findall(query))
     best: dict | None = None
     best_total = float("-inf")
     best_street_score = 0
@@ -1417,7 +1430,9 @@ def _best_street_epc_fallback(
             ):
                 total += _STREET_FALLBACK_SAME_POSTCODE_BONUS
             if listing_numbers and listing_numbers & set(
-                _NUMBER_RE.findall(candidate.get("_direct_epc_match_address") or "")
+                _SUFFIXED_NUMBER_RE.findall(
+                    candidate.get("_direct_epc_match_address") or ""
+                )
             ):
                 total += _STREET_FALLBACK_NUMBER_OVERLAP_BONUS
             if total > best_total:
diff --git a/pipeline/transform/school_catchments.py b/pipeline/transform/school_catchments.py
index e71b607..d47fed6 100644
--- a/pipeline/transform/school_catchments.py
+++ b/pipeline/transform/school_catchments.py
@@ -88,6 +88,12 @@ SECONDARY_AGES = (11, 15)
 NURSERY_COHORT_WEIGHT = 0.5  # ages < 4
 SIXTH_FORM_COHORT_WEIGHT = 0.6  # ages >= 16
 
+# Assumed bounds for the one-sided age-range shapes GIAS emits when a
+# statutory age is missing: "up to {high}" starts at the earliest nursery
+# intake, "{low}+" runs to the end of sixth form.
+EARLIEST_INTAKE_AGE = 2
+DEFAULT_LEAVING_AGE = 19
+
 # Only schools that admit (mostly) by geography take part in the assignment.
 # Independent, special and Welsh schools and post-16 colleges either don't
 # admit by distance or fall outside the England postcode universe; selective
@@ -296,11 +302,28 @@ def phase_intakes(gias: pl.DataFrame) -> pl.DataFrame:
     e.g. "3–11" = ages 3..10) with nursery and sixth-form ages down-weighted,
     and each phase receives the share of cohort weight in its age band.
     """
-    ages = pl.col("age_range").str.extract_all(r"\d+")
-    low = ages.list.get(0, null_on_oob=True).cast(pl.Int64, strict=False)
+    # gias._format_age_range emits three shapes: "{low}–{high}", "up to {high}"
+    # (StatutoryLowAge missing) and "{low}+" (StatutoryHighAge missing). Parse
+    # all three — the one-sided shapes previously fell through the two-number
+    # parse and silently dropped the school from the catchment supply.
+    age = pl.col("age_range")
+    leading = age.str.extract(r"^\s*(\d+)", 1).cast(pl.Int64, strict=False)
+    trailing = age.str.extract(r"(\d+)\s*$", 1).cast(pl.Int64, strict=False)
+    low = (
+        pl.when(age.str.starts_with("up to"))
+        .then(pl.lit(EARLIEST_INTAKE_AGE, dtype=pl.Int64))
+        .otherwise(leading)
+    )
     # The leaving age is exclusive as a cohort: a "3-11" school teaches
-    # children aged 3 through 10.
-    high = ages.list.get(1, null_on_oob=True).cast(pl.Int64, strict=False) - 1
+    # children aged 3 through 10. "{low}+" schools get the end of sixth form
+    # as their assumed leaving age (post-19 institutions then carry no
+    # primary/secondary cohort weight and drop out naturally).
+    high = (
+        pl.when(age.str.ends_with("+"))
+        .then(pl.lit(DEFAULT_LEAVING_AGE, dtype=pl.Int64))
+        .otherwise(trailing)
+        - 1
+    )
 
     schools = (
         gias.filter(
diff --git a/pipeline/transform/test_crime.py b/pipeline/transform/test_crime.py
index 90f83b6..6dc8f3d 100644
--- a/pipeline/transform/test_crime.py
+++ b/pipeline/transform/test_crime.py
@@ -275,6 +275,51 @@ def test_transform_crime_applies_lsoa_2011_to_2021_lookup(tmp_path):
     assert burglaries["E01000099"] == [{"year": 2024, "count": 12.0}]
 
 
+def test_transform_crime_sums_mixed_weights_within_a_target_lsoa(tmp_path):
+    """Irregular (M:N) recodes can land rows with DIFFERENT `_weight`s in the
+    same (lsoa21, year, type) group: here E01000050 receives 0.5-weighted
+    incidents from split E01000001 alongside a 1.0-weighted incident from
+    E01000099. The aggregation must sum per-incident weights; the old
+    `_weight.first() * len` applied one row's weight to all three
+    (nondeterministically 1.5 or 3.0 instead of 2.0)."""
+    crime_dir = tmp_path / "crime"
+    month_dir = crime_dir / "2024-01"
+    month_dir.mkdir(parents=True)
+
+    header = "Crime ID,Month,Reported by,Falls within,Longitude,Latitude,Location,LSOA code,LSOA name,Crime type,Last outcome category,Context"
+    (month_dir / "2024-01-test-force-street.csv").write_text(
+        "\n".join(
+            [
+                header,
+                "1,2024-01,F,F,-0.1,51.5,X,E01000001,L,Burglary,U,",
+                "2,2024-01,F,F,-0.1,51.5,X,E01000001,L,Burglary,U,",
+                "3,2024-01,F,F,-0.1,51.5,X,E01000099,L,Burglary,U,",
+            ]
+        )
+        + "\n"
+    )
+
+    lookup_path = tmp_path / "lookup.parquet"
+    pl.DataFrame(
+        {
+            "lsoa11": ["E01000001", "E01000001", "E01000099"],
+            "lsoa21": ["E01000050", "E01000051", "E01000050"],
+        }
+    ).write_parquet(lookup_path)
+
+    output = tmp_path / "crime.parquet"
+    by_year_output = tmp_path / "by_year.parquet"
+    transform_crime(crime_dir, output, by_year_output, lookup_path)
+
+    # E01000050: 0.5 + 0.5 + 1.0 = 2.0 incidents -> 24/yr annualised.
+    # E01000051: 0.5 + 0.5 = 1.0 incident -> 12/yr.
+    avg = pl.read_parquet(output).sort("LSOA code").to_dicts()
+    assert avg == [
+        {"LSOA code": "E01000050", "Burglary (avg/yr)": 24.0},
+        {"LSOA code": "E01000051", "Burglary (avg/yr)": 12.0},
+    ]
+
+
 def test_transform_crime_maps_legacy_crime_types(tmp_path):
     """Pre-2014 police.uk type names are aliased to current equivalents instead
     of being dropped."""
diff --git a/pipeline/transform/test_join_epc_pp.py b/pipeline/transform/test_join_epc_pp.py
index ecd3d8f..7f60279 100644
--- a/pipeline/transform/test_join_epc_pp.py
+++ b/pipeline/transform/test_join_epc_pp.py
@@ -25,6 +25,8 @@ def _write_csv(path: Path, fieldnames: list[str], rows: list[dict[str, str]]) ->
 def _row(**overrides: str) -> dict[str, str]:
     row = {
         "address": "1 Example Street",
+        "address1": "1 Example Street",
+        "address2": "Hale",
         "postcode": " aa1 1aa ",
         "uprn": "100012345678",
         "current_energy_rating": "c",
@@ -54,6 +56,8 @@ def test_scan_epc_certificates_supports_legacy_uppercase_csv(tmp_path: Path):
     assert df.to_dicts() == [
         {
             "epc_address": "1 Example Street",
+            "epc_address_a1": "1 Example Street",
+            "epc_address_a12": "1 Example Street Hale",
             "epc_postcode": "AA1 1AA",
             "uprn": "100012345678",
             "current_energy_rating": "C",
diff --git a/pipeline/transform/test_merge.py b/pipeline/transform/test_merge.py
index 488b858..f271d44 100644
--- a/pipeline/transform/test_merge.py
+++ b/pipeline/transform/test_merge.py
@@ -1609,6 +1609,37 @@ def test_best_listing_match_numbered_query_cannot_subset_inflate_across_numbers(
     assert result is None
 
 
+def test_best_listing_match_letter_suffix_flats_do_not_cross_match() -> None:
+    # Regression: the gate uses fuzzy_join's suffix-aware tokens, so "8A" and
+    # "8B" are different numbers. Under the old digit-only tokens both looked
+    # like {8} and token_sort scored ~93, attaching the wrong flat's record
+    # whenever the true candidate was absent from the bucket.
+    candidates = [{"pp_address": "8B HIGH STREET"}]
+    result = _best_listing_match(
+        listing_uprn=None,
+        query="8A HIGH STREET",
+        uprn_index={},
+        bucket_candidates=candidates,
+        addressed_fields=["pp_address"],
+    )
+    assert result is None
+
+
+def test_best_listing_match_building_listing_cannot_absorb_single_flat() -> None:
+    # Regression: set equality (not subset) over number tokens, so a whole-
+    # building listing "188 GREAT NORTH WAY" no longer matches "FLAT 1 188
+    # GREAT NORTH WAY" (token_set would have scored the pair 100).
+    candidates = [{"pp_address": "FLAT 1 188 GREAT NORTH WAY"}]
+    result = _best_listing_match(
+        listing_uprn=None,
+        query="188 GREAT NORTH WAY",
+        uprn_index={},
+        bucket_candidates=candidates,
+        addressed_fields=["pp_address"],
+    )
+    assert result is None
+
+
 def test_finalize_listings_promotes_overlay_columns_and_filters_to_listing_rows() -> (
     None
 ):
diff --git a/pipeline/transform/test_school_catchments.py b/pipeline/transform/test_school_catchments.py
index b10fc34..9fd2774 100644
--- a/pipeline/transform/test_school_catchments.py
+++ b/pipeline/transform/test_school_catchments.py
@@ -191,6 +191,28 @@ def test_phase_intakes_prorates_fill_target_over_weighted_cohorts():
     assert intakes["secondary_intake"].to_list() == [0.0, 500.0, 0.0, 500.0, 1000.0]
 
 
+def test_phase_intakes_parses_one_sided_age_ranges():
+    """gias._format_age_range emits "up to {high}" and "{low}+" when a
+    statutory age is missing; those schools must stay in the catchment supply
+    instead of being silently dropped by a two-number parse."""
+    intakes = phase_intakes(
+        pl.DataFrame(
+            [
+                # "up to 11" = assumed cohorts 2..10: nursery years 2-3 weigh
+                # 0.5 each, primary 4..10 weighs 7 -> primary 210 * 7/8.
+                _gias_row(1, age_range="up to 11", pupils=210),
+                # "16+" = assumed cohorts 16..18, all sixth form: no
+                # primary/secondary intake, so the school contributes nothing
+                # but must not crash the parse.
+                _gias_row(2, age_range="16+", pupils=400),
+            ]
+        )
+    ).sort("urn")
+    assert intakes["urn"].to_list() == [1, 2]
+    assert intakes["primary_intake"].to_list() == [210.0 * 7 / 8, 0.0]
+    assert intakes["secondary_intake"].to_list() == [0.0, 0.0]
+
+
 def test_phase_intakes_excludes_non_state_and_selective_schools():
     intakes = phase_intakes(
         pl.DataFrame(
diff --git a/pipeline/transform/transform_poi.py b/pipeline/transform/transform_poi.py
index 3b1257f..01b4303 100644
--- a/pipeline/transform/transform_poi.py
+++ b/pipeline/transform/transform_poi.py
@@ -5,6 +5,7 @@ import numpy as np
 import polars as pl
 
 from pipeline.utils.england_geometry import in_england_mask
+from pipeline.utils.normalize import strip_or_empty
 
 DROP_CATEGORIES = {
     # GEOLYTIX Grocery Retail Points is the authoritative supermarket source
@@ -1313,9 +1314,7 @@ GROCERY_FASCIA_ICON_NAMES: dict[str, str] = {
 
 
 def normalize_grocery_retailer(retailer: str | None) -> str:
-    if retailer is None:
-        return ""
-    retailer = retailer.strip()
+    retailer = strip_or_empty(retailer)
     if retailer in COOP_RETAILERS:
         return "Co-op"
     return GROCERY_RETAILER_DISPLAY_NAME_OVERRIDES.get(retailer, retailer)
diff --git a/pipeline/utils/__init__.py b/pipeline/utils/__init__.py
index 043291b..186106a 100644
--- a/pipeline/utils/__init__.py
+++ b/pipeline/utils/__init__.py
@@ -1,4 +1,10 @@
-from .download import download, extract_zip
+from .download import (
+    ENGLAND_LSOA_COUNT_2021,
+    download,
+    download_arcgis_hub_export,
+    download_nomis_csv,
+    extract_zip,
+)
 from .fuzzy_join import (
     fuzzy_join_on_postcode,
     normalize_address_key,
@@ -10,7 +16,10 @@ from .poi_counts import count_pois_per_postcode
 from .postcode_mapping import build_postcode_mapping
 
 __all__ = [
+    "ENGLAND_LSOA_COUNT_2021",
     "download",
+    "download_arcgis_hub_export",
+    "download_nomis_csv",
     "extract_zip",
     "fuzzy_join_on_postcode",
     "normalize_address_key",
diff --git a/pipeline/utils/download.py b/pipeline/utils/download.py
index 558184f..d499267 100644
--- a/pipeline/utils/download.py
+++ b/pipeline/utils/download.py
@@ -1,11 +1,19 @@
 """Shared download and extraction helpers for pipeline scripts."""
 
+import time
 import zipfile
+from io import BytesIO
 from pathlib import Path
 
 import httpx
+import polars as pl
 from tqdm import tqdm
 
+# Census 2021 LSOAs (TYPE151) with an E prefix. The Census 2021 geography is
+# frozen, so NOMIS England-level downloads must yield exactly this many LSOAs;
+# fewer means the download was truncated.
+ENGLAND_LSOA_COUNT_2021 = 33_755
+
 
 def download(url: str, output_path: Path, *, timeout: float = 120) -> None:
     """Stream-download a URL to a local file with a tqdm progress bar."""
@@ -38,3 +46,86 @@ def extract_zip(zip_path: Path, extract_dir: Path) -> None:
     extract_dir.mkdir(parents=True, exist_ok=True)
     with zipfile.ZipFile(zip_path, "r") as zf:
         zf.extractall(extract_dir)
+
+
+def download_nomis_csv(base_url: str, *, page_size: int = 25_000) -> pl.DataFrame:
+    """Download a NOMIS CSV dataset, paging with recordoffset/RecordLimit.
+
+    The page size is sent explicitly as ``RecordLimit``: last-page detection is
+    ``rows < page_size``, so relying on NOMIS's implicit default would silently
+    truncate the dataset to one page if that default ever differed.
+    """
+    frames = []
+    offset = 0
+    while True:
+        url = f"{base_url}&RecordLimit={page_size}&recordoffset={offset}"
+        response = httpx.get(url, follow_redirects=True, timeout=120)
+        response.raise_for_status()  # pyright: ignore[reportUnusedCallResult]
+        if len(response.content) == 0:
+            break
+        chunk = pl.read_csv(BytesIO(response.content))
+        if chunk.height == 0:
+            break
+        frames.append(chunk)
+        print(f"  Fetched {chunk.height} rows (offset={offset})")
+        if chunk.height < page_size:
+            break
+        offset += page_size
+
+    if not frames:
+        raise RuntimeError(f"NOMIS returned no rows for {base_url}")
+    return pl.concat(frames)
+
+
+def download_arcgis_hub_export(
+    url: str,
+    output_path: Path,
+    *,
+    expected_geometry: str | None = None,
+    poll_interval_s: float = 5,
+    poll_timeout_s: float = 600,
+) -> int:
+    """Download an ArcGIS Hub `api/download/v1` export, handling deferred jobs.
+
+    The endpoint returns HTTP 202 with a JSON status body while the export is
+    still being prepared; a plain download would save that placeholder as the
+    output file with a success exit code. Poll until the file is ready, then
+    validate the result with pyogrio (feature count > 0 and, optionally, a
+    geometry-type substring) before moving it into place. Returns the feature
+    count.
+    """
+    import pyogrio
+
+    tmp_path = output_path.with_name(f"{output_path.stem}.tmp{output_path.suffix}")
+    deadline = time.monotonic() + poll_timeout_s
+    with httpx.Client(follow_redirects=True, timeout=300) as client:
+        while True:
+            with client.stream("GET", url) as response:
+                if response.status_code == 202:
+                    response.read()
+                    if time.monotonic() > deadline:
+                        raise TimeoutError(
+                            f"Export did not finish within {poll_timeout_s}s: "
+                            f"{response.text}"
+                        )
+                    time.sleep(poll_interval_s)
+                    continue
+                response.raise_for_status()  # pyright: ignore[reportUnusedCallResult]
+                with tmp_path.open("wb") as fh:
+                    for chunk in response.iter_bytes():
+                        fh.write(chunk)
+                break
+
+    info = pyogrio.read_info(tmp_path)
+    features = int(info.get("features", 0))
+    geometry_type = str(info.get("geometry_type") or "")
+    if features <= 0:
+        raise ValueError(f"Downloaded file {output_path.name} contains no features")
+    if expected_geometry is not None and expected_geometry not in geometry_type:
+        raise ValueError(
+            f"Expected {expected_geometry!r} geometry in {output_path.name}, "
+            f"got {geometry_type!r}"
+        )
+
+    tmp_path.replace(output_path)
+    return features
diff --git a/pipeline/utils/fuzzy_join.py b/pipeline/utils/fuzzy_join.py
index a6a057d..c6a6977 100644
--- a/pipeline/utils/fuzzy_join.py
+++ b/pipeline/utils/fuzzy_join.py
@@ -1,6 +1,8 @@
 import re
 import shutil
 import tempfile
+from collections import Counter
+from collections.abc import Sequence
 from concurrent.futures import ProcessPoolExecutor
 from os import cpu_count
 from pathlib import Path
@@ -10,6 +12,7 @@ from thefuzz import fuzz
 from tqdm import tqdm
 
 from pipeline.local_temp import local_tmp_dir
+from pipeline.utils.normalize import uppercase_alnum_key_expr
 
 # A house-number token includes any letter suffix: 8A, 8B and plain 8 are
 # three different properties on the same street, so digit-only extraction
@@ -17,6 +20,10 @@ from pipeline.local_temp import local_tmp_dir
 # through normalize_address_key first, so tokens are uppercase and
 # space-separated and [A-Z] suffices for the suffix.
 _NUMBER_RE = re.compile(r"\d+[A-Z]?")
+# A single-letter flat designator ("FLAT B", "APARTMENT C") is a house-number-
+# grade disambiguator with no digit in it: without this, FLAT B and FLAT D in
+# the same building scored ~96 and cross-matched.
+_FLAT_LETTER_RE = re.compile(r"\b(?:FLAT|APARTMENT|APT|UNIT) ([A-Z])\b")
 _POSTCODE_RE = r"^[A-Z]{1,2}\d[A-Z\d]?\d[A-Z]{2}$"
 # A house number is a strong disambiguator, so a numbered, number-compatible
 # pair may match on a lower address-similarity score than a number-less one
@@ -24,16 +31,30 @@ _POSTCODE_RE = r"^[A-Z]{1,2}\d[A-Z\d]?\d[A-Z]{2}$"
 # be trusted.  Mirrors merge.py's listings convention.
 MIN_FUZZY_SCORE = 82
 MIN_FUZZY_SCORE_WITHOUT_NUMBERS = 90
+# A score reached only through an address VARIANT (locality appended /
+# secondary address lines dropped) accepts a match the primary strings alone
+# would reject, so it must clear a near-exact bar: in the miss audit >99% of
+# genuine variant recoveries scored 100, while the rare false variant matches
+# scored in the 80s.
+MIN_VARIANT_SCORE = 90
+
+# Tokens that mark a sub-unit of a building. A variant whose added/dropped
+# tokens include one of these could score a single flat's certificate as if it
+# were the whole building, so such variants are inadmissible.
+_FLAT_TOKENS = {
+    "FLAT",
+    "FLATS",
+    "APARTMENT",
+    "APT",
+    "UNIT",
+    "MAISONETTE",
+    "STUDIO",
+    "ROOM",
+}
 
 
 def normalize_address_key(s: pl.Expr) -> pl.Expr:
-    normalized = (
-        s.cast(pl.String)
-        .str.to_uppercase()
-        .str.replace_all(r"[^0-9A-Z]+", " ")
-        .str.replace_all(r"\s+", " ")
-        .str.strip_chars()
-    )
+    normalized = uppercase_alnum_key_expr(s)
     return pl.when(normalized.str.contains(r"[A-Z]")).then(normalized).otherwise(None)
 
 
@@ -58,6 +79,8 @@ def fuzzy_join_on_postcode(
     right_postcode_col: str,
     min_score: int = MIN_FUZZY_SCORE,
     min_score_without_numbers: int = MIN_FUZZY_SCORE_WITHOUT_NUMBERS,
+    left_variant_cols: Sequence[str] = (),
+    right_variant_cols: Sequence[str] = (),
 ) -> pl.LazyFrame:
     """Fuzzy join two LazyFrames by matching addresses within postcode buckets.
 
@@ -66,6 +89,19 @@ def fuzzy_join_on_postcode(
     columns (index, address, postcode) via projection pushdown, and the
     final join reads the remaining columns lazily.
 
+    ``left_variant_cols`` / ``right_variant_cols`` name alternative address
+    columns for the same property (e.g. the EPC's first address line without
+    its locality suffix, or the price-paid address with its locality
+    appended). A pair is scored as the best token_sort_ratio over all
+    admissible variant combinations: source registers frequently disagree
+    only on a trailing village/locality token, which alone drags short
+    addresses below the match threshold. The number-compatibility gate is
+    always evaluated on the primary addresses, and `_admissible_variants`
+    rejects any variant whose added/dropped tokens carry digits or flat
+    designators, so a variant can never bypass the gate or score a single
+    flat as its whole building. Variant-only scores must clear
+    ``MIN_VARIANT_SCORE``.
+
     Returns a LazyFrame with all left and right columns, plus a
     ``_match_score`` (UInt8) audit column holding the token_sort_ratio of
     the accepted match (exact matches score 100).  Unmatched rows have null
@@ -90,6 +126,10 @@ def fuzzy_join_on_postcode(
                 normalize_postcode_key(pl.col(left_postcode_col)).alias(
                     "_left_postcode"
                 ),
+                *(
+                    normalize_address_key(pl.col(col)).alias(f"_left_variant_{i}")
+                    for i, col in enumerate(left_variant_cols)
+                ),
             )
             .collect(engine="streaming")
         )
@@ -104,30 +144,45 @@ def fuzzy_join_on_postcode(
                 normalize_postcode_key(pl.col(right_postcode_col)).alias(
                     "_right_postcode"
                 ),
+                *(
+                    normalize_address_key(pl.col(col)).alias(f"_right_variant_{i}")
+                    for i, col in enumerate(right_variant_cols)
+                ),
             )
             .unique(subset=["_right_address", "_right_postcode"], keep="first")
             .collect(engine="streaming")
         )
 
+        left_variant_names = [f"_left_variant_{i}" for i in range(len(left_variant_cols))]
+        right_variant_names = [
+            f"_right_variant_{i}" for i in range(len(right_variant_cols))
+        ]
+
         # Group right side by postcode for fast lookup
-        right_by_postcode: dict[str, list[tuple[int, str]]] = {}
-        for idx, postcode, address in zip(
+        right_by_postcode: dict[str, list[tuple[int, str, tuple[str, ...]]]] = {}
+        for idx, postcode, address, *variants in zip(
             right_match["_right_idx"],
             right_match["_right_postcode"],
             right_match["_right_address"],
+            *(right_match[name] for name in right_variant_names),
         ):
             if address is not None and postcode is not None:
-                right_by_postcode.setdefault(postcode, []).append((idx, address))
+                right_by_postcode.setdefault(postcode, []).append(
+                    (idx, address, _admissible_variants(address, variants))
+                )
 
         # Group left side by postcode
-        left_by_postcode: dict[str, list[tuple[int, str]]] = {}
-        for idx, postcode, address in zip(
+        left_by_postcode: dict[str, list[tuple[int, str, tuple[str, ...]]]] = {}
+        for idx, postcode, address, *variants in zip(
             left_match["_left_idx"],
             left_match["_left_postcode"],
             left_match["_left_address"],
+            *(left_match[name] for name in left_variant_names),
         ):
             if address is not None and postcode is not None:
-                left_by_postcode.setdefault(postcode, []).append((idx, address))
+                left_by_postcode.setdefault(postcode, []).append(
+                    (idx, address, _admissible_variants(address, variants))
+                )
 
         del left_match, right_match
 
@@ -145,7 +200,12 @@ def fuzzy_join_on_postcode(
 
         # Score all pairwise matches in parallel, then greedily assign from
         # highest score downward so best pairs lock in first.
-        all_pairs: list[tuple[int, int, int]] = []  # (score, left_idx, right_idx)
+        # Pair tuples are (score, exact, left_idx, right_idx); `exact` marks a
+        # literally-equal primary pair so it wins greedy ties against a pair
+        # that merely token-sorts to the same score (e.g. "APARTMENT 3 1 HIGH
+        # ST" vs "APARTMENT 1 3 HIGH ST" both score 100 against each other's
+        # certificates, but each has a literal twin).
+        all_pairs: list[tuple[int, int, int, int]] = []
         with ProcessPoolExecutor(max_workers=cpu_count()) as executor:
             for pairs in tqdm(
                 executor.map(_score_bucket, tasks, chunksize=64),
@@ -156,8 +216,9 @@ def fuzzy_join_on_postcode(
 
         del tasks, left_by_postcode, right_by_postcode
 
-        # Sort descending by score so best matches are assigned first
-        all_pairs.sort(key=lambda t: (t[0], -t[1]), reverse=True)
+        # Sort so the best matches are assigned first: score, then literal
+        # equality, then stable left-index order.
+        all_pairs.sort(key=lambda t: (t[0], t[1], -t[2]), reverse=True)
 
         # Keep the score alongside each accepted pair: it is emitted as the
         # _match_score audit column so downstream consumers can distinguish
@@ -166,7 +227,7 @@ def fuzzy_join_on_postcode(
         matched_left: set[int] = set()
         matched_right: set[int] = set()
 
-        for score, left_idx, right_idx in all_pairs:
+        for score, _exact, left_idx, right_idx in all_pairs:
             if left_idx in matched_left or right_idx in matched_right:
                 continue
             matches.append((left_idx, right_idx, score))
@@ -208,40 +269,102 @@ def fuzzy_join_on_postcode(
     return result.lazy()
 
 
+def _number_tokens(address: str) -> set[str]:
+    tokens = set(_NUMBER_RE.findall(address))
+    tokens.update(_FLAT_LETTER_RE.findall(address))
+    return tokens
+
+
 def _numbers_compatible(a: str, b: str) -> bool:
     """Check that the number tokens (house/flat numbers, including any letter
-    suffix) of two addresses are IDENTICAL sets.
+    suffix, plus single-letter flat designators) of two addresses are
+    IDENTICAL sets.
 
     Equality, not subset: subset logic let "188 GREAT NORTH WAY" absorb
     "FLAT 1 188 GREAT NORTH WAY" ({188} is a subset of {1, 188}), attaching a
     single flat's EPC facts to the whole building — tens of thousands of
     wrong-property matches. Likewise digit-only tokens made "8A" and "8B"
-    both look like {8} and match each other (and plain "8"). Precision over
-    recall: a pair whose two sources genuinely disagree on number tokens is
-    safer left unmatched.
+    both look like {8} and match each other (and plain "8"), and ungated
+    letter flats let "FLAT D 39 X ST" cross-match "FLAT F 39 X ST" at ~96.
+    Precision over recall: a pair whose two sources genuinely disagree on
+    number tokens is safer left unmatched.
 
     One side numbered, the other not -> incompatible. Neither numbered ->
     compatible; such pairs are scored against the stricter no-numbers
     threshold instead.
     """
-    nums_a = set(_NUMBER_RE.findall(a))
-    nums_b = set(_NUMBER_RE.findall(b))
+    nums_a = _number_tokens(a)
+    nums_b = _number_tokens(b)
     if not nums_a and not nums_b:
         return True
     return nums_a == nums_b
 
 
+def _admissible_variants(
+    primary: str, variants: Sequence[str | None]
+) -> tuple[str, ...]:
+    """Variants of ``primary`` that are safe to score against the other side.
+
+    A variant may only ADD or DROP whole tokens relative to the primary (one
+    word multiset must contain the other) — never substitute, so a register
+    row whose address lines disagree with the combined address can't smuggle
+    in a different street. The number gate runs on the primary addresses
+    only, so the added/dropped tokens must additionally carry no digits
+    (house numbers) and no flat designator (a "Flat 1"-style secondary line
+    dropped from an EPC address would otherwise let a single flat score as
+    the whole building). The remaining admissible difference is exactly the
+    harmless kind variants exist for: trailing locality/village/town words.
+    """
+    primary_words = Counter(primary.split())
+    admissible: list[str] = []
+    for variant in variants:
+        if not variant or variant == primary:
+            continue
+        variant_words = Counter(variant.split())
+        if not (variant_words <= primary_words or primary_words <= variant_words):
+            continue
+        changed = (primary_words - variant_words) + (variant_words - primary_words)
+        if any(
+            any(ch.isdigit() for ch in token) or token in _FLAT_TOKENS
+            for token in changed
+        ):
+            continue
+        admissible.append(variant)
+    return tuple(dict.fromkeys(admissible))
+
+
 def _score_bucket(
-    args: tuple[list[tuple[int, str]], list[tuple[int, str]], int, int],
-) -> list[tuple[int, int, int]]:
+    args: tuple[
+        list[tuple[int, str, tuple[str, ...]]],
+        list[tuple[int, str, tuple[str, ...]]],
+        int,
+        int,
+    ],
+) -> list[tuple[int, int, int, int]]:
     """Score all address pairs within a single postcode bucket."""
     left_entries, right_entries, min_score, min_score_without_numbers = args
     pairs = []
-    for left_row, left_address in left_entries:
-        for right_row, right_address in right_entries:
+    for left_row, left_address, left_variants in left_entries:
+        for right_row, right_address, right_variants in right_entries:
             if not _numbers_compatible(left_address, right_address):
                 continue
             score = fuzz.token_sort_ratio(left_address, right_address)
+            # Variant pairs recover same-property matches where one register
+            # carries a locality suffix the other lacks; a variant-only score
+            # must clear the near-exact MIN_VARIANT_SCORE bar.
+            if score < 100 and (left_variants or right_variants):
+                for left_variant in (left_address, *left_variants):
+                    for right_variant in (right_address, *right_variants):
+                        if (
+                            left_variant is left_address
+                            and right_variant is right_address
+                        ):
+                            continue
+                        variant_score = fuzz.token_sort_ratio(
+                            left_variant, right_variant
+                        )
+                        if variant_score >= MIN_VARIANT_SCORE and variant_score > score:
+                            score = variant_score
             # Number-less pairs (named houses, building-name flats) lack the
             # house-number disambiguator, so require a near-exact match.
             threshold = (
@@ -250,5 +373,7 @@ def _score_bucket(
                 else min_score_without_numbers
             )
             if score >= threshold:
-                pairs.append((score, left_row, right_row))
+                pairs.append(
+                    (score, int(left_address == right_address), left_row, right_row)
+                )
     return pairs
diff --git a/pipeline/utils/normalize.py b/pipeline/utils/normalize.py
new file mode 100644
index 0000000..4032e3d
--- /dev/null
+++ b/pipeline/utils/normalize.py
@@ -0,0 +1,70 @@
+"""Shared low-level text-normalization primitives.
+
+Address matching (``pipeline.utils.fuzzy_join``, ``pipeline.transform.merge``),
+POI retailer cleanup (``pipeline.transform.transform_poi``) and school-name
+matching (``pipeline.check_school_cutoffs``) each layer domain-specific rules
+on top of these. The primitives are deliberately tiny and single-purpose so
+that composing them preserves every caller's existing output byte-for-byte.
+"""
+
+import re
+
+import polars as pl
+
+# One character outside [a-z0-9 ]. Callers lowercase first; each offending
+# character becomes a single space (runs are NOT merged here — callers apply
+# word-level rules and then collapse_whitespace).
+_NON_ALNUM_LOWER_RE = re.compile(r"[^a-z0-9 ]")
+
+# Any digit marks a token as number-bearing (house/flat numbers, including
+# letter-suffixed forms such as 8A, which still contain a digit).
+_DIGIT_RE = re.compile(r"\d")
+
+
+def collapse_whitespace(s: str) -> str:
+    """Collapse every whitespace run to a single space and strip the ends."""
+    return " ".join(s.split())
+
+
+def strip_or_empty(s: str | None) -> str:
+    """Strip leading/trailing whitespace, mapping None to ``""``.
+
+    Interior whitespace is preserved (unlike :func:`collapse_whitespace`) so
+    the result can be looked up verbatim against curated dictionary keys.
+    """
+    return "" if s is None else s.strip()
+
+
+def replace_non_alnum_lower(s: str) -> str:
+    """Replace each character outside [a-z0-9 ] with a single space.
+
+    Expects already-lowercased input (uppercase letters are replaced too).
+    Replacement is per character, not per run; callers collapse whitespace
+    afterwards.
+    """
+    return _NON_ALNUM_LOWER_RE.sub(" ", s)
+
+
+def drop_digit_tokens(s: str) -> str:
+    """Drop whitespace-separated tokens that contain any digit.
+
+    ``"10A HIGH STREET" -> "HIGH STREET"``. The surviving tokens are rejoined
+    with single spaces, so whitespace collapses as a side effect.
+    """
+    return " ".join(token for token in s.split() if not _DIGIT_RE.search(token))
+
+
+def uppercase_alnum_key_expr(s: pl.Expr) -> pl.Expr:
+    """Polars expression: uppercase, replace each non-alphanumeric run with a
+    single space, collapse whitespace, and strip the ends.
+
+    Non-ASCII letters fall outside [0-9A-Z] after uppercasing and become
+    spaces (``"Café 1" -> "CAF 1"``).
+    """
+    return (
+        s.cast(pl.String)
+        .str.to_uppercase()
+        .str.replace_all(r"[^0-9A-Z]+", " ")
+        .str.replace_all(r"\s+", " ")
+        .str.strip_chars()
+    )
diff --git a/pipeline/utils/test_fuzzy_join.py b/pipeline/utils/test_fuzzy_join.py
index 6d8f180..8063e9b 100644
--- a/pipeline/utils/test_fuzzy_join.py
+++ b/pipeline/utils/test_fuzzy_join.py
@@ -1,7 +1,7 @@
 import polars as pl
 
 from pipeline.utils import fuzzy_join_on_postcode, normalize_postcode_key
-from pipeline.utils.fuzzy_join import _numbers_compatible
+from pipeline.utils.fuzzy_join import _admissible_variants, _numbers_compatible
 
 
 def test_fuzzy_join_on_postcode_matches_addresses_within_postcode():
@@ -165,7 +165,7 @@ def test_fuzzy_join_rejects_mid_score_number_less_match():
 
 
 def test_fuzzy_join_matches_numbered_pair_at_baseline_threshold():
-    # "10 ACACIA AVENUE" vs "FLAT A 10 ACACIA AVENUE" scores exactly 82 and the
+    # "10 ACACIA AVENUE" vs "10 ACACIA AVENUE OAKHAM" scores exactly 82 and the
     # house number is compatible, so the numbered baseline (>= 82) still matches.
     left = pl.LazyFrame(
         {
@@ -175,7 +175,7 @@ def test_fuzzy_join_matches_numbered_pair_at_baseline_threshold():
     )
     right = pl.LazyFrame(
         {
-            "right_address": ["Flat A, 10 Acacia Avenue"],
+            "right_address": ["10 Acacia Avenue, Oakham"],
             "right_postcode": ["AB1 2CD"],
         }
     )
@@ -189,7 +189,7 @@ def test_fuzzy_join_matches_numbered_pair_at_baseline_threshold():
         right_postcode_col="right_postcode",
     ).collect()
 
-    assert result["right_address"].to_list() == ["Flat A, 10 Acacia Avenue"]
+    assert result["right_address"].to_list() == ["10 Acacia Avenue, Oakham"]
 
 
 def test_fuzzy_join_matches_high_score_number_less_pair():
@@ -244,6 +244,151 @@ def test_numbers_compatible_number_less_and_one_sided_pairs():
     assert not _numbers_compatible("ROSE COTTAGE", "8 HIGH STREET")
 
 
+def test_numbers_compatible_gates_single_letter_flats():
+    # "FLAT D" and "FLAT F" are different flats even with identical street
+    # numbers; ungated they token_sort to ~96 and cross-matched. The letter is
+    # a pseudo-number token, so it also blocks a flat matching the bare
+    # building address.
+    assert not _numbers_compatible(
+        "FLAT D 39 GERTRUDE STREET", "FLAT F 39 GERTRUDE STREET"
+    )
+    assert _numbers_compatible(
+        "FLAT D 39 GERTRUDE STREET", "39 GERTRUDE STREET FLAT D"
+    )
+    assert not _numbers_compatible("FLAT B ROSE COURT", "ROSE COURT")
+    # A letter glued to a number ("A3") is a unit name, not a flat letter.
+    assert _numbers_compatible("FLAT A3 CHESHAM HEIGHTS", "FLAT A3 CHESHAM HEIGHTS")
+
+
+def test_admissible_variants_allows_locality_suffix_only():
+    # Locality words may differ between a variant and its primary; digits and
+    # flat designators may not (the gate ran on the primary only).
+    assert _admissible_variants(
+        "12 OAK ROAD", ["12 OAK ROAD HALE", "12 OAK ROAD"]
+    ) == ("12 OAK ROAD HALE",)
+    # Dropping "FLAT 1" (digit) or "FLAT B" (flat designator) is inadmissible:
+    # the variant would score a single flat as the whole building.
+    assert (
+        _admissible_variants("FLAT 1 188 GREAT NORTH WAY", ["188 GREAT NORTH WAY"])
+        == ()
+    )
+    assert _admissible_variants("FLAT B ROSE COURT", ["ROSE COURT"]) == ()
+    assert _admissible_variants("12 OAK ROAD", [None, "12 OAK ROAD"]) == ()
+    # Substitution is never admissible: a register row whose address1
+    # disagrees with the combined address must not smuggle in a different
+    # street for scoring.
+    assert _admissible_variants("12 OAK ROAD", ["12 ELM ROAD"]) == ()
+    assert (
+        _admissible_variants("1 TOTALLY DIFFERENT ROAD", ["1 EXAMPLE STREET"]) == ()
+    )
+
+
+def test_fuzzy_join_variant_recovers_locality_suffix_mismatch():
+    # The EPC register stores "12 Oak Road, Hale" (address1 + locality line)
+    # while price-paid has the bare "12 Oak Road": token_sort scores 81 < 82
+    # and the match was lost. The EPC's address1-only variant scores 100.
+    left = pl.LazyFrame(
+        {
+            "left_address": ["12 Oak Road"],
+            "left_postcode": ["AB1 2CD"],
+            "left_with_locality": ["12 Oak Road Hale"],
+        }
+    )
+    right = pl.LazyFrame(
+        {
+            "right_address": ["12 Oak Road, Hale"],
+            "right_postcode": ["AB1 2CD"],
+            "right_address1": ["12 Oak Road"],
+        }
+    )
+
+    unmatched = fuzzy_join_on_postcode(
+        left=left,
+        right=right,
+        left_address_col="left_address",
+        right_address_col="right_address",
+        left_postcode_col="left_postcode",
+        right_postcode_col="right_postcode",
+    ).collect()
+    assert unmatched["_match_score"].to_list() == [None]
+
+    result = fuzzy_join_on_postcode(
+        left=left,
+        right=right,
+        left_address_col="left_address",
+        right_address_col="right_address",
+        left_postcode_col="left_postcode",
+        right_postcode_col="right_postcode",
+        left_variant_cols=["left_with_locality"],
+        right_variant_cols=["right_address1"],
+    ).collect()
+    assert result["_match_score"].to_list() == [100]
+
+
+def test_fuzzy_join_variant_cannot_unlock_a_flat_for_its_building():
+    # The EPC's secondary line carries the flat designator; dropping it would
+    # score the flat's certificate 100 against the whole-building price-paid
+    # address. The variant must be ruled inadmissible and the pair unmatched.
+    left = pl.LazyFrame(
+        {
+            "left_address": ["188 Great North Way"],
+            "left_postcode": ["AB1 2CD"],
+        }
+    )
+    right = pl.LazyFrame(
+        {
+            "right_address": ["Flat 1, 188 Great North Way"],
+            "right_postcode": ["AB1 2CD"],
+            "right_address1": ["188 Great North Way"],
+        }
+    )
+
+    result = fuzzy_join_on_postcode(
+        left=left,
+        right=right,
+        left_address_col="left_address",
+        right_address_col="right_address",
+        left_postcode_col="left_postcode",
+        right_postcode_col="right_postcode",
+        right_variant_cols=["right_address1"],
+    ).collect()
+
+    assert result["_match_score"].to_list() == [None]
+
+
+def test_fuzzy_join_variant_score_must_be_near_exact():
+    # A score reached only through a variant must clear MIN_VARIANT_SCORE
+    # (90): "2 MYRTLE COTTAGES" vs "2 LEITH VIEW COTTAGES" type pairs scored
+    # in the 80s via variants and were false matches.
+    left = pl.LazyFrame(
+        {
+            "left_address": ["2 Myrtle Cottages"],
+            "left_postcode": ["AB1 2CD"],
+            "left_with_locality": ["2 Myrtle Cottages Dorking"],
+        }
+    )
+    right = pl.LazyFrame(
+        {
+            "right_address": ["2 Leith View Cottages, North Holmwood"],
+            "right_postcode": ["AB1 2CD"],
+            "right_address1": ["2 Leith View Cottages"],
+        }
+    )
+
+    result = fuzzy_join_on_postcode(
+        left=left,
+        right=right,
+        left_address_col="left_address",
+        right_address_col="right_address",
+        left_postcode_col="left_postcode",
+        right_postcode_col="right_postcode",
+        left_variant_cols=["left_with_locality"],
+        right_variant_cols=["right_address1"],
+    ).collect()
+
+    assert result["_match_score"].to_list() == [None]
+
+
 def test_fuzzy_join_rejects_wrong_letter_suffix_match():
     # End-to-end guard for the 8A/8B class of wrong-property matches: the only
     # candidate in the postcode bucket differs solely in the number suffix, so
@@ -294,7 +439,7 @@ def test_fuzzy_join_emits_match_score_column():
                 "10 HIGH STREET",
                 # Scores exactly 82 against "10 Acacia Avenue" (see
                 # test_fuzzy_join_matches_numbered_pair_at_baseline_threshold).
-                "Flat A, 10 Acacia Avenue",
+                "10 Acacia Avenue, Oakham",
             ],
             "right_postcode": ["AB1 2CD", "EF3 4GH"],
         }
diff --git a/pipeline/utils/test_normalize.py b/pipeline/utils/test_normalize.py
new file mode 100644
index 0000000..26a50cd
--- /dev/null
+++ b/pipeline/utils/test_normalize.py
@@ -0,0 +1,158 @@
+import polars as pl
+
+from pipeline.check_school_cutoffs import normalize_la, normalize_name
+from pipeline.transform.merge import _street_only_address
+from pipeline.transform.transform_poi import normalize_grocery_retailer
+from pipeline.utils.fuzzy_join import normalize_address_key
+from pipeline.utils.normalize import (
+    collapse_whitespace,
+    drop_digit_tokens,
+    replace_non_alnum_lower,
+    strip_or_empty,
+    uppercase_alnum_key_expr,
+)
+
+# --- Primitives -------------------------------------------------------------
+
+
+def test_collapse_whitespace():
+    assert collapse_whitespace("") == ""
+    assert collapse_whitespace("   ") == ""
+    assert collapse_whitespace("a  b") == "a b"
+    assert collapse_whitespace("  a \t b \n c  ") == "a b c"
+    # str.split() also splits on unicode whitespace (non-breaking space).
+    assert collapse_whitespace("a\u00a0b") == "a b"
+
+
+def test_strip_or_empty():
+    assert strip_or_empty(None) == ""
+    assert strip_or_empty("") == ""
+    assert strip_or_empty("  x  ") == "x"
+    # Interior whitespace is preserved, unlike collapse_whitespace.
+    assert strip_or_empty("  a   b  ") == "a   b"
+
+
+def test_replace_non_alnum_lower():
+    assert replace_non_alnum_lower("") == ""
+    assert replace_non_alnum_lower("abc 123") == "abc 123"
+    # Per-character replacement: runs are not merged.
+    assert replace_non_alnum_lower("a--b") == "a  b"
+    # Existing spaces are kept as-is.
+    assert replace_non_alnum_lower("a , b") == "a   b"
+    # Uppercase and accented letters fall outside [a-z0-9 ].
+    assert replace_non_alnum_lower("École") == " cole"
+
+
+def test_drop_digit_tokens():
+    assert drop_digit_tokens("") == ""
+    assert drop_digit_tokens("10A HIGH STREET") == "HIGH STREET"
+    assert drop_digit_tokens("8B") == ""
+    assert drop_digit_tokens("12 34") == ""
+    assert drop_digit_tokens("KINGSWOOD") == "KINGSWOOD"
+    # Whitespace collapses as a side effect of the token rejoin.
+    assert drop_digit_tokens("  A   B  ") == "A B"
+
+
+def test_uppercase_alnum_key_expr():
+    values = [
+        "Flat 2, 10 High Street",
+        "  12   High-Street ",
+        "",
+        None,
+        "Café 1",
+        "st mary's-court",
+    ]
+    out = (
+        pl.DataFrame({"a": values}, schema={"a": pl.String})
+        .select(uppercase_alnum_key_expr(pl.col("a")))
+        .to_series()
+        .to_list()
+    )
+    assert out == [
+        "FLAT 2 10 HIGH STREET",
+        "12 HIGH STREET",
+        "",
+        None,
+        "CAF 1",
+        "ST MARY S COURT",
+    ]
+
+
+# --- Characterization of the call sites built on the primitives ------------
+# Expected values were captured from the pre-refactor implementations and
+# must never change: each wrapper's output is byte-for-byte pinned.
+
+
+def test_normalize_address_key_characterization():
+    values = [
+        "Flat 2, 10 High Street",
+        "  12   High-Street ",
+        "123",  # digits only: no letter -> null
+        "",  # empty -> null
+        None,  # null in, null out
+        "Café 1",
+        "st mary's-court",
+        "ALREADY NORMAL",
+    ]
+    out = (
+        pl.DataFrame({"a": values}, schema={"a": pl.String})
+        .select(normalize_address_key(pl.col("a")))
+        .to_series()
+        .to_list()
+    )
+    assert out == [
+        "FLAT 2 10 HIGH STREET",
+        "12 HIGH STREET",
+        None,
+        None,
+        None,
+        "CAF 1",
+        "ST MARY S COURT",
+        "ALREADY NORMAL",
+    ]
+
+
+def test_street_only_address_characterization():
+    assert _street_only_address("10A HIGH STREET") == "HIGH STREET"
+    assert _street_only_address("FLAT 1 188 GREAT NORTH WAY") == "FLAT GREAT NORTH WAY"
+    assert _street_only_address("") == ""
+    assert _street_only_address("OLDSTEAD ROAD") == "OLDSTEAD ROAD"
+    assert _street_only_address("  A   B  ") == "A B"
+    assert _street_only_address("12 34") == ""
+    assert _street_only_address("8B") == ""
+
+
+def test_normalize_grocery_retailer_characterization():
+    assert normalize_grocery_retailer(None) == ""
+    assert normalize_grocery_retailer("") == ""
+    assert normalize_grocery_retailer("  Tesco Express  ") == "Tesco Express"
+    assert normalize_grocery_retailer("Sainsburys") == "Sainsbury's"
+    assert normalize_grocery_retailer("Lincolnshire Co-operative") == "Co-op"
+    # Only edge whitespace is stripped; interior whitespace must survive so
+    # near-miss names fall through the exact dictionary lookups unchanged.
+    assert normalize_grocery_retailer("Bob's  Shop") == "Bob's  Shop"
+    assert normalize_grocery_retailer(" Marks and Spencer ") == "M&S"
+
+
+def test_normalize_name_characterization():
+    assert normalize_name("St. Mary's C of E Primary School") == (
+        "st marys primary school"
+    )
+    assert normalize_name("St. Mary's C of E Primary School", True) == "st marys"
+    assert normalize_name("") == ""
+    assert normalize_name("Ham & High School") == "ham high school"
+    assert normalize_name("Ham & High School", True) == "ham"
+    # Accented characters become spaces, splitting the word.
+    assert normalize_name("École Élémentaire") == "cole l mentaire"
+    assert normalize_name("  THE   KING'S  ACADEMY  ") == "kings academy"
+    assert normalize_name("Holy Trinity RC Voluntary Aided School") == (
+        "holy trinity school"
+    )
+    assert normalize_name("st. john's") == "st johns"
+
+
+def test_normalize_la_characterization():
+    assert normalize_la("City of Westminster") == "westminster"
+    assert normalize_la("Brighton & Hove") == "brighton and hove"
+    assert normalize_la("  Kingston upon Thames  ") == "kingston upon thames"
+    assert normalize_la("") == ""
diff --git a/server-rs/src/checkout_sessions.rs b/server-rs/src/checkout_sessions.rs
deleted file mode 100644
index 27a03dd..0000000
--- a/server-rs/src/checkout_sessions.rs
+++ /dev/null
@@ -1,1659 +0,0 @@
-use std::sync::LazyLock;
-use std::time::{SystemTime, UNIX_EPOCH};
-
-use anyhow::{anyhow, Context};
-use serde_json::Value;
-use tokio::sync::Mutex;
-use tracing::warn;
-
-use crate::auth::PocketBaseUser;
-use crate::pocketbase::get_superuser_token;
-use crate::pocketbase_locks::acquire_pocketbase_lock;
-use crate::routes::pricing::{count_licensed_users, price_for_count};
-use crate::state::AppState;
-
-pub const CHECKOUT_CURRENCY: &str = "gbp";
-
-const CHECKOUT_SESSION_TTL_SECS: u64 = 31 * 60;
-const CHECKOUT_PRODUCT_NAME: &str = "Perfect Postcodes Lifetime License";
-const CHECKOUT_COLLECTION: &str = "checkout_sessions";
-const CHECKOUT_PRICING_LOCK_NAME: &str = "checkout:pricing";
-const CHECKOUT_PRICING_LOCK_TTL_SECS: u64 = 5 * 60;
-const REFERRAL_DISCOUNT_PERCENT: u64 = 30;
-
-static CHECKOUT_RESERVATION_LOCK: LazyLock<Mutex<()>> = LazyLock::new(|| Mutex::new(()));
-
-pub enum CheckoutStart {
-    Free,
-    Stripe { url: String },
-}
-
-pub enum CheckoutCompletion {
-    Grant(VerifiedCheckout),
-    AlreadyHandled(VerifiedCheckout),
-    Rejected(String),
-}
-
-pub enum PaymentReversalOutcome {
-    Applied {
-        user_id: String,
-    },
-    AlreadyHandled {
-        user_id: String,
-    },
-    IgnoredPartialRefund {
-        user_id: String,
-        refunded_amount_pence: u64,
-        paid_amount_pence: u64,
-    },
-    NoMatchingCheckout,
-    NotReversible {
-        user_id: String,
-        status: String,
-    },
-}
-
-pub enum PaymentReinstatementOutcome {
-    Applied { user_id: String },
-    AlreadyHandled { user_id: String },
-    Ignored { user_id: String, reason: String },
-    NoMatchingCheckout,
-}
-
-pub struct VerifiedCheckout {
-    pub reservation_id: String,
-    pub user_id: String,
-    pub stripe_session_id: String,
-    pub payment_intent_id: String,
-    pub paid_amount_pence: u64,
-    pub referral_invite_id: String,
-}
-
-#[derive(Debug)]
-struct PendingCheckout {
-    id: String,
-    user_id: String,
-    stripe_session_id: String,
-    checkout_url: String,
-    amount_pence: u64,
-    expected_total_pence: u64,
-    currency: String,
-    referral_invite_id: String,
-    status: String,
-    payment_intent_id: String,
-    paid_amount_pence: u64,
-    reversal_reason: String,
-}
-
-pub fn now_unix_secs() -> u64 {
-    SystemTime::now()
-        .duration_since(UNIX_EPOCH)
-        .unwrap_or_default()
-        .as_secs()
-}
-
-pub async fn start_license_checkout(
-    state: &AppState,
-    user: &PocketBaseUser,
-    success_url: &str,
-    cancel_url: &str,
-    discount_coupon_id: Option<&str>,
-    referral_invite_id: Option<&str>,
-) -> anyhow::Result<CheckoutStart> {
-    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
-    let pricing_lock = acquire_pocketbase_lock(
-        state,
-        CHECKOUT_PRICING_LOCK_NAME,
-        CHECKOUT_PRICING_LOCK_TTL_SECS,
-    )
-    .await?;
-    let result = start_license_checkout_locked(
-        state,
-        user,
-        success_url,
-        cancel_url,
-        discount_coupon_id,
-        referral_invite_id,
-    )
-    .await;
-    if let Err(err) = pricing_lock.release().await {
-        warn!("Failed to release checkout pricing lock: {err}");
-    }
-    result
-}
-
-async fn start_license_checkout_locked(
-    state: &AppState,
-    user: &PocketBaseUser,
-    success_url: &str,
-    cancel_url: &str,
-    discount_coupon_id: Option<&str>,
-    referral_invite_id: Option<&str>,
-) -> anyhow::Result<CheckoutStart> {
-    let now = now_unix_secs();
-    expire_stale_pending_checkouts(state, now).await?;
-
-    if let Some(existing) = find_active_checkout_for_user(
-        state,
-        &user.id,
-        discount_coupon_id.unwrap_or_default(),
-        referral_invite_id.unwrap_or_default(),
-        now,
-    )
-    .await?
-    {
-        if !existing.checkout_url.is_empty() {
-            return Ok(CheckoutStart::Stripe {
-                url: existing.checkout_url,
-            });
-        }
-        if let Err(err) = mark_checkout_status(state, &existing.id, "failed").await {
-            warn!(
-                reservation_id = %existing.id,
-                "Failed to fail incomplete checkout reservation: {err}"
-            );
-        }
-    }
-
-    let licensed_count = count_licensed_users(state).await?;
-    let pending_count = count_active_pending_checkouts(state, now).await?;
-    let price_pence = price_for_count(licensed_count + pending_count);
-
-    if price_pence == 0 {
-        grant_license(state, &user.id).await?;
-        return Ok(CheckoutStart::Free);
-    }
-
-    let expires_at_unix = now + CHECKOUT_SESSION_TTL_SECS;
-    let expected_total_pence = expected_total_for_checkout(price_pence, discount_coupon_id);
-    let reservation_id = create_pending_checkout(
-        state,
-        PendingCheckoutInput {
-            user_id: &user.id,
-            amount_pence: price_pence,
-            expected_total_pence,
-            currency: CHECKOUT_CURRENCY,
-            discount_coupon_id: discount_coupon_id.unwrap_or_default(),
-            referral_invite_id: referral_invite_id.unwrap_or_default(),
-            expires_at_unix,
-        },
-    )
-    .await?;
-
-    if let Some(invite_id) = referral_invite_id.filter(|id| !id.is_empty()) {
-        if let Err(err) =
-            reserve_referral_invite(state, invite_id, &user.id, &reservation_id, expires_at_unix)
-                .await
-        {
-            if let Err(mark_err) = mark_checkout_status(state, &reservation_id, "failed").await {
-                warn!(
-                    reservation_id,
-                    "Failed to mark checkout reservation failed: {mark_err}"
-                );
-            }
-            return Err(err);
-        }
-    }
-
-    let stripe_result = create_stripe_session(
-        state,
-        user,
-        &reservation_id,
-        price_pence,
-        success_url,
-        cancel_url,
-        expires_at_unix,
-        discount_coupon_id,
-    )
-    .await;
-
-    let (stripe_session_id, url) = match stripe_result {
-        Ok(session) => session,
-        Err(err) => {
-            if let Err(mark_err) = mark_checkout_status(state, &reservation_id, "failed").await {
-                warn!(
-                    reservation_id,
-                    "Failed to mark checkout reservation failed: {mark_err}"
-                );
-            }
-            if let Some(invite_id) = referral_invite_id.filter(|id| !id.is_empty()) {
-                if let Err(release_err) =
-                    release_referral_invite_reservation(state, invite_id, &reservation_id).await
-                {
-                    warn!(
-                        reservation_id,
-                        referral_invite_id = invite_id,
-                        "Failed to release referral invite reservation: {release_err}"
-                    );
-                }
-            }
-            return Err(err);
-        }
-    };
-
-    if let Err(err) = attach_stripe_session(state, &reservation_id, &stripe_session_id, &url).await
-    {
-        if let Err(mark_err) = mark_checkout_status(state, &reservation_id, "failed").await {
-            warn!(
-                reservation_id,
-                "Failed to mark checkout reservation failed: {mark_err}"
-            );
-        }
-        if let Some(invite_id) = referral_invite_id.filter(|id| !id.is_empty()) {
-            if let Err(release_err) =
-                release_referral_invite_reservation(state, invite_id, &reservation_id).await
-            {
-                warn!(
-                    reservation_id,
-                    referral_invite_id = invite_id,
-                    "Failed to release referral invite reservation: {release_err}"
-                );
-            }
-        }
-        return Err(err);
-    }
-
-    Ok(CheckoutStart::Stripe { url })
-}
-
-pub async fn verify_checkout_completion(
-    state: &AppState,
-    session: &Value,
-) -> anyhow::Result<CheckoutCompletion> {
-    let session_id = match session["id"].as_str() {
-        Some(id) if is_safe_stripe_session_id(id) => id,
-        _ => {
-            return Ok(CheckoutCompletion::Rejected(
-                "missing or invalid session id".into(),
-            ))
-        }
-    };
-    let payment_intent_id = match session["payment_intent"].as_str() {
-        Some(id) if is_safe_stripe_session_id(id) => id,
-        _ => {
-            return Ok(CheckoutCompletion::Rejected(
-                "missing or invalid payment intent id".into(),
-            ))
-        }
-    };
-
-    let checkout = match find_checkout_by_stripe_session(state, session_id).await? {
-        Some(checkout) => checkout,
-        None => {
-            return Ok(CheckoutCompletion::Rejected(
-                "checkout session has no reservation".into(),
-            ))
-        }
-    };
-
-    let already_completed = checkout.status == "completed";
-    if !already_completed && checkout.status != "pending" && checkout.status != "expired" {
-        return Ok(CheckoutCompletion::Rejected(format!(
-            "checkout reservation is {}",
-            checkout.status
-        )));
-    }
-    if checkout.stripe_session_id != session_id {
-        mark_checkout_status(state, &checkout.id, "invalid").await?;
-        return Ok(CheckoutCompletion::Rejected(
-            "checkout reservation session id mismatch".into(),
-        ));
-    }
-
-    let client_reference_id = session["client_reference_id"].as_str().unwrap_or_default();
-    if client_reference_id != checkout.user_id {
-        mark_checkout_status(state, &checkout.id, "invalid").await?;
-        return Ok(CheckoutCompletion::Rejected(
-            "checkout client_reference_id mismatch".into(),
-        ));
-    }
-
-    let payment_status = session["payment_status"].as_str().unwrap_or_default();
-    if payment_status != "paid" {
-        return Ok(CheckoutCompletion::Rejected(format!(
-            "checkout payment_status is {payment_status}"
-        )));
-    }
-
-    let currency = session["currency"]
-        .as_str()
-        .unwrap_or_default()
-        .to_ascii_lowercase();
-    if currency != checkout.currency {
-        mark_checkout_status(state, &checkout.id, "invalid").await?;
-        return Ok(CheckoutCompletion::Rejected(
-            "checkout currency mismatch".into(),
-        ));
-    }
-
-    let amount_subtotal = match number_field(session, "amount_subtotal") {
-        Some(amount) => amount,
-        None => {
-            mark_checkout_status(state, &checkout.id, "invalid").await?;
-            return Ok(CheckoutCompletion::Rejected(
-                "checkout amount_subtotal missing".into(),
-            ));
-        }
-    };
-    if amount_subtotal != checkout.amount_pence {
-        mark_checkout_status(state, &checkout.id, "invalid").await?;
-        return Ok(CheckoutCompletion::Rejected(
-            "checkout amount_subtotal mismatch".into(),
-        ));
-    }
-
-    let amount_total = match number_field(session, "amount_total") {
-        Some(amount) => amount,
-        None => {
-            mark_checkout_status(state, &checkout.id, "invalid").await?;
-            return Ok(CheckoutCompletion::Rejected(
-                "checkout amount_total missing".into(),
-            ));
-        }
-    };
-    if amount_total != checkout.expected_total_pence {
-        mark_checkout_status(state, &checkout.id, "invalid").await?;
-        return Ok(CheckoutCompletion::Rejected(
-            "checkout amount_total mismatch".into(),
-        ));
-    }
-
-    let verified = VerifiedCheckout {
-        reservation_id: checkout.id,
-        user_id: checkout.user_id,
-        stripe_session_id: session_id.to_string(),
-        payment_intent_id: payment_intent_id.to_string(),
-        paid_amount_pence: amount_total,
-        referral_invite_id: checkout.referral_invite_id,
-    };
-
-    if already_completed {
-        Ok(CheckoutCompletion::AlreadyHandled(verified))
-    } else {
-        Ok(CheckoutCompletion::Grant(verified))
-    }
-}
-
-pub async fn mark_checkout_completed(
-    state: &AppState,
-    reservation_id: &str,
-    paid_amount_pence: u64,
-    payment_intent_id: &str,
-) -> anyhow::Result<()> {
-    if !is_safe_stripe_session_id(payment_intent_id) {
-        return Err(anyhow!("invalid Stripe payment intent id"));
-    }
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "status": "completed",
-            "paid_amount_pence": paid_amount_pence,
-            "completed_at_unix": now_unix_secs().to_string(),
-            "stripe_payment_intent_id": payment_intent_id,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase checkout completion update failed")
-}
-
-pub async fn complete_verified_checkout(
-    state: &AppState,
-    checkout: &VerifiedCheckout,
-) -> anyhow::Result<()> {
-    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
-    let pricing_lock = acquire_pocketbase_lock(
-        state,
-        CHECKOUT_PRICING_LOCK_NAME,
-        CHECKOUT_PRICING_LOCK_TTL_SECS,
-    )
-    .await?;
-    let result = complete_verified_checkout_locked(state, checkout).await;
-    if let Err(err) = pricing_lock.release().await {
-        warn!("Failed to release checkout pricing lock: {err}");
-    }
-    result
-}
-
-async fn complete_verified_checkout_locked(
-    state: &AppState,
-    checkout: &VerifiedCheckout,
-) -> anyhow::Result<()> {
-    let live_checkout = find_checkout_by_stripe_session(state, &checkout.stripe_session_id)
-        .await?
-        .ok_or_else(|| anyhow!("checkout reservation disappeared before completion"))?;
-
-    if live_checkout.status == "completed" {
-        if !checkout.referral_invite_id.is_empty() {
-            mark_referral_invite_used(
-                state,
-                &checkout.referral_invite_id,
-                &checkout.user_id,
-                &checkout.reservation_id,
-            )
-            .await?;
-        }
-        return Ok(());
-    }
-    if live_checkout.id != checkout.reservation_id
-        || live_checkout.user_id != checkout.user_id
-        || live_checkout.referral_invite_id != checkout.referral_invite_id
-    {
-        mark_checkout_status(state, &checkout.reservation_id, "invalid").await?;
-        return Err(anyhow!("checkout reservation changed before completion"));
-    }
-    if live_checkout.status != "pending" && live_checkout.status != "expired" {
-        return Err(anyhow!("checkout reservation is {}", live_checkout.status));
-    }
-
-    grant_license(state, &checkout.user_id).await?;
-    mark_checkout_completed(
-        state,
-        &checkout.reservation_id,
-        checkout.paid_amount_pence,
-        &checkout.payment_intent_id,
-    )
-    .await?;
-    if !checkout.referral_invite_id.is_empty() {
-        mark_referral_invite_used(
-            state,
-            &checkout.referral_invite_id,
-            &checkout.user_id,
-            &checkout.reservation_id,
-        )
-        .await?;
-    }
-    Ok(())
-}
-
-pub async fn grant_license_with_pricing_lock(
-    state: &AppState,
-    user_id: &str,
-) -> anyhow::Result<()> {
-    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
-    let pricing_lock = acquire_pocketbase_lock(
-        state,
-        CHECKOUT_PRICING_LOCK_NAME,
-        CHECKOUT_PRICING_LOCK_TTL_SECS,
-    )
-    .await?;
-    let result = grant_license(state, user_id).await;
-    if let Err(err) = pricing_lock.release().await {
-        warn!("Failed to release checkout pricing lock: {err}");
-    }
-    result
-}
-
-pub async fn grant_license(state: &AppState, user_id: &str) -> anyhow::Result<()> {
-    set_user_subscription(state, user_id, "licensed").await
-}
-
-pub async fn reverse_license_for_payment_intent(
-    state: &AppState,
-    payment_intent_id: &str,
-    reason: &str,
-    refunded_amount_pence: Option<u64>,
-) -> anyhow::Result<PaymentReversalOutcome> {
-    if !is_safe_stripe_session_id(payment_intent_id) {
-        return Err(anyhow!("invalid Stripe payment intent id"));
-    }
-    if !is_safe_reversal_reason(reason) {
-        return Err(anyhow!("invalid Stripe reversal reason"));
-    }
-
-    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
-    let checkout = match find_checkout_by_payment_intent_or_checkout_session(
-        state,
-        payment_intent_id,
-    )
-    .await?
-    {
-        Some(checkout) => checkout,
-        None => return Ok(PaymentReversalOutcome::NoMatchingCheckout),
-    };
-
-    let paid_amount_pence = checkout
-        .paid_amount_pence
-        .max(checkout.expected_total_pence);
-    if let Some(refunded_amount_pence) = refunded_amount_pence {
-        if refunded_amount_pence < paid_amount_pence {
-            return Ok(PaymentReversalOutcome::IgnoredPartialRefund {
-                user_id: checkout.user_id,
-                refunded_amount_pence,
-                paid_amount_pence,
-            });
-        }
-    }
-
-    if checkout.status == "reversed" {
-        return Ok(PaymentReversalOutcome::AlreadyHandled {
-            user_id: checkout.user_id,
-        });
-    }
-
-    if matches!(checkout.status.as_str(), "pending" | "expired" | "failed") {
-        mark_checkout_reversed(state, &checkout.id, reason, payment_intent_id).await?;
-        return Ok(PaymentReversalOutcome::Applied {
-            user_id: checkout.user_id,
-        });
-    }
-
-    if checkout.status != "completed" {
-        return Ok(PaymentReversalOutcome::NotReversible {
-            user_id: checkout.user_id,
-            status: checkout.status,
-        });
-    }
-
-    let has_other_license = has_other_completed_checkout_for_user(
-        state,
-        &checkout.user_id,
-        &checkout.id,
-        payment_intent_id,
-    )
-    .await?;
-    if !has_other_license {
-        revoke_license(state, &checkout.user_id).await?;
-    }
-    mark_checkout_reversed(state, &checkout.id, reason, payment_intent_id).await?;
-
-    Ok(PaymentReversalOutcome::Applied {
-        user_id: checkout.user_id,
-    })
-}
-
-pub async fn reinstate_license_for_payment_intent(
-    state: &AppState,
-    payment_intent_id: &str,
-    reason: &str,
-) -> anyhow::Result<PaymentReinstatementOutcome> {
-    if !is_safe_stripe_session_id(payment_intent_id) {
-        return Err(anyhow!("invalid Stripe payment intent id"));
-    }
-    if !is_safe_reversal_reason(reason) {
-        return Err(anyhow!("invalid Stripe reinstatement reason"));
-    }
-
-    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
-    let checkout = match find_checkout_by_payment_intent_or_checkout_session(
-        state,
-        payment_intent_id,
-    )
-    .await?
-    {
-        Some(checkout) => checkout,
-        None => return Ok(PaymentReinstatementOutcome::NoMatchingCheckout),
-    };
-
-    if checkout.status == "completed" {
-        return Ok(PaymentReinstatementOutcome::AlreadyHandled {
-            user_id: checkout.user_id,
-        });
-    }
-
-    if checkout.status != "reversed" {
-        return Ok(PaymentReinstatementOutcome::Ignored {
-            user_id: checkout.user_id,
-            reason: format!("checkout status is {}", checkout.status),
-        });
-    }
-
-    if !checkout.reversal_reason.starts_with("charge.dispute.") {
-        return Ok(PaymentReinstatementOutcome::Ignored {
-            user_id: checkout.user_id,
-            reason: format!("checkout was reversed by {}", checkout.reversal_reason),
-        });
-    }
-
-    grant_license(state, &checkout.user_id).await?;
-    mark_checkout_reinstated(state, &checkout.id, reason).await?;
-
-    Ok(PaymentReinstatementOutcome::Applied {
-        user_id: checkout.user_id,
-    })
-}
-
-async fn revoke_license(state: &AppState, user_id: &str) -> anyhow::Result<()> {
-    set_user_subscription(state, user_id, "free").await
-}
-
-async fn set_user_subscription(
-    state: &AppState,
-    user_id: &str,
-    subscription: &str,
-) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/users/records/{user_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({ "subscription": subscription }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase license update failed")?;
-
-    state.token_cache.invalidate_by_user_id(user_id);
-    Ok(())
-}
-
-pub async fn mark_referral_invite_used(
-    state: &AppState,
-    invite_id: &str,
-    user_id: &str,
-    reservation_id: &str,
-) -> anyhow::Result<()> {
-    if invite_id.is_empty() {
-        return Ok(());
-    }
-    if !is_safe_pocketbase_id(invite_id)
-        || !is_safe_pocketbase_id(user_id)
-        || !is_safe_pocketbase_id(reservation_id)
-    {
-        return Err(anyhow!("invalid PocketBase id"));
-    }
-
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let invite = fetch_invite_record(state, pb_url, &token, invite_id).await?;
-
-    // A verified Stripe payment must not lose entitlement just because local
-    // invite reservation bookkeeping expired or moved before webhook delivery.
-    match referral_invite_completion_action(&invite, user_id, reservation_id) {
-        ReferralInviteCompletionAction::AlreadyRecorded => return Ok(()),
-        ReferralInviteCompletionAction::AlreadyUsedByAnother => {
-            warn!(
-                invite_id,
-                user_id,
-                existing_used_by = invite["used_by_id"].as_str().unwrap_or_default(),
-                "Referral invite was already used by another account; preserving verified checkout entitlement"
-            );
-            return Ok(());
-        }
-        ReferralInviteCompletionAction::Record {
-            reservation_reassigned,
-        } => {
-            if reservation_reassigned {
-                warn!(
-                    invite_id,
-                    user_id,
-                    reservation_id,
-                    reserved_by_id = invite["reserved_by_id"].as_str().unwrap_or_default(),
-                    reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default(),
-                    "Referral invite reservation moved before webhook completion; verified checkout will consume it"
-                );
-            }
-        }
-    }
-
-    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "used_by_id": user_id,
-            "used_at": now_unix_secs().to_string(),
-            "reserved_by_id": "",
-            "reserved_checkout_id": "",
-            "reserved_until_unix": 0,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase invite usage update failed")
-}
-
-#[derive(Debug, PartialEq, Eq)]
-enum ReferralInviteCompletionAction {
-    AlreadyRecorded,
-    AlreadyUsedByAnother,
-    Record { reservation_reassigned: bool },
-}
-
-fn referral_invite_completion_action(
-    invite: &Value,
-    user_id: &str,
-    reservation_id: &str,
-) -> ReferralInviteCompletionAction {
-    let existing_used_by = invite["used_by_id"].as_str().unwrap_or_default();
-    if existing_used_by == user_id {
-        return ReferralInviteCompletionAction::AlreadyRecorded;
-    }
-    if !existing_used_by.is_empty() {
-        return ReferralInviteCompletionAction::AlreadyUsedByAnother;
-    }
-
-    let reserved_by_id = invite["reserved_by_id"].as_str().unwrap_or_default();
-    let reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default();
-    let reservation_reassigned = (!reserved_by_id.is_empty() && reserved_by_id != user_id)
-        || (!reserved_checkout_id.is_empty() && reserved_checkout_id != reservation_id);
-
-    ReferralInviteCompletionAction::Record {
-        reservation_reassigned,
-    }
-}
-
-async fn fetch_invite_record(
-    state: &AppState,
-    pb_url: &str,
-    token: &str,
-    invite_id: &str,
-) -> anyhow::Result<Value> {
-    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    resp.json().await.map_err(Into::into)
-}
-
-async fn reserve_referral_invite(
-    state: &AppState,
-    invite_id: &str,
-    user_id: &str,
-    reservation_id: &str,
-    reserved_until_unix: u64,
-) -> anyhow::Result<()> {
-    if !is_safe_pocketbase_id(invite_id)
-        || !is_safe_pocketbase_id(user_id)
-        || !is_safe_pocketbase_id(reservation_id)
-    {
-        return Err(anyhow!("invalid PocketBase id"));
-    }
-
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let invite = fetch_invite_record(state, pb_url, &token, invite_id).await?;
-    let used_by = invite["used_by_id"].as_str().unwrap_or_default();
-    if !used_by.is_empty() {
-        return Err(anyhow!("referral invite already used"));
-    }
-
-    let now = now_unix_secs();
-    let reserved_by_id = invite["reserved_by_id"].as_str().unwrap_or_default();
-    let reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default();
-    let existing_reserved_until = number_field(&invite, "reserved_until_unix").unwrap_or(0);
-    let reservation_is_live = existing_reserved_until >= now;
-    if reservation_is_live
-        && !reserved_checkout_id.is_empty()
-        && reserved_checkout_id != reservation_id
-    {
-        return Err(anyhow!("referral invite already has an active checkout"));
-    }
-    if reservation_is_live && !reserved_by_id.is_empty() && reserved_by_id != user_id {
-        return Err(anyhow!("referral invite reserved by another account"));
-    }
-
-    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "reserved_by_id": user_id,
-            "reserved_checkout_id": reservation_id,
-            "reserved_until_unix": reserved_until_unix,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase invite reservation update failed")
-}
-
-async fn release_referral_invite_reservation(
-    state: &AppState,
-    invite_id: &str,
-    reservation_id: &str,
-) -> anyhow::Result<()> {
-    if !is_safe_pocketbase_id(invite_id) || !is_safe_pocketbase_id(reservation_id) {
-        return Err(anyhow!("invalid PocketBase id"));
-    }
-
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let invite = fetch_invite_record(state, pb_url, &token, invite_id).await?;
-    let used_by = invite["used_by_id"].as_str().unwrap_or_default();
-    let reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default();
-    if !used_by.is_empty() || reserved_checkout_id != reservation_id {
-        return Ok(());
-    }
-
-    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "reserved_by_id": "",
-            "reserved_checkout_id": "",
-            "reserved_until_unix": 0,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase invite reservation release failed")
-}
-
-pub async fn active_referral_checkout_user(
-    state: &AppState,
-    invite_id: &str,
-) -> anyhow::Result<Option<String>> {
-    if !is_safe_pocketbase_id(invite_id) {
-        return Err(anyhow!("invalid PocketBase invite id"));
-    }
-
-    let now = now_unix_secs();
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = format!(
-        "status=\"pending\" && expires_at_unix>={now} && referral_invite_id=\"{}\"",
-        invite_id
-    );
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    Ok(body["items"]
-        .as_array()
-        .and_then(|items| items.first())
-        .and_then(|item| item["user"].as_str())
-        .map(str::to_string))
-}
-
-async fn count_active_pending_checkouts(state: &AppState, now: u64) -> anyhow::Result<u64> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = format!("status=\"pending\" && expires_at_unix>={now}");
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    Ok(body["totalItems"].as_u64().unwrap_or(0))
-}
-
-async fn find_active_checkout_for_user(
-    state: &AppState,
-    user_id: &str,
-    discount_coupon_id: &str,
-    referral_invite_id: &str,
-    now: u64,
-) -> anyhow::Result<Option<PendingCheckout>> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = active_checkout_filter(user_id, discount_coupon_id, referral_invite_id, now)?;
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    let item = body["items"]
-        .as_array()
-        .and_then(|items| items.first())
-        .cloned();
-
-    item.map(parse_pending_checkout).transpose()
-}
-
-fn active_checkout_filter(
-    user_id: &str,
-    discount_coupon_id: &str,
-    referral_invite_id: &str,
-    now: u64,
-) -> anyhow::Result<String> {
-    if !is_safe_pocketbase_id(user_id) {
-        return Err(anyhow!("invalid PocketBase user id"));
-    }
-    if !discount_coupon_id.is_empty() && !is_safe_stripe_session_id(discount_coupon_id) {
-        return Err(anyhow!("invalid Stripe coupon id"));
-    }
-    if !referral_invite_id.is_empty() && !is_safe_pocketbase_id(referral_invite_id) {
-        return Err(anyhow!("invalid PocketBase referral invite id"));
-    }
-
-    Ok(format!(
-        "status=\"pending\" && expires_at_unix>={now} && user=\"{user_id}\" && discount_coupon_id=\"{discount_coupon_id}\" && referral_invite_id=\"{referral_invite_id}\""
-    ))
-}
-
-async fn expire_stale_pending_checkouts(state: &AppState, now: u64) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = format!("status=\"pending\" && expires_at_unix<{now}");
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=50",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    let Some(items) = body["items"].as_array() else {
-        return Ok(());
-    };
-
-    for item in items {
-        let Some(id) = item["id"].as_str() else {
-            continue;
-        };
-        if let Err(err) = mark_checkout_status(state, id, "expired").await {
-            warn!(
-                reservation_id = id,
-                "Failed to expire checkout reservation: {err}"
-            );
-        }
-        if let Some(invite_id) = item["referral_invite_id"]
-            .as_str()
-            .filter(|invite_id| !invite_id.is_empty())
-        {
-            if let Err(err) = release_referral_invite_reservation(state, invite_id, id).await {
-                warn!(
-                    reservation_id = id,
-                    referral_invite_id = invite_id,
-                    "Failed to release expired referral invite reservation: {err}"
-                );
-            }
-        }
-    }
-
-    Ok(())
-}
-
-struct PendingCheckoutInput<'a> {
-    user_id: &'a str,
-    amount_pence: u64,
-    expected_total_pence: u64,
-    currency: &'a str,
-    discount_coupon_id: &'a str,
-    referral_invite_id: &'a str,
-    expires_at_unix: u64,
-}
-
-async fn create_pending_checkout(
-    state: &AppState,
-    input: PendingCheckoutInput<'_>,
-) -> anyhow::Result<String> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records");
-    let resp = state
-        .http_client
-        .post(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "user": input.user_id,
-            "stripe_session_id": "",
-            "stripe_payment_intent_id": "",
-            "checkout_url": "",
-            "amount_pence": input.amount_pence,
-            "expected_total_pence": input.expected_total_pence,
-            "currency": input.currency,
-            "discount_coupon_id": input.discount_coupon_id,
-            "referral_invite_id": input.referral_invite_id,
-            "status": "pending",
-            "expires_at_unix": input.expires_at_unix,
-            "paid_amount_pence": 0,
-            "completed_at_unix": "",
-            "reversal_reason": "",
-        }))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    body["id"]
-        .as_str()
-        .map(str::to_string)
-        .ok_or_else(|| anyhow!("PocketBase checkout reservation missing id"))
-}
-
-/// Fetch a Stripe coupon and ensure its `percent_off` matches the expected
-/// referral discount AND that it has no `amount_off` override. This blocks a
-/// misconfigured (or maliciously swapped) coupon ID from quietly granting a
-/// larger discount than the server's pricing math assumed.
-async fn verify_stripe_coupon_discount(state: &AppState, coupon_id: &str) -> anyhow::Result<()> {
-    if !is_safe_stripe_session_id(coupon_id) {
-        return Err(anyhow!("unsafe stripe coupon id"));
-    }
-    let url = format!(
-        "https://api.stripe.com/v1/coupons/{}",
-        urlencoding::encode(coupon_id)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .basic_auth(&state.stripe_secret_key, None::<&str>)
-        .send()
-        .await
-        .context("Stripe coupon fetch failed")?;
-    ensure_success_ref(&resp)
-        .await
-        .context("Stripe coupon fetch returned error")?;
-    let body: Value = resp
-        .json()
-        .await
-        .context("Failed to parse Stripe coupon response")?;
-
-    let valid = body["valid"].as_bool().unwrap_or(false);
-    if !valid {
-        return Err(anyhow!("stripe coupon is not valid"));
-    }
-    if body["amount_off"].is_number() {
-        return Err(anyhow!(
-            "stripe coupon uses amount_off; only percent_off is permitted"
-        ));
-    }
-    let percent_off = body["percent_off"]
-        .as_f64()
-        .ok_or_else(|| anyhow!("stripe coupon missing percent_off"))?;
-    if percent_off.is_nan() || (percent_off - REFERRAL_DISCOUNT_PERCENT as f64).abs() > 0.001 {
-        return Err(anyhow!(
-            "stripe coupon percent_off ({percent_off}) does not match expected {REFERRAL_DISCOUNT_PERCENT}"
-        ));
-    }
-    Ok(())
-}
-
-#[allow(clippy::too_many_arguments)]
-async fn create_stripe_session(
-    state: &AppState,
-    user: &PocketBaseUser,
-    reservation_id: &str,
-    price_pence: u64,
-    success_url: &str,
-    cancel_url: &str,
-    expires_at_unix: u64,
-    discount_coupon_id: Option<&str>,
-) -> anyhow::Result<(String, String)> {
-    if let Some(coupon_id) = discount_coupon_id.filter(|id| !id.is_empty()) {
-        verify_stripe_coupon_discount(state, coupon_id).await?;
-    }
-
-    let mut form_params = vec![
-        ("mode", "payment".to_string()),
-        ("payment_method_types[0]", "card".to_string()),
-        (
-            "line_items[0][price_data][unit_amount]",
-            price_pence.to_string(),
-        ),
-        (
-            "line_items[0][price_data][currency]",
-            CHECKOUT_CURRENCY.to_string(),
-        ),
-        (
-            "line_items[0][price_data][product_data][name]",
-            CHECKOUT_PRODUCT_NAME.to_string(),
-        ),
-        ("line_items[0][quantity]", "1".to_string()),
-        ("success_url", success_url.to_string()),
-        ("cancel_url", cancel_url.to_string()),
-        ("expires_at", expires_at_unix.to_string()),
-        ("client_reference_id", user.id.clone()),
-        ("customer_email", user.email.clone()),
-        ("metadata[pending_checkout_id]", reservation_id.to_string()),
-        ("metadata[expected_amount_pence]", price_pence.to_string()),
-        (
-            "metadata[expected_total_pence]",
-            expected_total_for_checkout(price_pence, discount_coupon_id).to_string(),
-        ),
-        ("metadata[expected_currency]", CHECKOUT_CURRENCY.to_string()),
-    ];
-
-    if let Some(coupon_id) = discount_coupon_id.filter(|id| !id.is_empty()) {
-        form_params.push(("discounts[0][coupon]", coupon_id.to_string()));
-        form_params.push(("metadata[discount_coupon_id]", coupon_id.to_string()));
-    }
-
-    let resp = state
-        .http_client
-        .post("https://api.stripe.com/v1/checkout/sessions")
-        .basic_auth(&state.stripe_secret_key, None::<&str>)
-        .form(&form_params)
-        .send()
-        .await
-        .context("Stripe checkout request failed")?;
-
-    ensure_success_ref(&resp)
-        .await
-        .context("Stripe checkout failed")?;
-
-    let body: Value = resp
-        .json()
-        .await
-        .context("Failed to parse Stripe response")?;
-    let session_id = body["id"]
-        .as_str()
-        .filter(|id| is_safe_stripe_session_id(id))
-        .map(str::to_string)
-        .ok_or_else(|| anyhow!("Stripe session missing valid id"))?;
-    let url = body["url"]
-        .as_str()
-        .map(str::to_string)
-        .filter(|url| !url.is_empty())
-        .ok_or_else(|| anyhow!("Stripe session missing URL"))?;
-
-    Ok((session_id, url))
-}
-
-async fn attach_stripe_session(
-    state: &AppState,
-    reservation_id: &str,
-    stripe_session_id: &str,
-    checkout_url: &str,
-) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "stripe_session_id": stripe_session_id,
-            "checkout_url": checkout_url,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase checkout session attach failed")
-}
-
-async fn mark_checkout_status(
-    state: &AppState,
-    reservation_id: &str,
-    status: &str,
-) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({ "status": status }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .with_context(|| format!("PocketBase checkout status update failed for {reservation_id}"))
-}
-
-async fn mark_checkout_reversed(
-    state: &AppState,
-    reservation_id: &str,
-    reason: &str,
-    payment_intent_id: &str,
-) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "status": "reversed",
-            "reversal_reason": reason,
-            "stripe_payment_intent_id": payment_intent_id,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .with_context(|| format!("PocketBase checkout reversal update failed for {reservation_id}"))
-}
-
-async fn mark_checkout_reinstated(
-    state: &AppState,
-    reservation_id: &str,
-    _reason: &str,
-) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "status": "completed",
-            "reversal_reason": "",
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp).await.with_context(|| {
-        format!("PocketBase checkout reinstatement update failed for {reservation_id}")
-    })
-}
-
-async fn find_checkout_by_stripe_session(
-    state: &AppState,
-    stripe_session_id: &str,
-) -> anyhow::Result<Option<PendingCheckout>> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = format!("stripe_session_id=\"{}\"", stripe_session_id);
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    let item = body["items"]
-        .as_array()
-        .and_then(|items| items.first())
-        .cloned();
-
-    item.map(parse_pending_checkout).transpose()
-}
-
-async fn find_checkout_by_payment_intent(
-    state: &AppState,
-    payment_intent_id: &str,
-) -> anyhow::Result<Option<PendingCheckout>> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = format!("stripe_payment_intent_id=\"{}\"", payment_intent_id);
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    let item = body["items"]
-        .as_array()
-        .and_then(|items| items.first())
-        .cloned();
-
-    item.map(parse_pending_checkout).transpose()
-}
-
-async fn find_checkout_by_payment_intent_or_checkout_session(
-    state: &AppState,
-    payment_intent_id: &str,
-) -> anyhow::Result<Option<PendingCheckout>> {
-    if let Some(checkout) = find_checkout_by_payment_intent(state, payment_intent_id).await? {
-        return Ok(Some(checkout));
-    }
-
-    let Some(session_id) =
-        fetch_stripe_checkout_session_id_for_payment_intent(state, payment_intent_id).await?
-    else {
-        return Ok(None);
-    };
-
-    let Some(mut checkout) = find_checkout_by_stripe_session(state, &session_id).await? else {
-        return Ok(None);
-    };
-
-    if checkout.payment_intent_id.is_empty() {
-        attach_payment_intent_to_checkout(state, &checkout.id, payment_intent_id).await?;
-        checkout.payment_intent_id = payment_intent_id.to_string();
-    } else if checkout.payment_intent_id != payment_intent_id {
-        mark_checkout_status(state, &checkout.id, "invalid").await?;
-        return Err(anyhow!(
-            "checkout reservation payment intent changed before reversal"
-        ));
-    }
-
-    Ok(Some(checkout))
-}
-
-async fn fetch_stripe_checkout_session_id_for_payment_intent(
-    state: &AppState,
-    payment_intent_id: &str,
-) -> anyhow::Result<Option<String>> {
-    let url = format!(
-        "https://api.stripe.com/v1/checkout/sessions?payment_intent={}&limit=1",
-        urlencoding::encode(payment_intent_id)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .basic_auth(&state.stripe_secret_key, None::<&str>)
-        .send()
-        .await
-        .context("Stripe checkout session lookup failed")?;
-
-    ensure_success_ref(&resp)
-        .await
-        .context("Stripe checkout session lookup returned error")?;
-
-    let body: Value = resp
-        .json()
-        .await
-        .context("Failed to parse Stripe checkout session lookup")?;
-    Ok(body["data"]
-        .as_array()
-        .and_then(|items| items.first())
-        .and_then(|item| item["id"].as_str())
-        .filter(|id| is_safe_stripe_session_id(id))
-        .map(str::to_string))
-}
-
-async fn attach_payment_intent_to_checkout(
-    state: &AppState,
-    reservation_id: &str,
-    payment_intent_id: &str,
-) -> anyhow::Result<()> {
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
-    let resp = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&serde_json::json!({
-            "stripe_payment_intent_id": payment_intent_id,
-        }))
-        .send()
-        .await?;
-
-    ensure_success(resp)
-        .await
-        .context("PocketBase checkout payment intent attach failed")
-}
-
-async fn has_other_completed_checkout_for_user(
-    state: &AppState,
-    user_id: &str,
-    reservation_id: &str,
-    payment_intent_id: &str,
-) -> anyhow::Result<bool> {
-    if !is_safe_pocketbase_id(user_id) || !is_safe_pocketbase_id(reservation_id) {
-        return Err(anyhow!("invalid PocketBase id"));
-    }
-    if !is_safe_stripe_session_id(payment_intent_id) {
-        return Err(anyhow!("invalid Stripe payment intent id"));
-    }
-
-    let token = get_superuser_token(state).await?;
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let filter = format!("user=\"{user_id}\" && status=\"completed\"");
-    let url = format!(
-        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=50",
-        urlencoding::encode(&filter)
-    );
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await?;
-
-    ensure_success_ref(&resp).await?;
-
-    let body: Value = resp.json().await?;
-    let Some(items) = body["items"].as_array() else {
-        return Ok(false);
-    };
-
-    Ok(items.iter().any(|item| {
-        let other_id = item["id"].as_str().unwrap_or_default();
-        let other_payment_intent = item["stripe_payment_intent_id"]
-            .as_str()
-            .unwrap_or_default();
-        other_id != reservation_id && other_payment_intent != payment_intent_id
-    }))
-}
-
-fn parse_pending_checkout(item: Value) -> anyhow::Result<PendingCheckout> {
-    Ok(PendingCheckout {
-        id: item["id"]
-            .as_str()
-            .ok_or_else(|| anyhow!("checkout reservation missing id"))?
-            .to_string(),
-        user_id: item["user"]
-            .as_str()
-            .ok_or_else(|| anyhow!("checkout reservation missing user"))?
-            .to_string(),
-        stripe_session_id: item["stripe_session_id"]
-            .as_str()
-            .unwrap_or_default()
-            .to_string(),
-        checkout_url: item["checkout_url"]
-            .as_str()
-            .unwrap_or_default()
-            .to_string(),
-        amount_pence: number_field(&item, "amount_pence")
-            .ok_or_else(|| anyhow!("checkout reservation missing amount_pence"))?,
-        expected_total_pence: number_field(&item, "expected_total_pence")
-            .ok_or_else(|| anyhow!("checkout reservation missing expected_total_pence"))?,
-        currency: item["currency"]
-            .as_str()
-            .unwrap_or_default()
-            .to_ascii_lowercase(),
-        referral_invite_id: item["referral_invite_id"]
-            .as_str()
-            .unwrap_or_default()
-            .to_string(),
-        status: item["status"].as_str().unwrap_or_default().to_string(),
-        payment_intent_id: item["stripe_payment_intent_id"]
-            .as_str()
-            .unwrap_or_default()
-            .to_string(),
-        paid_amount_pence: number_field(&item, "paid_amount_pence").unwrap_or(0),
-        reversal_reason: item["reversal_reason"]
-            .as_str()
-            .unwrap_or_default()
-            .to_string(),
-    })
-}
-
-fn expected_total_for_checkout(amount_pence: u64, discount_coupon_id: Option<&str>) -> u64 {
-    if discount_coupon_id.is_some_and(|id| !id.is_empty()) {
-        return ((amount_pence * (100 - REFERRAL_DISCOUNT_PERCENT)) / 100).max(1);
-    }
-    amount_pence
-}
-
-fn number_field(value: &Value, field: &str) -> Option<u64> {
-    value[field].as_u64().or_else(|| {
-        value[field]
-            .as_f64()
-            .filter(|n| n.is_finite() && *n >= 0.0 && n.fract() == 0.0)
-            .map(|n| n as u64)
-    })
-}
-
-fn is_safe_stripe_session_id(id: &str) -> bool {
-    !id.is_empty()
-        && id.len() <= 128
-        && id
-            .bytes()
-            .all(|b| b.is_ascii_alphanumeric() || b == b'_' || b == b'-')
-}
-
-fn is_safe_pocketbase_id(id: &str) -> bool {
-    !id.is_empty() && id.len() <= 32 && id.bytes().all(|b| b.is_ascii_alphanumeric())
-}
-
-fn is_safe_reversal_reason(reason: &str) -> bool {
-    !reason.is_empty()
-        && reason.len() <= 128
-        && reason
-            .bytes()
-            .all(|b| b.is_ascii_alphanumeric() || b == b'_' || b == b'-' || b == b'.')
-}
-
-async fn ensure_success(resp: reqwest::Response) -> anyhow::Result<()> {
-    if resp.status().is_success() {
-        return Ok(());
-    }
-
-    let status = resp.status();
-    let text = resp.text().await.unwrap_or_default();
-    Err(anyhow!("upstream returned {status}: {text}"))
-}
-
-async fn ensure_success_ref(resp: &reqwest::Response) -> anyhow::Result<()> {
-    if resp.status().is_success() {
-        return Ok(());
-    }
-
-    Err(anyhow!("upstream returned {}", resp.status()))
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn active_checkout_filter_includes_empty_context_for_standard_checkout() {
-        let filter = active_checkout_filter("abc123", "", "", 42).unwrap();
-        assert_eq!(
-            filter,
-            "status=\"pending\" && expires_at_unix>=42 && user=\"abc123\" && discount_coupon_id=\"\" && referral_invite_id=\"\""
-        );
-    }
-
-    #[test]
-    fn active_checkout_filter_includes_referral_context() {
-        let filter = active_checkout_filter("user123", "coupon_30", "invite123", 99).unwrap();
-        assert_eq!(
-            filter,
-            "status=\"pending\" && expires_at_unix>=99 && user=\"user123\" && discount_coupon_id=\"coupon_30\" && referral_invite_id=\"invite123\""
-        );
-    }
-
-    #[test]
-    fn active_checkout_filter_rejects_unsafe_context_values() {
-        assert!(active_checkout_filter("user123", "bad\"coupon", "", 1).is_err());
-        assert!(active_checkout_filter("user123", "", "bad-invite", 1).is_err());
-        assert!(active_checkout_filter("bad-user", "", "", 1).is_err());
-    }
-
-    #[test]
-    fn expected_total_for_referral_discount_rounds_down_like_stripe_amount_math() {
-        assert_eq!(expected_total_for_checkout(999, Some("coupon_30")), 699);
-        assert_eq!(expected_total_for_checkout(1, Some("coupon_30")), 1);
-        assert_eq!(expected_total_for_checkout(999, None), 999);
-    }
-
-    #[test]
-    fn referral_invite_completion_records_available_invite() {
-        let invite = serde_json::json!({
-            "used_by_id": "",
-            "reserved_by_id": "",
-            "reserved_checkout_id": "",
-        });
-
-        assert_eq!(
-            referral_invite_completion_action(&invite, "user123", "checkout123"),
-            ReferralInviteCompletionAction::Record {
-                reservation_reassigned: false
-            }
-        );
-    }
-
-    #[test]
-    fn referral_invite_completion_records_reassigned_reservation() {
-        let invite = serde_json::json!({
-            "used_by_id": "",
-            "reserved_by_id": "otheruser",
-            "reserved_checkout_id": "othercheckout",
-        });
-
-        assert_eq!(
-            referral_invite_completion_action(&invite, "user123", "checkout123"),
-            ReferralInviteCompletionAction::Record {
-                reservation_reassigned: true
-            }
-        );
-    }
-
-    #[test]
-    fn referral_invite_completion_detects_existing_usage() {
-        let used_by_same_user = serde_json::json!({ "used_by_id": "user123" });
-        let used_by_another_user = serde_json::json!({ "used_by_id": "otheruser" });
-
-        assert_eq!(
-            referral_invite_completion_action(&used_by_same_user, "user123", "checkout123"),
-            ReferralInviteCompletionAction::AlreadyRecorded
-        );
-        assert_eq!(
-            referral_invite_completion_action(&used_by_another_user, "user123", "checkout123"),
-            ReferralInviteCompletionAction::AlreadyUsedByAnother
-        );
-    }
-}
diff --git a/server-rs/src/checkout_sessions/lifecycle.rs b/server-rs/src/checkout_sessions/lifecycle.rs
new file mode 100644
index 0000000..9d515a9
--- /dev/null
+++ b/server-rs/src/checkout_sessions/lifecycle.rs
@@ -0,0 +1,589 @@
+//! The checkout session state machine: starting a checkout (with pricing and
+//! reservation under a cross-instance lock), verifying Stripe's completion
+//! payload, completing/granting, and reversing/reinstating after refunds or
+//! disputes.
+
+use std::sync::LazyLock;
+
+use anyhow::{anyhow, Context};
+use serde_json::Value;
+use tokio::sync::Mutex;
+use tracing::warn;
+
+use crate::auth::PocketBaseUser;
+use crate::pocketbase::get_superuser_token;
+use crate::pocketbase_locks::acquire_pocketbase_lock;
+use crate::routes::pricing::{count_licensed_users, price_for_count};
+use crate::state::AppState;
+
+use super::records::{
+    attach_stripe_session, count_active_pending_checkouts, create_pending_checkout,
+    expire_stale_pending_checkouts, find_active_checkout_for_user,
+    find_checkout_by_payment_intent_or_checkout_session, find_checkout_by_stripe_session,
+    has_other_completed_checkout_for_user, mark_checkout_completed, mark_checkout_reinstated,
+    mark_checkout_reversed, mark_checkout_status, PendingCheckoutInput,
+};
+use super::referral::{
+    mark_referral_invite_used, release_referral_invite_reservation, reserve_referral_invite,
+};
+use super::stripe::create_stripe_session;
+use super::{
+    ensure_success, is_safe_reversal_reason, is_safe_stripe_session_id, now_unix_secs,
+    number_field, CheckoutCompletion, CheckoutStart, PaymentReinstatementOutcome,
+    PaymentReversalOutcome, VerifiedCheckout, CHECKOUT_CURRENCY, REFERRAL_DISCOUNT_PERCENT,
+};
+
+const CHECKOUT_SESSION_TTL_SECS: u64 = 31 * 60;
+const CHECKOUT_PRICING_LOCK_NAME: &str = "checkout:pricing";
+const CHECKOUT_PRICING_LOCK_TTL_SECS: u64 = 5 * 60;
+
+static CHECKOUT_RESERVATION_LOCK: LazyLock<Mutex<()>> = LazyLock::new(|| Mutex::new(()));
+
+pub async fn start_license_checkout(
+    state: &AppState,
+    user: &PocketBaseUser,
+    success_url: &str,
+    cancel_url: &str,
+    discount_coupon_id: Option<&str>,
+    referral_invite_id: Option<&str>,
+) -> anyhow::Result<CheckoutStart> {
+    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
+    let pricing_lock = acquire_pocketbase_lock(
+        state,
+        CHECKOUT_PRICING_LOCK_NAME,
+        CHECKOUT_PRICING_LOCK_TTL_SECS,
+    )
+    .await?;
+    let result = start_license_checkout_locked(
+        state,
+        user,
+        success_url,
+        cancel_url,
+        discount_coupon_id,
+        referral_invite_id,
+    )
+    .await;
+    if let Err(err) = pricing_lock.release().await {
+        warn!("Failed to release checkout pricing lock: {err}");
+    }
+    result
+}
+
+async fn start_license_checkout_locked(
+    state: &AppState,
+    user: &PocketBaseUser,
+    success_url: &str,
+    cancel_url: &str,
+    discount_coupon_id: Option<&str>,
+    referral_invite_id: Option<&str>,
+) -> anyhow::Result<CheckoutStart> {
+    let now = now_unix_secs();
+    expire_stale_pending_checkouts(state, now).await?;
+
+    if let Some(existing) = find_active_checkout_for_user(
+        state,
+        &user.id,
+        discount_coupon_id.unwrap_or_default(),
+        referral_invite_id.unwrap_or_default(),
+        now,
+    )
+    .await?
+    {
+        if !existing.checkout_url.is_empty() {
+            return Ok(CheckoutStart::Stripe {
+                url: existing.checkout_url,
+            });
+        }
+        if let Err(err) = mark_checkout_status(state, &existing.id, "failed").await {
+            warn!(
+                reservation_id = %existing.id,
+                "Failed to fail incomplete checkout reservation: {err}"
+            );
+        }
+    }
+
+    let licensed_count = count_licensed_users(state).await?;
+    let pending_count = count_active_pending_checkouts(state, now).await?;
+    let price_pence = price_for_count(licensed_count + pending_count);
+
+    if price_pence == 0 {
+        grant_license(state, &user.id).await?;
+        return Ok(CheckoutStart::Free);
+    }
+
+    let expires_at_unix = now + CHECKOUT_SESSION_TTL_SECS;
+    let expected_total_pence = expected_total_for_checkout(price_pence, discount_coupon_id);
+    let reservation_id = create_pending_checkout(
+        state,
+        PendingCheckoutInput {
+            user_id: &user.id,
+            amount_pence: price_pence,
+            expected_total_pence,
+            currency: CHECKOUT_CURRENCY,
+            discount_coupon_id: discount_coupon_id.unwrap_or_default(),
+            referral_invite_id: referral_invite_id.unwrap_or_default(),
+            expires_at_unix,
+        },
+    )
+    .await?;
+
+    if let Some(invite_id) = referral_invite_id.filter(|id| !id.is_empty()) {
+        if let Err(err) =
+            reserve_referral_invite(state, invite_id, &user.id, &reservation_id, expires_at_unix)
+                .await
+        {
+            if let Err(mark_err) = mark_checkout_status(state, &reservation_id, "failed").await {
+                warn!(
+                    reservation_id,
+                    "Failed to mark checkout reservation failed: {mark_err}"
+                );
+            }
+            return Err(err);
+        }
+    }
+
+    let stripe_result = create_stripe_session(
+        state,
+        user,
+        &reservation_id,
+        price_pence,
+        success_url,
+        cancel_url,
+        expires_at_unix,
+        discount_coupon_id,
+    )
+    .await;
+
+    let (stripe_session_id, url) = match stripe_result {
+        Ok(session) => session,
+        Err(err) => {
+            if let Err(mark_err) = mark_checkout_status(state, &reservation_id, "failed").await {
+                warn!(
+                    reservation_id,
+                    "Failed to mark checkout reservation failed: {mark_err}"
+                );
+            }
+            if let Some(invite_id) = referral_invite_id.filter(|id| !id.is_empty()) {
+                if let Err(release_err) =
+                    release_referral_invite_reservation(state, invite_id, &reservation_id).await
+                {
+                    warn!(
+                        reservation_id,
+                        referral_invite_id = invite_id,
+                        "Failed to release referral invite reservation: {release_err}"
+                    );
+                }
+            }
+            return Err(err);
+        }
+    };
+
+    if let Err(err) = attach_stripe_session(state, &reservation_id, &stripe_session_id, &url).await
+    {
+        if let Err(mark_err) = mark_checkout_status(state, &reservation_id, "failed").await {
+            warn!(
+                reservation_id,
+                "Failed to mark checkout reservation failed: {mark_err}"
+            );
+        }
+        if let Some(invite_id) = referral_invite_id.filter(|id| !id.is_empty()) {
+            if let Err(release_err) =
+                release_referral_invite_reservation(state, invite_id, &reservation_id).await
+            {
+                warn!(
+                    reservation_id,
+                    referral_invite_id = invite_id,
+                    "Failed to release referral invite reservation: {release_err}"
+                );
+            }
+        }
+        return Err(err);
+    }
+
+    Ok(CheckoutStart::Stripe { url })
+}
+
+pub async fn verify_checkout_completion(
+    state: &AppState,
+    session: &Value,
+) -> anyhow::Result<CheckoutCompletion> {
+    let session_id = match session["id"].as_str() {
+        Some(id) if is_safe_stripe_session_id(id) => id,
+        _ => {
+            return Ok(CheckoutCompletion::Rejected(
+                "missing or invalid session id".into(),
+            ))
+        }
+    };
+    let payment_intent_id = match session["payment_intent"].as_str() {
+        Some(id) if is_safe_stripe_session_id(id) => id,
+        _ => {
+            return Ok(CheckoutCompletion::Rejected(
+                "missing or invalid payment intent id".into(),
+            ))
+        }
+    };
+
+    let checkout = match find_checkout_by_stripe_session(state, session_id).await? {
+        Some(checkout) => checkout,
+        None => {
+            return Ok(CheckoutCompletion::Rejected(
+                "checkout session has no reservation".into(),
+            ))
+        }
+    };
+
+    let already_completed = checkout.status == "completed";
+    if !already_completed && checkout.status != "pending" && checkout.status != "expired" {
+        return Ok(CheckoutCompletion::Rejected(format!(
+            "checkout reservation is {}",
+            checkout.status
+        )));
+    }
+    if checkout.stripe_session_id != session_id {
+        mark_checkout_status(state, &checkout.id, "invalid").await?;
+        return Ok(CheckoutCompletion::Rejected(
+            "checkout reservation session id mismatch".into(),
+        ));
+    }
+
+    let client_reference_id = session["client_reference_id"].as_str().unwrap_or_default();
+    if client_reference_id != checkout.user_id {
+        mark_checkout_status(state, &checkout.id, "invalid").await?;
+        return Ok(CheckoutCompletion::Rejected(
+            "checkout client_reference_id mismatch".into(),
+        ));
+    }
+
+    let payment_status = session["payment_status"].as_str().unwrap_or_default();
+    if payment_status != "paid" {
+        return Ok(CheckoutCompletion::Rejected(format!(
+            "checkout payment_status is {payment_status}"
+        )));
+    }
+
+    let currency = session["currency"]
+        .as_str()
+        .unwrap_or_default()
+        .to_ascii_lowercase();
+    if currency != checkout.currency {
+        mark_checkout_status(state, &checkout.id, "invalid").await?;
+        return Ok(CheckoutCompletion::Rejected(
+            "checkout currency mismatch".into(),
+        ));
+    }
+
+    let amount_subtotal = match number_field(session, "amount_subtotal") {
+        Some(amount) => amount,
+        None => {
+            mark_checkout_status(state, &checkout.id, "invalid").await?;
+            return Ok(CheckoutCompletion::Rejected(
+                "checkout amount_subtotal missing".into(),
+            ));
+        }
+    };
+    if amount_subtotal != checkout.amount_pence {
+        mark_checkout_status(state, &checkout.id, "invalid").await?;
+        return Ok(CheckoutCompletion::Rejected(
+            "checkout amount_subtotal mismatch".into(),
+        ));
+    }
+
+    let amount_total = match number_field(session, "amount_total") {
+        Some(amount) => amount,
+        None => {
+            mark_checkout_status(state, &checkout.id, "invalid").await?;
+            return Ok(CheckoutCompletion::Rejected(
+                "checkout amount_total missing".into(),
+            ));
+        }
+    };
+    if amount_total != checkout.expected_total_pence {
+        mark_checkout_status(state, &checkout.id, "invalid").await?;
+        return Ok(CheckoutCompletion::Rejected(
+            "checkout amount_total mismatch".into(),
+        ));
+    }
+
+    let verified = VerifiedCheckout {
+        reservation_id: checkout.id,
+        user_id: checkout.user_id,
+        stripe_session_id: session_id.to_string(),
+        payment_intent_id: payment_intent_id.to_string(),
+        paid_amount_pence: amount_total,
+        referral_invite_id: checkout.referral_invite_id,
+    };
+
+    if already_completed {
+        Ok(CheckoutCompletion::AlreadyHandled(verified))
+    } else {
+        Ok(CheckoutCompletion::Grant(verified))
+    }
+}
+
+pub async fn complete_verified_checkout(
+    state: &AppState,
+    checkout: &VerifiedCheckout,
+) -> anyhow::Result<()> {
+    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
+    let pricing_lock = acquire_pocketbase_lock(
+        state,
+        CHECKOUT_PRICING_LOCK_NAME,
+        CHECKOUT_PRICING_LOCK_TTL_SECS,
+    )
+    .await?;
+    let result = complete_verified_checkout_locked(state, checkout).await;
+    if let Err(err) = pricing_lock.release().await {
+        warn!("Failed to release checkout pricing lock: {err}");
+    }
+    result
+}
+
+async fn complete_verified_checkout_locked(
+    state: &AppState,
+    checkout: &VerifiedCheckout,
+) -> anyhow::Result<()> {
+    let live_checkout = find_checkout_by_stripe_session(state, &checkout.stripe_session_id)
+        .await?
+        .ok_or_else(|| anyhow!("checkout reservation disappeared before completion"))?;
+
+    if live_checkout.status == "completed" {
+        if !checkout.referral_invite_id.is_empty() {
+            mark_referral_invite_used(
+                state,
+                &checkout.referral_invite_id,
+                &checkout.user_id,
+                &checkout.reservation_id,
+            )
+            .await?;
+        }
+        return Ok(());
+    }
+    if live_checkout.id != checkout.reservation_id
+        || live_checkout.user_id != checkout.user_id
+        || live_checkout.referral_invite_id != checkout.referral_invite_id
+    {
+        mark_checkout_status(state, &checkout.reservation_id, "invalid").await?;
+        return Err(anyhow!("checkout reservation changed before completion"));
+    }
+    if live_checkout.status != "pending" && live_checkout.status != "expired" {
+        return Err(anyhow!("checkout reservation is {}", live_checkout.status));
+    }
+
+    grant_license(state, &checkout.user_id).await?;
+    mark_checkout_completed(
+        state,
+        &checkout.reservation_id,
+        checkout.paid_amount_pence,
+        &checkout.payment_intent_id,
+    )
+    .await?;
+    if !checkout.referral_invite_id.is_empty() {
+        mark_referral_invite_used(
+            state,
+            &checkout.referral_invite_id,
+            &checkout.user_id,
+            &checkout.reservation_id,
+        )
+        .await?;
+    }
+    Ok(())
+}
+
+pub async fn grant_license_with_pricing_lock(
+    state: &AppState,
+    user_id: &str,
+) -> anyhow::Result<()> {
+    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
+    let pricing_lock = acquire_pocketbase_lock(
+        state,
+        CHECKOUT_PRICING_LOCK_NAME,
+        CHECKOUT_PRICING_LOCK_TTL_SECS,
+    )
+    .await?;
+    let result = grant_license(state, user_id).await;
+    if let Err(err) = pricing_lock.release().await {
+        warn!("Failed to release checkout pricing lock: {err}");
+    }
+    result
+}
+
+pub async fn grant_license(state: &AppState, user_id: &str) -> anyhow::Result<()> {
+    set_user_subscription(state, user_id, "licensed").await
+}
+
+pub async fn reverse_license_for_payment_intent(
+    state: &AppState,
+    payment_intent_id: &str,
+    reason: &str,
+    refunded_amount_pence: Option<u64>,
+) -> anyhow::Result<PaymentReversalOutcome> {
+    if !is_safe_stripe_session_id(payment_intent_id) {
+        return Err(anyhow!("invalid Stripe payment intent id"));
+    }
+    if !is_safe_reversal_reason(reason) {
+        return Err(anyhow!("invalid Stripe reversal reason"));
+    }
+
+    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
+    let checkout = match find_checkout_by_payment_intent_or_checkout_session(
+        state,
+        payment_intent_id,
+    )
+    .await?
+    {
+        Some(checkout) => checkout,
+        None => return Ok(PaymentReversalOutcome::NoMatchingCheckout),
+    };
+
+    let paid_amount_pence = checkout
+        .paid_amount_pence
+        .max(checkout.expected_total_pence);
+    if let Some(refunded_amount_pence) = refunded_amount_pence {
+        if refunded_amount_pence < paid_amount_pence {
+            return Ok(PaymentReversalOutcome::IgnoredPartialRefund {
+                user_id: checkout.user_id,
+                refunded_amount_pence,
+                paid_amount_pence,
+            });
+        }
+    }
+
+    if checkout.status == "reversed" {
+        return Ok(PaymentReversalOutcome::AlreadyHandled {
+            user_id: checkout.user_id,
+        });
+    }
+
+    if matches!(checkout.status.as_str(), "pending" | "expired" | "failed") {
+        mark_checkout_reversed(state, &checkout.id, reason, payment_intent_id).await?;
+        return Ok(PaymentReversalOutcome::Applied {
+            user_id: checkout.user_id,
+        });
+    }
+
+    if checkout.status != "completed" {
+        return Ok(PaymentReversalOutcome::NotReversible {
+            user_id: checkout.user_id,
+            status: checkout.status,
+        });
+    }
+
+    let has_other_license = has_other_completed_checkout_for_user(
+        state,
+        &checkout.user_id,
+        &checkout.id,
+        payment_intent_id,
+    )
+    .await?;
+    if !has_other_license {
+        revoke_license(state, &checkout.user_id).await?;
+    }
+    mark_checkout_reversed(state, &checkout.id, reason, payment_intent_id).await?;
+
+    Ok(PaymentReversalOutcome::Applied {
+        user_id: checkout.user_id,
+    })
+}
+
+pub async fn reinstate_license_for_payment_intent(
+    state: &AppState,
+    payment_intent_id: &str,
+    reason: &str,
+) -> anyhow::Result<PaymentReinstatementOutcome> {
+    if !is_safe_stripe_session_id(payment_intent_id) {
+        return Err(anyhow!("invalid Stripe payment intent id"));
+    }
+    if !is_safe_reversal_reason(reason) {
+        return Err(anyhow!("invalid Stripe reinstatement reason"));
+    }
+
+    let _guard = CHECKOUT_RESERVATION_LOCK.lock().await;
+    let checkout = match find_checkout_by_payment_intent_or_checkout_session(
+        state,
+        payment_intent_id,
+    )
+    .await?
+    {
+        Some(checkout) => checkout,
+        None => return Ok(PaymentReinstatementOutcome::NoMatchingCheckout),
+    };
+
+    if checkout.status == "completed" {
+        return Ok(PaymentReinstatementOutcome::AlreadyHandled {
+            user_id: checkout.user_id,
+        });
+    }
+
+    if checkout.status != "reversed" {
+        return Ok(PaymentReinstatementOutcome::Ignored {
+            user_id: checkout.user_id,
+            reason: format!("checkout status is {}", checkout.status),
+        });
+    }
+
+    if !checkout.reversal_reason.starts_with("charge.dispute.") {
+        return Ok(PaymentReinstatementOutcome::Ignored {
+            user_id: checkout.user_id,
+            reason: format!("checkout was reversed by {}", checkout.reversal_reason),
+        });
+    }
+
+    grant_license(state, &checkout.user_id).await?;
+    mark_checkout_reinstated(state, &checkout.id, reason).await?;
+
+    Ok(PaymentReinstatementOutcome::Applied {
+        user_id: checkout.user_id,
+    })
+}
+
+async fn revoke_license(state: &AppState, user_id: &str) -> anyhow::Result<()> {
+    set_user_subscription(state, user_id, "free").await
+}
+
+async fn set_user_subscription(
+    state: &AppState,
+    user_id: &str,
+    subscription: &str,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/users/records/{user_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({ "subscription": subscription }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase license update failed")?;
+
+    state.token_cache.invalidate_by_user_id(user_id);
+    Ok(())
+}
+
+pub(super) fn expected_total_for_checkout(
+    amount_pence: u64,
+    discount_coupon_id: Option<&str>,
+) -> u64 {
+    if discount_coupon_id.is_some_and(|id| !id.is_empty()) {
+        return ((amount_pence * (100 - REFERRAL_DISCOUNT_PERCENT)) / 100).max(1);
+    }
+    amount_pence
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn expected_total_for_referral_discount_rounds_down_like_stripe_amount_math() {
+        assert_eq!(expected_total_for_checkout(999, Some("coupon_30")), 699);
+        assert_eq!(expected_total_for_checkout(1, Some("coupon_30")), 1);
+        assert_eq!(expected_total_for_checkout(999, None), 999);
+    }
+}
diff --git a/server-rs/src/checkout_sessions/mod.rs b/server-rs/src/checkout_sessions/mod.rs
new file mode 100644
index 0000000..90701e7
--- /dev/null
+++ b/server-rs/src/checkout_sessions/mod.rs
@@ -0,0 +1,133 @@
+//! Checkout sessions: Stripe-backed lifetime-license purchases reserved and
+//! recorded in the PocketBase `checkout_sessions` collection.
+//!
+//! Split by concern:
+//! - [`lifecycle`]: the session state machine (start, verify, complete,
+//!   reverse, reinstate) and license granting
+//! - [`records`]: PocketBase `checkout_sessions` record handling
+//! - [`referral`]: referral invite reservation/consumption bookkeeping
+//! - [`stripe`]: Stripe API interaction (sessions, coupons, lookups)
+
+mod lifecycle;
+mod records;
+mod referral;
+mod stripe;
+#[cfg(test)]
+mod tests;
+
+pub use lifecycle::{
+    complete_verified_checkout, grant_license_with_pricing_lock,
+    reinstate_license_for_payment_intent, reverse_license_for_payment_intent,
+    start_license_checkout, verify_checkout_completion,
+};
+pub use referral::active_referral_checkout_user;
+
+use std::time::{SystemTime, UNIX_EPOCH};
+
+use anyhow::anyhow;
+use serde_json::Value;
+
+pub const CHECKOUT_CURRENCY: &str = "gbp";
+
+const CHECKOUT_COLLECTION: &str = "checkout_sessions";
+const REFERRAL_DISCOUNT_PERCENT: u64 = 30;
+
+pub enum CheckoutStart {
+    Free,
+    Stripe { url: String },
+}
+
+pub enum CheckoutCompletion {
+    Grant(VerifiedCheckout),
+    AlreadyHandled(VerifiedCheckout),
+    Rejected(String),
+}
+
+pub enum PaymentReversalOutcome {
+    Applied {
+        user_id: String,
+    },
+    AlreadyHandled {
+        user_id: String,
+    },
+    IgnoredPartialRefund {
+        user_id: String,
+        refunded_amount_pence: u64,
+        paid_amount_pence: u64,
+    },
+    NoMatchingCheckout,
+    NotReversible {
+        user_id: String,
+        status: String,
+    },
+}
+
+pub enum PaymentReinstatementOutcome {
+    Applied { user_id: String },
+    AlreadyHandled { user_id: String },
+    Ignored { user_id: String, reason: String },
+    NoMatchingCheckout,
+}
+
+pub struct VerifiedCheckout {
+    pub reservation_id: String,
+    pub user_id: String,
+    pub stripe_session_id: String,
+    pub payment_intent_id: String,
+    pub paid_amount_pence: u64,
+    pub referral_invite_id: String,
+}
+
+pub fn now_unix_secs() -> u64 {
+    SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs()
+}
+
+fn number_field(value: &Value, field: &str) -> Option<u64> {
+    value[field].as_u64().or_else(|| {
+        value[field]
+            .as_f64()
+            .filter(|n| n.is_finite() && *n >= 0.0 && n.fract() == 0.0)
+            .map(|n| n as u64)
+    })
+}
+
+fn is_safe_stripe_session_id(id: &str) -> bool {
+    !id.is_empty()
+        && id.len() <= 128
+        && id
+            .bytes()
+            .all(|b| b.is_ascii_alphanumeric() || b == b'_' || b == b'-')
+}
+
+fn is_safe_pocketbase_id(id: &str) -> bool {
+    !id.is_empty() && id.len() <= 32 && id.bytes().all(|b| b.is_ascii_alphanumeric())
+}
+
+fn is_safe_reversal_reason(reason: &str) -> bool {
+    !reason.is_empty()
+        && reason.len() <= 128
+        && reason
+            .bytes()
+            .all(|b| b.is_ascii_alphanumeric() || b == b'_' || b == b'-' || b == b'.')
+}
+
+async fn ensure_success(resp: reqwest::Response) -> anyhow::Result<()> {
+    if resp.status().is_success() {
+        return Ok(());
+    }
+
+    let status = resp.status();
+    let text = resp.text().await.unwrap_or_default();
+    Err(anyhow!("upstream returned {status}: {text}"))
+}
+
+async fn ensure_success_ref(resp: &reqwest::Response) -> anyhow::Result<()> {
+    if resp.status().is_success() {
+        return Ok(());
+    }
+
+    Err(anyhow!("upstream returned {}", resp.status()))
+}
diff --git a/server-rs/src/checkout_sessions/records.rs b/server-rs/src/checkout_sessions/records.rs
new file mode 100644
index 0000000..40edbe0
--- /dev/null
+++ b/server-rs/src/checkout_sessions/records.rs
@@ -0,0 +1,564 @@
+//! PocketBase `checkout_sessions` record handling: creating reservations,
+//! status transitions, and lookups by Stripe session / payment intent.
+
+use anyhow::{anyhow, Context};
+use serde_json::Value;
+use tracing::warn;
+
+use crate::pocketbase::get_superuser_token;
+use crate::state::AppState;
+
+use super::referral::release_referral_invite_reservation;
+use super::stripe::fetch_stripe_checkout_session_id_for_payment_intent;
+use super::{
+    ensure_success, ensure_success_ref, is_safe_pocketbase_id, is_safe_stripe_session_id,
+    now_unix_secs, number_field, CHECKOUT_COLLECTION,
+};
+
+#[derive(Debug)]
+pub(super) struct PendingCheckout {
+    pub(super) id: String,
+    pub(super) user_id: String,
+    pub(super) stripe_session_id: String,
+    pub(super) checkout_url: String,
+    pub(super) amount_pence: u64,
+    pub(super) expected_total_pence: u64,
+    pub(super) currency: String,
+    pub(super) referral_invite_id: String,
+    pub(super) status: String,
+    pub(super) payment_intent_id: String,
+    pub(super) paid_amount_pence: u64,
+    pub(super) reversal_reason: String,
+}
+
+pub async fn mark_checkout_completed(
+    state: &AppState,
+    reservation_id: &str,
+    paid_amount_pence: u64,
+    payment_intent_id: &str,
+) -> anyhow::Result<()> {
+    if !is_safe_stripe_session_id(payment_intent_id) {
+        return Err(anyhow!("invalid Stripe payment intent id"));
+    }
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "status": "completed",
+            "paid_amount_pence": paid_amount_pence,
+            "completed_at_unix": now_unix_secs().to_string(),
+            "stripe_payment_intent_id": payment_intent_id,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase checkout completion update failed")
+}
+
+pub(super) async fn count_active_pending_checkouts(
+    state: &AppState,
+    now: u64,
+) -> anyhow::Result<u64> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = format!("status=\"pending\" && expires_at_unix>={now}");
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    Ok(body["totalItems"].as_u64().unwrap_or(0))
+}
+
+pub(super) async fn find_active_checkout_for_user(
+    state: &AppState,
+    user_id: &str,
+    discount_coupon_id: &str,
+    referral_invite_id: &str,
+    now: u64,
+) -> anyhow::Result<Option<PendingCheckout>> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = active_checkout_filter(user_id, discount_coupon_id, referral_invite_id, now)?;
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    let item = body["items"]
+        .as_array()
+        .and_then(|items| items.first())
+        .cloned();
+
+    item.map(parse_pending_checkout).transpose()
+}
+
+fn active_checkout_filter(
+    user_id: &str,
+    discount_coupon_id: &str,
+    referral_invite_id: &str,
+    now: u64,
+) -> anyhow::Result<String> {
+    if !is_safe_pocketbase_id(user_id) {
+        return Err(anyhow!("invalid PocketBase user id"));
+    }
+    if !discount_coupon_id.is_empty() && !is_safe_stripe_session_id(discount_coupon_id) {
+        return Err(anyhow!("invalid Stripe coupon id"));
+    }
+    if !referral_invite_id.is_empty() && !is_safe_pocketbase_id(referral_invite_id) {
+        return Err(anyhow!("invalid PocketBase referral invite id"));
+    }
+
+    Ok(format!(
+        "status=\"pending\" && expires_at_unix>={now} && user=\"{user_id}\" && discount_coupon_id=\"{discount_coupon_id}\" && referral_invite_id=\"{referral_invite_id}\""
+    ))
+}
+
+pub(super) async fn expire_stale_pending_checkouts(
+    state: &AppState,
+    now: u64,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = format!("status=\"pending\" && expires_at_unix<{now}");
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=50",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    let Some(items) = body["items"].as_array() else {
+        return Ok(());
+    };
+
+    for item in items {
+        let Some(id) = item["id"].as_str() else {
+            continue;
+        };
+        if let Err(err) = mark_checkout_status(state, id, "expired").await {
+            warn!(
+                reservation_id = id,
+                "Failed to expire checkout reservation: {err}"
+            );
+        }
+        if let Some(invite_id) = item["referral_invite_id"]
+            .as_str()
+            .filter(|invite_id| !invite_id.is_empty())
+        {
+            if let Err(err) = release_referral_invite_reservation(state, invite_id, id).await {
+                warn!(
+                    reservation_id = id,
+                    referral_invite_id = invite_id,
+                    "Failed to release expired referral invite reservation: {err}"
+                );
+            }
+        }
+    }
+
+    Ok(())
+}
+
+pub(super) struct PendingCheckoutInput<'a> {
+    pub(super) user_id: &'a str,
+    pub(super) amount_pence: u64,
+    pub(super) expected_total_pence: u64,
+    pub(super) currency: &'a str,
+    pub(super) discount_coupon_id: &'a str,
+    pub(super) referral_invite_id: &'a str,
+    pub(super) expires_at_unix: u64,
+}
+
+pub(super) async fn create_pending_checkout(
+    state: &AppState,
+    input: PendingCheckoutInput<'_>,
+) -> anyhow::Result<String> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records");
+    let resp = state
+        .http_client
+        .post(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "user": input.user_id,
+            "stripe_session_id": "",
+            "stripe_payment_intent_id": "",
+            "checkout_url": "",
+            "amount_pence": input.amount_pence,
+            "expected_total_pence": input.expected_total_pence,
+            "currency": input.currency,
+            "discount_coupon_id": input.discount_coupon_id,
+            "referral_invite_id": input.referral_invite_id,
+            "status": "pending",
+            "expires_at_unix": input.expires_at_unix,
+            "paid_amount_pence": 0,
+            "completed_at_unix": "",
+            "reversal_reason": "",
+        }))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    body["id"]
+        .as_str()
+        .map(str::to_string)
+        .ok_or_else(|| anyhow!("PocketBase checkout reservation missing id"))
+}
+
+pub(super) async fn attach_stripe_session(
+    state: &AppState,
+    reservation_id: &str,
+    stripe_session_id: &str,
+    checkout_url: &str,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "stripe_session_id": stripe_session_id,
+            "checkout_url": checkout_url,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase checkout session attach failed")
+}
+
+pub(super) async fn mark_checkout_status(
+    state: &AppState,
+    reservation_id: &str,
+    status: &str,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({ "status": status }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .with_context(|| format!("PocketBase checkout status update failed for {reservation_id}"))
+}
+
+pub(super) async fn mark_checkout_reversed(
+    state: &AppState,
+    reservation_id: &str,
+    reason: &str,
+    payment_intent_id: &str,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "status": "reversed",
+            "reversal_reason": reason,
+            "stripe_payment_intent_id": payment_intent_id,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .with_context(|| format!("PocketBase checkout reversal update failed for {reservation_id}"))
+}
+
+pub(super) async fn mark_checkout_reinstated(
+    state: &AppState,
+    reservation_id: &str,
+    _reason: &str,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "status": "completed",
+            "reversal_reason": "",
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp).await.with_context(|| {
+        format!("PocketBase checkout reinstatement update failed for {reservation_id}")
+    })
+}
+
+pub(super) async fn find_checkout_by_stripe_session(
+    state: &AppState,
+    stripe_session_id: &str,
+) -> anyhow::Result<Option<PendingCheckout>> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = format!("stripe_session_id=\"{}\"", stripe_session_id);
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    let item = body["items"]
+        .as_array()
+        .and_then(|items| items.first())
+        .cloned();
+
+    item.map(parse_pending_checkout).transpose()
+}
+
+async fn find_checkout_by_payment_intent(
+    state: &AppState,
+    payment_intent_id: &str,
+) -> anyhow::Result<Option<PendingCheckout>> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = format!("stripe_payment_intent_id=\"{}\"", payment_intent_id);
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    let item = body["items"]
+        .as_array()
+        .and_then(|items| items.first())
+        .cloned();
+
+    item.map(parse_pending_checkout).transpose()
+}
+
+pub(super) async fn find_checkout_by_payment_intent_or_checkout_session(
+    state: &AppState,
+    payment_intent_id: &str,
+) -> anyhow::Result<Option<PendingCheckout>> {
+    if let Some(checkout) = find_checkout_by_payment_intent(state, payment_intent_id).await? {
+        return Ok(Some(checkout));
+    }
+
+    let Some(session_id) =
+        fetch_stripe_checkout_session_id_for_payment_intent(state, payment_intent_id).await?
+    else {
+        return Ok(None);
+    };
+
+    let Some(mut checkout) = find_checkout_by_stripe_session(state, &session_id).await? else {
+        return Ok(None);
+    };
+
+    if checkout.payment_intent_id.is_empty() {
+        attach_payment_intent_to_checkout(state, &checkout.id, payment_intent_id).await?;
+        checkout.payment_intent_id = payment_intent_id.to_string();
+    } else if checkout.payment_intent_id != payment_intent_id {
+        mark_checkout_status(state, &checkout.id, "invalid").await?;
+        return Err(anyhow!(
+            "checkout reservation payment intent changed before reversal"
+        ));
+    }
+
+    Ok(Some(checkout))
+}
+
+async fn attach_payment_intent_to_checkout(
+    state: &AppState,
+    reservation_id: &str,
+    payment_intent_id: &str,
+) -> anyhow::Result<()> {
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records/{reservation_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "stripe_payment_intent_id": payment_intent_id,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase checkout payment intent attach failed")
+}
+
+pub(super) async fn has_other_completed_checkout_for_user(
+    state: &AppState,
+    user_id: &str,
+    reservation_id: &str,
+    payment_intent_id: &str,
+) -> anyhow::Result<bool> {
+    if !is_safe_pocketbase_id(user_id) || !is_safe_pocketbase_id(reservation_id) {
+        return Err(anyhow!("invalid PocketBase id"));
+    }
+    if !is_safe_stripe_session_id(payment_intent_id) {
+        return Err(anyhow!("invalid Stripe payment intent id"));
+    }
+
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = format!("user=\"{user_id}\" && status=\"completed\"");
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=50",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    let Some(items) = body["items"].as_array() else {
+        return Ok(false);
+    };
+
+    Ok(items.iter().any(|item| {
+        let other_id = item["id"].as_str().unwrap_or_default();
+        let other_payment_intent = item["stripe_payment_intent_id"]
+            .as_str()
+            .unwrap_or_default();
+        other_id != reservation_id && other_payment_intent != payment_intent_id
+    }))
+}
+
+fn parse_pending_checkout(item: Value) -> anyhow::Result<PendingCheckout> {
+    Ok(PendingCheckout {
+        id: item["id"]
+            .as_str()
+            .ok_or_else(|| anyhow!("checkout reservation missing id"))?
+            .to_string(),
+        user_id: item["user"]
+            .as_str()
+            .ok_or_else(|| anyhow!("checkout reservation missing user"))?
+            .to_string(),
+        stripe_session_id: item["stripe_session_id"]
+            .as_str()
+            .unwrap_or_default()
+            .to_string(),
+        checkout_url: item["checkout_url"]
+            .as_str()
+            .unwrap_or_default()
+            .to_string(),
+        amount_pence: number_field(&item, "amount_pence")
+            .ok_or_else(|| anyhow!("checkout reservation missing amount_pence"))?,
+        expected_total_pence: number_field(&item, "expected_total_pence")
+            .ok_or_else(|| anyhow!("checkout reservation missing expected_total_pence"))?,
+        currency: item["currency"]
+            .as_str()
+            .unwrap_or_default()
+            .to_ascii_lowercase(),
+        referral_invite_id: item["referral_invite_id"]
+            .as_str()
+            .unwrap_or_default()
+            .to_string(),
+        status: item["status"].as_str().unwrap_or_default().to_string(),
+        payment_intent_id: item["stripe_payment_intent_id"]
+            .as_str()
+            .unwrap_or_default()
+            .to_string(),
+        paid_amount_pence: number_field(&item, "paid_amount_pence").unwrap_or(0),
+        reversal_reason: item["reversal_reason"]
+            .as_str()
+            .unwrap_or_default()
+            .to_string(),
+    })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn active_checkout_filter_includes_empty_context_for_standard_checkout() {
+        let filter = active_checkout_filter("abc123", "", "", 42).unwrap();
+        assert_eq!(
+            filter,
+            "status=\"pending\" && expires_at_unix>=42 && user=\"abc123\" && discount_coupon_id=\"\" && referral_invite_id=\"\""
+        );
+    }
+
+    #[test]
+    fn active_checkout_filter_includes_referral_context() {
+        let filter = active_checkout_filter("user123", "coupon_30", "invite123", 99).unwrap();
+        assert_eq!(
+            filter,
+            "status=\"pending\" && expires_at_unix>=99 && user=\"user123\" && discount_coupon_id=\"coupon_30\" && referral_invite_id=\"invite123\""
+        );
+    }
+
+    #[test]
+    fn active_checkout_filter_rejects_unsafe_context_values() {
+        assert!(active_checkout_filter("user123", "bad\"coupon", "", 1).is_err());
+        assert!(active_checkout_filter("user123", "", "bad-invite", 1).is_err());
+        assert!(active_checkout_filter("bad-user", "", "", 1).is_err());
+    }
+}
diff --git a/server-rs/src/checkout_sessions/referral.rs b/server-rs/src/checkout_sessions/referral.rs
new file mode 100644
index 0000000..d503ea1
--- /dev/null
+++ b/server-rs/src/checkout_sessions/referral.rs
@@ -0,0 +1,312 @@
+//! Referral invite bookkeeping: reserving an invite for an in-flight checkout,
+//! releasing the reservation on failure/expiry, and recording final usage when
+//! a verified payment completes.
+
+use anyhow::{anyhow, Context};
+use serde_json::Value;
+use tracing::warn;
+
+use crate::pocketbase::get_superuser_token;
+use crate::state::AppState;
+
+use super::{
+    ensure_success, ensure_success_ref, is_safe_pocketbase_id, now_unix_secs, number_field,
+    CHECKOUT_COLLECTION,
+};
+
+pub async fn mark_referral_invite_used(
+    state: &AppState,
+    invite_id: &str,
+    user_id: &str,
+    reservation_id: &str,
+) -> anyhow::Result<()> {
+    if invite_id.is_empty() {
+        return Ok(());
+    }
+    if !is_safe_pocketbase_id(invite_id)
+        || !is_safe_pocketbase_id(user_id)
+        || !is_safe_pocketbase_id(reservation_id)
+    {
+        return Err(anyhow!("invalid PocketBase id"));
+    }
+
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let invite = fetch_invite_record(state, pb_url, &token, invite_id).await?;
+
+    // A verified Stripe payment must not lose entitlement just because local
+    // invite reservation bookkeeping expired or moved before webhook delivery.
+    match referral_invite_completion_action(&invite, user_id, reservation_id) {
+        ReferralInviteCompletionAction::AlreadyRecorded => return Ok(()),
+        ReferralInviteCompletionAction::AlreadyUsedByAnother => {
+            warn!(
+                invite_id,
+                user_id,
+                existing_used_by = invite["used_by_id"].as_str().unwrap_or_default(),
+                "Referral invite was already used by another account; preserving verified checkout entitlement"
+            );
+            return Ok(());
+        }
+        ReferralInviteCompletionAction::Record {
+            reservation_reassigned,
+        } => {
+            if reservation_reassigned {
+                warn!(
+                    invite_id,
+                    user_id,
+                    reservation_id,
+                    reserved_by_id = invite["reserved_by_id"].as_str().unwrap_or_default(),
+                    reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default(),
+                    "Referral invite reservation moved before webhook completion; verified checkout will consume it"
+                );
+            }
+        }
+    }
+
+    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "used_by_id": user_id,
+            "used_at": now_unix_secs().to_string(),
+            "reserved_by_id": "",
+            "reserved_checkout_id": "",
+            "reserved_until_unix": 0,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase invite usage update failed")
+}
+
+#[derive(Debug, PartialEq, Eq)]
+enum ReferralInviteCompletionAction {
+    AlreadyRecorded,
+    AlreadyUsedByAnother,
+    Record { reservation_reassigned: bool },
+}
+
+fn referral_invite_completion_action(
+    invite: &Value,
+    user_id: &str,
+    reservation_id: &str,
+) -> ReferralInviteCompletionAction {
+    let existing_used_by = invite["used_by_id"].as_str().unwrap_or_default();
+    if existing_used_by == user_id {
+        return ReferralInviteCompletionAction::AlreadyRecorded;
+    }
+    if !existing_used_by.is_empty() {
+        return ReferralInviteCompletionAction::AlreadyUsedByAnother;
+    }
+
+    let reserved_by_id = invite["reserved_by_id"].as_str().unwrap_or_default();
+    let reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default();
+    let reservation_reassigned = (!reserved_by_id.is_empty() && reserved_by_id != user_id)
+        || (!reserved_checkout_id.is_empty() && reserved_checkout_id != reservation_id);
+
+    ReferralInviteCompletionAction::Record {
+        reservation_reassigned,
+    }
+}
+
+async fn fetch_invite_record(
+    state: &AppState,
+    pb_url: &str,
+    token: &str,
+    invite_id: &str,
+) -> anyhow::Result<Value> {
+    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    resp.json().await.map_err(Into::into)
+}
+
+pub(super) async fn reserve_referral_invite(
+    state: &AppState,
+    invite_id: &str,
+    user_id: &str,
+    reservation_id: &str,
+    reserved_until_unix: u64,
+) -> anyhow::Result<()> {
+    if !is_safe_pocketbase_id(invite_id)
+        || !is_safe_pocketbase_id(user_id)
+        || !is_safe_pocketbase_id(reservation_id)
+    {
+        return Err(anyhow!("invalid PocketBase id"));
+    }
+
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let invite = fetch_invite_record(state, pb_url, &token, invite_id).await?;
+    let used_by = invite["used_by_id"].as_str().unwrap_or_default();
+    if !used_by.is_empty() {
+        return Err(anyhow!("referral invite already used"));
+    }
+
+    let now = now_unix_secs();
+    let reserved_by_id = invite["reserved_by_id"].as_str().unwrap_or_default();
+    let reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default();
+    let existing_reserved_until = number_field(&invite, "reserved_until_unix").unwrap_or(0);
+    let reservation_is_live = existing_reserved_until >= now;
+    if reservation_is_live
+        && !reserved_checkout_id.is_empty()
+        && reserved_checkout_id != reservation_id
+    {
+        return Err(anyhow!("referral invite already has an active checkout"));
+    }
+    if reservation_is_live && !reserved_by_id.is_empty() && reserved_by_id != user_id {
+        return Err(anyhow!("referral invite reserved by another account"));
+    }
+
+    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "reserved_by_id": user_id,
+            "reserved_checkout_id": reservation_id,
+            "reserved_until_unix": reserved_until_unix,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase invite reservation update failed")
+}
+
+pub(super) async fn release_referral_invite_reservation(
+    state: &AppState,
+    invite_id: &str,
+    reservation_id: &str,
+) -> anyhow::Result<()> {
+    if !is_safe_pocketbase_id(invite_id) || !is_safe_pocketbase_id(reservation_id) {
+        return Err(anyhow!("invalid PocketBase id"));
+    }
+
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let invite = fetch_invite_record(state, pb_url, &token, invite_id).await?;
+    let used_by = invite["used_by_id"].as_str().unwrap_or_default();
+    let reserved_checkout_id = invite["reserved_checkout_id"].as_str().unwrap_or_default();
+    if !used_by.is_empty() || reserved_checkout_id != reservation_id {
+        return Ok(());
+    }
+
+    let url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
+    let resp = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&serde_json::json!({
+            "reserved_by_id": "",
+            "reserved_checkout_id": "",
+            "reserved_until_unix": 0,
+        }))
+        .send()
+        .await?;
+
+    ensure_success(resp)
+        .await
+        .context("PocketBase invite reservation release failed")
+}
+
+pub async fn active_referral_checkout_user(
+    state: &AppState,
+    invite_id: &str,
+) -> anyhow::Result<Option<String>> {
+    if !is_safe_pocketbase_id(invite_id) {
+        return Err(anyhow!("invalid PocketBase invite id"));
+    }
+
+    let now = now_unix_secs();
+    let token = get_superuser_token(state).await?;
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let filter = format!(
+        "status=\"pending\" && expires_at_unix>={now} && referral_invite_id=\"{}\"",
+        invite_id
+    );
+    let url = format!(
+        "{pb_url}/api/collections/{CHECKOUT_COLLECTION}/records?filter={}&perPage=1",
+        urlencoding::encode(&filter)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await?;
+
+    ensure_success_ref(&resp).await?;
+
+    let body: Value = resp.json().await?;
+    Ok(body["items"]
+        .as_array()
+        .and_then(|items| items.first())
+        .and_then(|item| item["user"].as_str())
+        .map(str::to_string))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn referral_invite_completion_records_available_invite() {
+        let invite = serde_json::json!({
+            "used_by_id": "",
+            "reserved_by_id": "",
+            "reserved_checkout_id": "",
+        });
+
+        assert_eq!(
+            referral_invite_completion_action(&invite, "user123", "checkout123"),
+            ReferralInviteCompletionAction::Record {
+                reservation_reassigned: false
+            }
+        );
+    }
+
+    #[test]
+    fn referral_invite_completion_records_reassigned_reservation() {
+        let invite = serde_json::json!({
+            "used_by_id": "",
+            "reserved_by_id": "otheruser",
+            "reserved_checkout_id": "othercheckout",
+        });
+
+        assert_eq!(
+            referral_invite_completion_action(&invite, "user123", "checkout123"),
+            ReferralInviteCompletionAction::Record {
+                reservation_reassigned: true
+            }
+        );
+    }
+
+    #[test]
+    fn referral_invite_completion_detects_existing_usage() {
+        let used_by_same_user = serde_json::json!({ "used_by_id": "user123" });
+        let used_by_another_user = serde_json::json!({ "used_by_id": "otheruser" });
+
+        assert_eq!(
+            referral_invite_completion_action(&used_by_same_user, "user123", "checkout123"),
+            ReferralInviteCompletionAction::AlreadyRecorded
+        );
+        assert_eq!(
+            referral_invite_completion_action(&used_by_another_user, "user123", "checkout123"),
+            ReferralInviteCompletionAction::AlreadyUsedByAnother
+        );
+    }
+}
diff --git a/server-rs/src/checkout_sessions/stripe.rs b/server-rs/src/checkout_sessions/stripe.rs
new file mode 100644
index 0000000..9546b35
--- /dev/null
+++ b/server-rs/src/checkout_sessions/stripe.rs
@@ -0,0 +1,175 @@
+//! Stripe API interaction: creating checkout sessions, verifying coupon
+//! configuration, and looking up sessions by payment intent.
+
+use anyhow::{anyhow, Context};
+use serde_json::Value;
+
+use crate::auth::PocketBaseUser;
+use crate::state::AppState;
+
+use super::lifecycle::expected_total_for_checkout;
+use super::{
+    ensure_success_ref, is_safe_stripe_session_id, CHECKOUT_CURRENCY, REFERRAL_DISCOUNT_PERCENT,
+};
+
+const CHECKOUT_PRODUCT_NAME: &str = "Perfect Postcodes Lifetime License";
+
+/// Fetch a Stripe coupon and ensure its `percent_off` matches the expected
+/// referral discount AND that it has no `amount_off` override. This blocks a
+/// misconfigured (or maliciously swapped) coupon ID from quietly granting a
+/// larger discount than the server's pricing math assumed.
+async fn verify_stripe_coupon_discount(state: &AppState, coupon_id: &str) -> anyhow::Result<()> {
+    if !is_safe_stripe_session_id(coupon_id) {
+        return Err(anyhow!("unsafe stripe coupon id"));
+    }
+    let url = format!(
+        "https://api.stripe.com/v1/coupons/{}",
+        urlencoding::encode(coupon_id)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .basic_auth(&state.stripe_secret_key, None::<&str>)
+        .send()
+        .await
+        .context("Stripe coupon fetch failed")?;
+    ensure_success_ref(&resp)
+        .await
+        .context("Stripe coupon fetch returned error")?;
+    let body: Value = resp
+        .json()
+        .await
+        .context("Failed to parse Stripe coupon response")?;
+
+    let valid = body["valid"].as_bool().unwrap_or(false);
+    if !valid {
+        return Err(anyhow!("stripe coupon is not valid"));
+    }
+    if body["amount_off"].is_number() {
+        return Err(anyhow!(
+            "stripe coupon uses amount_off; only percent_off is permitted"
+        ));
+    }
+    let percent_off = body["percent_off"]
+        .as_f64()
+        .ok_or_else(|| anyhow!("stripe coupon missing percent_off"))?;
+    if percent_off.is_nan() || (percent_off - REFERRAL_DISCOUNT_PERCENT as f64).abs() > 0.001 {
+        return Err(anyhow!(
+            "stripe coupon percent_off ({percent_off}) does not match expected {REFERRAL_DISCOUNT_PERCENT}"
+        ));
+    }
+    Ok(())
+}
+
+#[allow(clippy::too_many_arguments)]
+pub(super) async fn create_stripe_session(
+    state: &AppState,
+    user: &PocketBaseUser,
+    reservation_id: &str,
+    price_pence: u64,
+    success_url: &str,
+    cancel_url: &str,
+    expires_at_unix: u64,
+    discount_coupon_id: Option<&str>,
+) -> anyhow::Result<(String, String)> {
+    if let Some(coupon_id) = discount_coupon_id.filter(|id| !id.is_empty()) {
+        verify_stripe_coupon_discount(state, coupon_id).await?;
+    }
+
+    let mut form_params = vec![
+        ("mode", "payment".to_string()),
+        ("payment_method_types[0]", "card".to_string()),
+        (
+            "line_items[0][price_data][unit_amount]",
+            price_pence.to_string(),
+        ),
+        (
+            "line_items[0][price_data][currency]",
+            CHECKOUT_CURRENCY.to_string(),
+        ),
+        (
+            "line_items[0][price_data][product_data][name]",
+            CHECKOUT_PRODUCT_NAME.to_string(),
+        ),
+        ("line_items[0][quantity]", "1".to_string()),
+        ("success_url", success_url.to_string()),
+        ("cancel_url", cancel_url.to_string()),
+        ("expires_at", expires_at_unix.to_string()),
+        ("client_reference_id", user.id.clone()),
+        ("customer_email", user.email.clone()),
+        ("metadata[pending_checkout_id]", reservation_id.to_string()),
+        ("metadata[expected_amount_pence]", price_pence.to_string()),
+        (
+            "metadata[expected_total_pence]",
+            expected_total_for_checkout(price_pence, discount_coupon_id).to_string(),
+        ),
+        ("metadata[expected_currency]", CHECKOUT_CURRENCY.to_string()),
+    ];
+
+    if let Some(coupon_id) = discount_coupon_id.filter(|id| !id.is_empty()) {
+        form_params.push(("discounts[0][coupon]", coupon_id.to_string()));
+        form_params.push(("metadata[discount_coupon_id]", coupon_id.to_string()));
+    }
+
+    let resp = state
+        .http_client
+        .post("https://api.stripe.com/v1/checkout/sessions")
+        .basic_auth(&state.stripe_secret_key, None::<&str>)
+        .form(&form_params)
+        .send()
+        .await
+        .context("Stripe checkout request failed")?;
+
+    ensure_success_ref(&resp)
+        .await
+        .context("Stripe checkout failed")?;
+
+    let body: Value = resp
+        .json()
+        .await
+        .context("Failed to parse Stripe response")?;
+    let session_id = body["id"]
+        .as_str()
+        .filter(|id| is_safe_stripe_session_id(id))
+        .map(str::to_string)
+        .ok_or_else(|| anyhow!("Stripe session missing valid id"))?;
+    let url = body["url"]
+        .as_str()
+        .map(str::to_string)
+        .filter(|url| !url.is_empty())
+        .ok_or_else(|| anyhow!("Stripe session missing URL"))?;
+
+    Ok((session_id, url))
+}
+
+pub(super) async fn fetch_stripe_checkout_session_id_for_payment_intent(
+    state: &AppState,
+    payment_intent_id: &str,
+) -> anyhow::Result<Option<String>> {
+    let url = format!(
+        "https://api.stripe.com/v1/checkout/sessions?payment_intent={}&limit=1",
+        urlencoding::encode(payment_intent_id)
+    );
+    let resp = state
+        .http_client
+        .get(&url)
+        .basic_auth(&state.stripe_secret_key, None::<&str>)
+        .send()
+        .await
+        .context("Stripe checkout session lookup failed")?;
+
+    ensure_success_ref(&resp)
+        .await
+        .context("Stripe checkout session lookup returned error")?;
+
+    let body: Value = resp
+        .json()
+        .await
+        .context("Failed to parse Stripe checkout session lookup")?;
+    Ok(body["data"]
+        .as_array()
+        .and_then(|items| items.first())
+        .and_then(|item| item["id"].as_str())
+        .filter(|id| is_safe_stripe_session_id(id))
+        .map(str::to_string))
+}
diff --git a/server-rs/src/checkout_sessions/tests.rs b/server-rs/src/checkout_sessions/tests.rs
new file mode 100644
index 0000000..fa76662
--- /dev/null
+++ b/server-rs/src/checkout_sessions/tests.rs
@@ -0,0 +1,688 @@
+//! Integration-style tests for the money paths: Stripe webhook verification →
+//! license granting, checkout reservation bookkeeping, and invite redemption.
+//!
+//! PocketBase (an external HTTP service in production) is replaced by an
+//! in-process axum mock listening on an ephemeral local port. The mock keeps
+//! records in memory, evaluates the small PocketBase filter subset the server
+//! actually uses, and records every mutating request so tests can assert that
+//! e.g. a replayed webhook does not grant a license twice.
+//!
+//! Stripe itself is NOT mocked (its API URL is hardcoded to
+//! `https://api.stripe.com`), so tests that reach the Stripe call assert the
+//! failure-cleanup behaviour instead: the reservation is marked `failed` and
+//! referral invite reservations are released.
+
+use std::collections::HashMap;
+use std::sync::atomic::{AtomicUsize, Ordering};
+use std::sync::{Arc, Mutex};
+use std::time::{SystemTime, UNIX_EPOCH};
+
+use axum::body::Bytes;
+use axum::extract::{Path, Query, State};
+use axum::http::{HeaderMap, HeaderValue, StatusCode};
+use axum::response::{IntoResponse, Response};
+use axum::routing::{get, post};
+use axum::{Extension, Json, Router};
+use hmac::{Hmac, KeyInit, Mac};
+use serde_json::{json, Value};
+use sha2::Sha256;
+
+use crate::auth::{OptionalUser, PocketBaseUser};
+use crate::routes::{post_redeem_invite, post_stripe_webhook};
+use crate::state::{AppState, SharedState};
+
+use super::start_license_checkout;
+
+// ---------------------------------------------------------------------------
+// Mock PocketBase
+// ---------------------------------------------------------------------------
+
+#[derive(Default)]
+struct MockPocketBase {
+    /// collection name → records (each a JSON object with an "id").
+    records: Mutex<HashMap<String, Vec<Value>>>,
+    next_id: AtomicUsize,
+    /// Every mutating request: (method, path, body).
+    log: Mutex<Vec<(String, String, Value)>>,
+}
+
+impl MockPocketBase {
+    fn seed(&self, collection: &str, mut record: Value) -> String {
+        let id = match record["id"].as_str() {
+            Some(id) if !id.is_empty() => id.to_string(),
+            _ => format!("mockid{:06}", self.next_id.fetch_add(1, Ordering::SeqCst)),
+        };
+        record["id"] = json!(id);
+        self.records
+            .lock()
+            .unwrap()
+            .entry(collection.to_string())
+            .or_default()
+            .push(record);
+        id
+    }
+
+    fn record(&self, collection: &str, id: &str) -> Option<Value> {
+        self.records
+            .lock()
+            .unwrap()
+            .get(collection)?
+            .iter()
+            .find(|record| record["id"].as_str() == Some(id))
+            .cloned()
+    }
+
+    fn records_in(&self, collection: &str) -> Vec<Value> {
+        self.records
+            .lock()
+            .unwrap()
+            .get(collection)
+            .cloned()
+            .unwrap_or_default()
+    }
+
+    /// Number of PATCHes that set the user's subscription to "licensed" —
+    /// i.e. how many times a license was granted.
+    fn license_grant_count(&self, user_id: &str) -> usize {
+        let path = format!("/api/collections/users/records/{user_id}");
+        self.log
+            .lock()
+            .unwrap()
+            .iter()
+            .filter(|(method, request_path, body)| {
+                method == "PATCH"
+                    && request_path == &path
+                    && body["subscription"].as_str() == Some("licensed")
+            })
+            .count()
+    }
+}
+
+/// Evaluate the PocketBase filter subset used by the server:
+/// `a="x" && b>=N && c<N && (d="" || d="y")`.
+fn record_matches(record: &Value, filter: &str) -> bool {
+    if filter.is_empty() {
+        return true;
+    }
+    filter
+        .split(" && ")
+        .all(|clause| clause_matches(record, clause.trim()))
+}
+
+fn clause_matches(record: &Value, clause: &str) -> bool {
+    if let Some(inner) = clause.strip_prefix('(').and_then(|c| c.strip_suffix(')')) {
+        return inner
+            .split(" || ")
+            .any(|alternative| clause_matches(record, alternative.trim()));
+    }
+    if let Some((field, value)) = clause.split_once(">=") {
+        return record_number(record, field) >= value.trim().parse::<i64>().unwrap_or(i64::MAX);
+    }
+    if let Some((field, value)) = clause.split_once('<') {
+        return record_number(record, field) < value.trim().parse::<i64>().unwrap_or(i64::MIN);
+    }
+    if let Some((field, value)) = clause.split_once('=') {
+        let expected = value.trim().trim_matches('"');
+        return record_string(record, field) == expected;
+    }
+    panic!("mock PocketBase cannot evaluate filter clause: {clause}");
+}
+
+fn record_number(record: &Value, field: &str) -> i64 {
+    let value = &record[field];
+    value
+        .as_i64()
+        .or_else(|| value.as_f64().map(|float| float as i64))
+        .or_else(|| value.as_str().and_then(|text| text.parse().ok()))
+        .unwrap_or(0)
+}
+
+fn record_string(record: &Value, field: &str) -> String {
+    let value = &record[field];
+    value.as_str().map(str::to_string).unwrap_or_else(|| {
+        if value.is_null() {
+            String::new()
+        } else {
+            value.to_string()
+        }
+    })
+}
+
+async fn auth_handler() -> Json<Value> {
+    Json(json!({ "token": "testsuperusertoken" }))
+}
+
+async fn list_records(
+    State(pb): State<Arc<MockPocketBase>>,
+    Path(collection): Path<String>,
+    Query(params): Query<HashMap<String, String>>,
+) -> Json<Value> {
+    let filter = params.get("filter").map(String::as_str).unwrap_or("");
+    let matching: Vec<Value> = pb
+        .records
+        .lock()
+        .unwrap()
+        .get(&collection)
+        .map(|records| {
+            records
+                .iter()
+                .filter(|record| record_matches(record, filter))
+                .cloned()
+                .collect()
+        })
+        .unwrap_or_default();
+    let total = matching.len();
+    let per_page = params
+        .get("perPage")
+        .and_then(|raw| raw.parse::<usize>().ok())
+        .unwrap_or(30);
+    let items: Vec<Value> = matching.into_iter().take(per_page).collect();
+    Json(json!({ "items": items, "totalItems": total }))
+}
+
+async fn create_record(
+    State(pb): State<Arc<MockPocketBase>>,
+    Path(collection): Path<String>,
+    Json(body): Json<Value>,
+) -> Response {
+    pb.log.lock().unwrap().push((
+        "POST".to_string(),
+        format!("/api/collections/{collection}/records"),
+        body.clone(),
+    ));
+
+    // Emulate the unique `name` constraint on the distributed-lock collection
+    // so concurrent acquisitions conflict like they do against real PocketBase.
+    if collection == "checkout_locks" {
+        let exists = pb
+            .records
+            .lock()
+            .unwrap()
+            .get(&collection)
+            .is_some_and(|records| records.iter().any(|record| record["name"] == body["name"]));
+        if exists {
+            return (
+                StatusCode::BAD_REQUEST,
+                Json(json!({ "message": "name must be unique" })),
+            )
+                .into_response();
+        }
+    }
+
+    let id = pb.seed(&collection, body);
+    Json(pb.record(&collection, &id).expect("record just created")).into_response()
+}
+
+async fn get_record(
+    State(pb): State<Arc<MockPocketBase>>,
+    Path((collection, id)): Path<(String, String)>,
+) -> Response {
+    match pb.record(&collection, &id) {
+        Some(record) => Json(record).into_response(),
+        None => StatusCode::NOT_FOUND.into_response(),
+    }
+}
+
+async fn patch_record(
+    State(pb): State<Arc<MockPocketBase>>,
+    Path((collection, id)): Path<(String, String)>,
+    Json(body): Json<Value>,
+) -> Response {
+    pb.log.lock().unwrap().push((
+        "PATCH".to_string(),
+        format!("/api/collections/{collection}/records/{id}"),
+        body.clone(),
+    ));
+
+    let mut records = pb.records.lock().unwrap();
+    let Some(record) = records.get_mut(&collection).and_then(|list| {
+        list.iter_mut()
+            .find(|record| record["id"].as_str() == Some(&id))
+    }) else {
+        return StatusCode::NOT_FOUND.into_response();
+    };
+    if let (Some(target), Some(updates)) = (record.as_object_mut(), body.as_object()) {
+        for (key, value) in updates {
+            target.insert(key.clone(), value.clone());
+        }
+    }
+    Json(record.clone()).into_response()
+}
+
+async fn delete_record(
+    State(pb): State<Arc<MockPocketBase>>,
+    Path((collection, id)): Path<(String, String)>,
+) -> Response {
+    pb.log.lock().unwrap().push((
+        "DELETE".to_string(),
+        format!("/api/collections/{collection}/records/{id}"),
+        Value::Null,
+    ));
+
+    let mut records = pb.records.lock().unwrap();
+    let Some(list) = records.get_mut(&collection) else {
+        return StatusCode::NOT_FOUND.into_response();
+    };
+    let before = list.len();
+    list.retain(|record| record["id"].as_str() != Some(&id));
+    if list.len() == before {
+        StatusCode::NOT_FOUND.into_response()
+    } else {
+        StatusCode::NO_CONTENT.into_response()
+    }
+}
+
+fn mock_pb_router(pb: Arc<MockPocketBase>) -> Router {
+    Router::new()
+        .route(
+            "/api/collections/_superusers/auth-with-password",
+            post(auth_handler),
+        )
+        .route(
+            "/api/collections/{collection}/records",
+            get(list_records).post(create_record),
+        )
+        .route(
+            "/api/collections/{collection}/records/{id}",
+            get(get_record).patch(patch_record).delete(delete_record),
+        )
+        .with_state(pb)
+}
+
+// ---------------------------------------------------------------------------
+// Test harness
+// ---------------------------------------------------------------------------
+
+struct TestEnv {
+    shared: Arc<SharedState>,
+    pb: Arc<MockPocketBase>,
+}
+
+async fn setup() -> TestEnv {
+    let pb = Arc::new(MockPocketBase::default());
+    let listener = tokio::net::TcpListener::bind("127.0.0.1:0")
+        .await
+        .expect("bind mock PocketBase listener");
+    let addr = listener.local_addr().expect("mock PocketBase address");
+    let router = mock_pb_router(pb.clone());
+    tokio::spawn(async move {
+        axum::serve(listener, router)
+            .await
+            .expect("mock PocketBase serve");
+    });
+
+    let state = AppState::for_tests(format!("http://{addr}"));
+    TestEnv {
+        shared: Arc::new(SharedState::new(state)),
+        pb,
+    }
+}
+
+fn now_unix() -> u64 {
+    SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .expect("clock after epoch")
+        .as_secs()
+}
+
+fn test_user(id: &str) -> PocketBaseUser {
+    PocketBaseUser {
+        id: id.to_string(),
+        email: format!("{id}@test.example"),
+        is_admin: false,
+        subscription: "free".to_string(),
+        newsletter: false,
+        can_see_listings: false,
+    }
+}
+
+fn seed_user(pb: &MockPocketBase, id: &str) {
+    pb.seed(
+        "users",
+        json!({
+            "id": id,
+            "email": format!("{id}@test.example"),
+            "subscription": "free",
+            "is_admin": false,
+        }),
+    );
+}
+
+fn seed_pending_checkout(pb: &MockPocketBase, user_id: &str, session_id: &str) -> String {
+    pb.seed(
+        "checkout_sessions",
+        json!({
+            "user": user_id,
+            "stripe_session_id": session_id,
+            "stripe_payment_intent_id": "",
+            "checkout_url": "https://checkout.stripe.test/session",
+            "amount_pence": 999,
+            "expected_total_pence": 999,
+            "currency": "gbp",
+            "discount_coupon_id": "",
+            "referral_invite_id": "",
+            "status": "pending",
+            "expires_at_unix": now_unix() + 1800,
+            "paid_amount_pence": 0,
+            "completed_at_unix": "",
+            "reversal_reason": "",
+        }),
+    )
+}
+
+fn checkout_completed_event(session_id: &str, user_id: &str, amount_total: u64) -> Vec<u8> {
+    serde_json::to_vec(&json!({
+        "id": "evt_test_1",
+        "type": "checkout.session.completed",
+        "data": { "object": {
+            "id": session_id,
+            "payment_intent": "pi_test_1",
+            "client_reference_id": user_id,
+            "payment_status": "paid",
+            "currency": "gbp",
+            "amount_subtotal": 999,
+            "amount_total": amount_total,
+        }}
+    }))
+    .expect("event serializes")
+}
+
+/// Sign a payload the way Stripe does: HMAC-SHA256 over `{timestamp}.{payload}`
+/// with the webhook secret, presented as `t=...,v1=...`.
+fn stripe_signature_header(payload: &[u8], secret: &str) -> String {
+    let timestamp = now_unix();
+    let mut mac =
+        Hmac::<Sha256>::new_from_slice(secret.as_bytes()).expect("HMAC accepts any key length");
+    mac.update(format!("{timestamp}.").as_bytes());
+    mac.update(payload);
+    let signature = hex::encode(mac.finalize().into_bytes());
+    format!("t={timestamp},v1={signature}")
+}
+
+async fn deliver_webhook(env: &TestEnv, payload: Vec<u8>, signature: Option<&str>) -> Response {
+    let mut headers = HeaderMap::new();
+    if let Some(signature) = signature {
+        headers.insert(
+            "stripe-signature",
+            HeaderValue::from_str(signature).expect("signature header value"),
+        );
+    }
+    post_stripe_webhook(State(env.shared.clone()), headers, Bytes::from(payload)).await
+}
+
+async fn redeem_invite(env: &TestEnv, user: PocketBaseUser, code: &str) -> Response {
+    post_redeem_invite(
+        State(env.shared.clone()),
+        Extension(OptionalUser(Some(user))),
+        Json(serde_json::from_value(json!({ "code": code })).expect("redeem request deserializes")),
+    )
+    .await
+}
+
+async fn response_json(response: Response) -> Value {
+    let bytes = axum::body::to_bytes(response.into_body(), 1 << 20)
+        .await
+        .expect("response body");
+    serde_json::from_slice(&bytes).expect("response body is JSON")
+}
+
+// ---------------------------------------------------------------------------
+// Stripe webhook → license granting
+// ---------------------------------------------------------------------------
+
+#[tokio::test]
+async fn webhook_with_valid_signature_grants_license() {
+    let env = setup().await;
+    seed_user(&env.pb, "user1");
+    let reservation_id = seed_pending_checkout(&env.pb, "user1", "cs_test_abc");
+
+    let payload = checkout_completed_event("cs_test_abc", "user1", 999);
+    let signature = stripe_signature_header(&payload, "whsec_test_secret");
+    let response = deliver_webhook(&env, payload, Some(&signature)).await;
+
+    assert_eq!(response.status(), StatusCode::OK);
+
+    let user = env.pb.record("users", "user1").expect("user exists");
+    assert_eq!(user["subscription"], json!("licensed"));
+
+    let checkout = env
+        .pb
+        .record("checkout_sessions", &reservation_id)
+        .expect("checkout exists");
+    assert_eq!(checkout["status"], json!("completed"));
+    assert_eq!(checkout["paid_amount_pence"], json!(999));
+    assert_eq!(checkout["stripe_payment_intent_id"], json!("pi_test_1"));
+
+    assert_eq!(env.pb.license_grant_count("user1"), 1);
+}
+
+#[tokio::test]
+async fn webhook_with_invalid_signature_is_rejected() {
+    let env = setup().await;
+    seed_user(&env.pb, "user1");
+    let reservation_id = seed_pending_checkout(&env.pb, "user1", "cs_test_bad");
+
+    let payload = checkout_completed_event("cs_test_bad", "user1", 999);
+
+    // Signed with the wrong secret.
+    let wrong_signature = stripe_signature_header(&payload, "whsec_wrong_secret");
+    let response = deliver_webhook(&env, payload.clone(), Some(&wrong_signature)).await;
+    assert_eq!(response.status(), StatusCode::BAD_REQUEST);
+
+    // Missing signature header entirely.
+    let response = deliver_webhook(&env, payload, None).await;
+    assert_eq!(response.status(), StatusCode::BAD_REQUEST);
+
+    let user = env.pb.record("users", "user1").expect("user exists");
+    assert_eq!(user["subscription"], json!("free"));
+    let checkout = env
+        .pb
+        .record("checkout_sessions", &reservation_id)
+        .expect("checkout exists");
+    assert_eq!(checkout["status"], json!("pending"));
+    assert_eq!(env.pb.license_grant_count("user1"), 0);
+}
+
+#[tokio::test]
+async fn replayed_webhook_does_not_double_grant() {
+    let env = setup().await;
+    seed_user(&env.pb, "user1");
+    let reservation_id = seed_pending_checkout(&env.pb, "user1", "cs_test_replay");
+
+    let payload = checkout_completed_event("cs_test_replay", "user1", 999);
+    let signature = stripe_signature_header(&payload, "whsec_test_secret");
+
+    let first = deliver_webhook(&env, payload.clone(), Some(&signature)).await;
+    assert_eq!(first.status(), StatusCode::OK);
+    let replay = deliver_webhook(&env, payload, Some(&signature)).await;
+    assert_eq!(replay.status(), StatusCode::OK);
+
+    let user = env.pb.record("users", "user1").expect("user exists");
+    assert_eq!(user["subscription"], json!("licensed"));
+    let checkout = env
+        .pb
+        .record("checkout_sessions", &reservation_id)
+        .expect("checkout exists");
+    assert_eq!(checkout["status"], json!("completed"));
+
+    // The replay must be acknowledged without granting a second time.
+    assert_eq!(env.pb.license_grant_count("user1"), 1);
+}
+
+#[tokio::test]
+async fn webhook_with_tampered_amount_is_rejected_and_marks_reservation_invalid() {
+    let env = setup().await;
+    seed_user(&env.pb, "user1");
+    let reservation_id = seed_pending_checkout(&env.pb, "user1", "cs_test_amount");
+
+    // Validly signed event whose amount_total does not match the reservation.
+    let payload = checkout_completed_event("cs_test_amount", "user1", 500);
+    let signature = stripe_signature_header(&payload, "whsec_test_secret");
+    let response = deliver_webhook(&env, payload, Some(&signature)).await;
+
+    // Rejections are acknowledged with 200 so Stripe stops retrying.
+    assert_eq!(response.status(), StatusCode::OK);
+
+    let user = env.pb.record("users", "user1").expect("user exists");
+    assert_eq!(user["subscription"], json!("free"));
+    let checkout = env
+        .pb
+        .record("checkout_sessions", &reservation_id)
+        .expect("checkout exists");
+    assert_eq!(checkout["status"], json!("invalid"));
+    assert_eq!(env.pb.license_grant_count("user1"), 0);
+}
+
+// ---------------------------------------------------------------------------
+// Checkout session creation (up to the hardcoded Stripe API call)
+// ---------------------------------------------------------------------------
+
+#[tokio::test]
+async fn checkout_start_reserves_then_marks_failed_when_stripe_is_unreachable() {
+    let env = setup().await;
+    seed_user(&env.pb, "user9");
+    let state = env.shared.load_state();
+    let user = test_user("user9");
+
+    // The Stripe API URL is hardcoded to https://api.stripe.com, so the
+    // session-creation call fails in tests (no network / dummy key). The
+    // reservation bookkeeping before and after that call is what we assert.
+    let result = start_license_checkout(
+        &state,
+        &user,
+        "https://x/success",
+        "https://x/cancel",
+        None,
+        None,
+    )
+    .await;
+    assert!(result.is_err(), "Stripe call must fail in tests");
+
+    let checkouts = env.pb.records_in("checkout_sessions");
+    assert_eq!(checkouts.len(), 1, "exactly one reservation created");
+    let checkout = &checkouts[0];
+    assert_eq!(checkout["user"], json!("user9"));
+    // 0 licensed users → public count 120 → second tier price (999p).
+    assert_eq!(checkout["amount_pence"], json!(999));
+    assert_eq!(checkout["expected_total_pence"], json!(999));
+    assert_eq!(checkout["currency"], json!("gbp"));
+    // The failed Stripe call must not leave a live pending reservation.
+    assert_eq!(checkout["status"], json!("failed"));
+
+    // The cross-instance pricing lock was released.
+    assert!(env.pb.records_in("checkout_locks").is_empty());
+    assert_eq!(env.pb.license_grant_count("user9"), 0);
+}
+
+// ---------------------------------------------------------------------------
+// Invite redemption
+// ---------------------------------------------------------------------------
+
+#[tokio::test]
+async fn admin_invite_redemption_grants_license() {
+    let env = setup().await;
+    seed_user(&env.pb, "user2");
+    let invite_id = env.pb.seed(
+        "invites",
+        json!({
+            "code": "admininvite1",
+            "invite_type": "admin",
+            "created_by": "adminuser1",
+            "used_by_id": "",
+            "used_at": "",
+        }),
+    );
+
+    let response = redeem_invite(&env, test_user("user2"), "admininvite1").await;
+    assert_eq!(response.status(), StatusCode::OK);
+    let body = response_json(response).await;
+    assert_eq!(body["result"], json!("licensed"));
+
+    let invite = env.pb.record("invites", &invite_id).expect("invite exists");
+    assert_eq!(invite["used_by_id"], json!("user2"));
+
+    let user = env.pb.record("users", "user2").expect("user exists");
+    assert_eq!(user["subscription"], json!("licensed"));
+    assert_eq!(env.pb.license_grant_count("user2"), 1);
+}
+
+#[tokio::test]
+async fn invalid_and_oversized_invite_codes_are_rejected() {
+    let env = setup().await;
+
+    // Non-alphanumeric characters.
+    let response = redeem_invite(&env, test_user("user2"), "bad-code!").await;
+    assert_eq!(response.status(), StatusCode::BAD_REQUEST);
+
+    // Longer than the 20-character limit.
+    let oversized = "a".repeat(21);
+    let response = redeem_invite(&env, test_user("user2"), &oversized).await;
+    assert_eq!(response.status(), StatusCode::BAD_REQUEST);
+
+    // Empty code.
+    let response = redeem_invite(&env, test_user("user2"), "").await;
+    assert_eq!(response.status(), StatusCode::BAD_REQUEST);
+
+    assert_eq!(env.pb.license_grant_count("user2"), 0);
+}
+
+#[tokio::test]
+async fn already_used_invite_is_rejected() {
+    let env = setup().await;
+    seed_user(&env.pb, "user2");
+    env.pb.seed(
+        "invites",
+        json!({
+            "code": "usedinvite12",
+            "invite_type": "admin",
+            "created_by": "adminuser1",
+            "used_by_id": "otheruser9",
+            "used_at": "1700000000",
+        }),
+    );
+
+    let response = redeem_invite(&env, test_user("user2"), "usedinvite12").await;
+    assert_eq!(response.status(), StatusCode::NOT_FOUND);
+
+    let user = env.pb.record("users", "user2").expect("user exists");
+    assert_eq!(user["subscription"], json!("free"));
+    assert_eq!(env.pb.license_grant_count("user2"), 0);
+}
+
+#[tokio::test]
+async fn referral_invite_redemption_releases_reservation_when_stripe_is_unreachable() {
+    let env = setup().await;
+    seed_user(&env.pb, "user3");
+    let invite_id = env.pb.seed(
+        "invites",
+        json!({
+            "code": "refcode12345",
+            "invite_type": "referral",
+            "created_by": "licenseduser1",
+            "used_by_id": "",
+            "used_at": "",
+            "reserved_by_id": "",
+            "reserved_checkout_id": "",
+            "reserved_until_unix": 0,
+        }),
+    );
+
+    // The redemption itself is valid; it fails only at the hardcoded Stripe
+    // call (coupon verification / session creation), which must roll back the
+    // reservation cleanly.
+    let response = redeem_invite(&env, test_user("user3"), "refcode12345").await;
+    assert_eq!(response.status(), StatusCode::BAD_GATEWAY);
+
+    let checkouts = env.pb.records_in("checkout_sessions");
+    assert_eq!(checkouts.len(), 1, "referral reservation was created");
+    assert_eq!(checkouts[0]["referral_invite_id"], json!(invite_id));
+    assert_eq!(checkouts[0]["status"], json!("failed"));
+
+    // The invite reservation was released and the invite is still unused.
+    let invite = env.pb.record("invites", &invite_id).expect("invite exists");
+    assert_eq!(invite["used_by_id"], json!(""));
+    assert_eq!(invite["reserved_checkout_id"], json!(""));
+    assert_eq!(invite["reserved_by_id"], json!(""));
+
+    assert_eq!(env.pb.license_grant_count("user3"), 0);
+}
diff --git a/server-rs/src/data/crime_by_year.rs b/server-rs/src/data/crime_by_year.rs
index b77eed6..73a96f4 100644
--- a/server-rs/src/data/crime_by_year.rs
+++ b/server-rs/src/data/crime_by_year.rs
@@ -182,8 +182,7 @@ impl CrimeByYearData {
         // Force-coverage calendar (optional column: legacy parquets predate it;
         // their postcodes are treated as fully covered). A row with an empty
         // list is meaningful — zero covered years — so it IS inserted.
-        let mut covered_years_by_postcode: FxHashMap<String, Vec<i32>> =
-            FxHashMap::default();
+        let mut covered_years_by_postcode: FxHashMap<String, Vec<i32>> = FxHashMap::default();
         if let Ok(col) = df.column(COVERAGE_COLUMN) {
             let list_ca = col
                 .list()
@@ -195,12 +194,12 @@ impl CrimeByYearData {
                 };
                 let mut years: Vec<i32> = Vec::with_capacity(inner.len());
                 if !inner.is_empty() {
-                    let structs = inner.struct_().with_context(|| {
-                        format!("Inner of '{COVERAGE_COLUMN}' is not a struct")
-                    })?;
-                    let year_field = structs.field_by_name("year").with_context(|| {
-                        format!("Missing 'year' field in '{COVERAGE_COLUMN}'")
-                    })?;
+                    let structs = inner
+                        .struct_()
+                        .with_context(|| format!("Inner of '{COVERAGE_COLUMN}' is not a struct"))?;
+                    let year_field = structs
+                        .field_by_name("year")
+                        .with_context(|| format!("Missing 'year' field in '{COVERAGE_COLUMN}'"))?;
                     for idx in 0..inner.len() {
                         match year_field.get(idx).ok() {
                             Some(AnyValue::Int32(y)) => years.push(y),
diff --git a/server-rs/src/data/places.rs b/server-rs/src/data/places.rs
index 4d4744a..7f0b662 100644
--- a/server-rs/src/data/places.rs
+++ b/server-rs/src/data/places.rs
@@ -742,6 +742,29 @@ impl PlaceData {
     }
 }
 
+#[cfg(test)]
+impl PlaceData {
+    /// Minimal empty instance for integration tests that need an `AppState`
+    /// but never touch place data.
+    pub(crate) fn empty_for_tests() -> Self {
+        PlaceData {
+            name: Vec::new(),
+            name_lower: Vec::new(),
+            name_search: Vec::new(),
+            place_type: InternedColumn::build(&[]),
+            type_rank: Vec::new(),
+            population: Vec::new(),
+            lat: Vec::new(),
+            lon: Vec::new(),
+            city: Vec::new(),
+            travel_destination: Vec::new(),
+            token_index: FxHashMap::default(),
+            token_prefix_index: FxHashMap::default(),
+            fuzzy_trigram_index: FxHashMap::default(),
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/server-rs/src/data/poi.rs b/server-rs/src/data/poi.rs
index b6c223b..4f29526 100644
--- a/server-rs/src/data/poi.rs
+++ b/server-rs/src/data/poi.rs
@@ -588,6 +588,29 @@ impl POIData {
     }
 }
 
+#[cfg(test)]
+impl POIData {
+    /// Minimal empty instance for integration tests that need an `AppState`
+    /// but never touch POI data.
+    pub(crate) fn empty_for_tests() -> Self {
+        POIData {
+            id_buffer: String::new(),
+            id_offsets: Vec::new(),
+            id_lengths: Vec::new(),
+            group: InternedColumn::build(&[]),
+            category: InternedColumn::build(&[]),
+            icon_category: InternedColumn::build(&[]),
+            name: Vec::new(),
+            lat: Vec::new(),
+            lng: Vec::new(),
+            emoji: InternedColumn::build(&[]),
+            priority: Vec::new(),
+            school_meta_idx: Vec::new(),
+            school_meta: Vec::new(),
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/server-rs/src/data/property.rs b/server-rs/src/data/property.rs
deleted file mode 100644
index dd04645..0000000
--- a/server-rs/src/data/property.rs
+++ /dev/null
@@ -1,3005 +0,0 @@
-use anyhow::{bail, Context};
-use polars::lazy::frame::LazyFrame;
-use polars::prelude::*;
-use rayon::prelude::*;
-use serde::Serialize;
-use std::path::Path;
-
-use rustc_hash::{FxHashMap, FxHashSet};
-
-use crate::consts::{H3_PRECOMPUTE_MAX, HISTOGRAM_BINS, NAN_U16, QUANT_SCALE};
-use crate::features::{self, Bounds};
-
-/// Upper bound on rows scored per query. Intersection keeps most candidate sets far below
-/// this; only a single very common road word (e.g. "high") approaches it, and the in-area
-/// priority sort keeps a refined query's matches ahead of the cut.
-const ADDRESS_SEARCH_CANDIDATE_LIMIT: usize = 150_000;
-const ADDRESS_SEARCH_PREFIX_MIN_LEN: usize = 4;
-const ADDRESS_SEARCH_PREFIX_MAX_LEN: usize = 8;
-const NO_POI_METRIC_ROW: u32 = u32::MAX;
-const MISSING_COORDINATE_SAMPLE_LIMIT: usize = 10;
-const COUNTRY_COLUMN_CANDIDATES: &[&str] = &[
-    "ctry25cd",
-    "ctry24cd",
-    "ctry23cd",
-    "ctry22cd",
-    "country_code",
-    "Country code",
-    "country",
-    "Country",
-];
-const ENGLAND_COUNTRY_VALUES: &[&str] = &["E92000001", "England", "ENGLAND", "england"];
-
-fn is_numeric_dtype(dtype: &DataType) -> bool {
-    matches!(
-        dtype,
-        DataType::Int8
-            | DataType::Int16
-            | DataType::Int32
-            | DataType::Int64
-            | DataType::UInt8
-            | DataType::UInt16
-            | DataType::UInt32
-            | DataType::UInt64
-            | DataType::Float32
-            | DataType::Float64
-            | DataType::Datetime(_, _)
-            | DataType::Date
-    )
-}
-
-fn is_datetime_dtype(dtype: &DataType) -> bool {
-    matches!(dtype, DataType::Datetime(_, _) | DataType::Date)
-}
-
-fn find_country_column(schema: &Schema) -> Option<String> {
-    COUNTRY_COLUMN_CANDIDATES
-        .iter()
-        .find_map(|&name| match schema.get(name) {
-            Some(dtype) if matches!(dtype, DataType::String) || dtype.is_categorical() => {
-                Some(name.to_string())
-            }
-            _ => None,
-        })
-}
-
-fn england_country_expr(country_column: &str) -> Expr {
-    ENGLAND_COUNTRY_VALUES.iter().skip(1).fold(
-        col(country_column)
-            .cast(DataType::String)
-            .eq(lit(ENGLAND_COUNTRY_VALUES[0])),
-        |expr, value| expr.or(col(country_column).cast(DataType::String).eq(lit(*value))),
-    )
-}
-
-#[derive(Debug, PartialEq, Eq)]
-struct MissingCoordinateSummary {
-    row_count: usize,
-    unique_postcode_count: usize,
-    sample_postcodes: Vec<String>,
-}
-
-fn summarize_missing_coordinate_postcodes<'a>(
-    postcodes: impl IntoIterator<Item = Option<&'a str>>,
-) -> MissingCoordinateSummary {
-    let mut row_count = 0usize;
-    let mut unique_postcodes = FxHashSet::default();
-
-    for postcode in postcodes {
-        row_count += 1;
-        if let Some(postcode) = postcode {
-            let trimmed = postcode.trim();
-            if !trimmed.is_empty() {
-                unique_postcodes.insert(trimmed.to_string());
-            }
-        }
-    }
-
-    let mut sample_postcodes: Vec<String> = unique_postcodes.iter().cloned().collect();
-    sample_postcodes.sort_unstable();
-    sample_postcodes.truncate(MISSING_COORDINATE_SAMPLE_LIMIT);
-
-    MissingCoordinateSummary {
-        row_count,
-        unique_postcode_count: unique_postcodes.len(),
-        sample_postcodes,
-    }
-}
-
-fn missing_england_coordinates_error(
-    summary: &MissingCoordinateSummary,
-    country_column: &str,
-) -> String {
-    let samples = if summary.sample_postcodes.is_empty() {
-        "none".to_string()
-    } else {
-        summary.sample_postcodes.join(", ")
-    };
-    format!(
-        "England property rows missing postcode coordinates after joining postcode data: {} rows across {} postcodes (country column '{}'). Sample postcodes: {}",
-        summary.row_count, summary.unique_postcode_count, country_column, samples
-    )
-}
-
-fn validate_no_england_rows_missing_coordinates(
-    combined_lf: &LazyFrame,
-    schema: &Schema,
-) -> anyhow::Result<()> {
-    let Some(country_column) = find_country_column(schema) else {
-        bail!(
-            "Postcode feature parquet has no reliable country column; cannot verify that rows with missing coordinates are outside England. Regenerate postcode.parquet with pipeline.transform.merge so it includes ctry25cd."
-        );
-    };
-
-    let missing_coordinates = col("lat").is_null().or(col("lon").is_null());
-    let offending_df = combined_lf
-        .clone()
-        .filter(missing_coordinates.and(england_country_expr(&country_column)))
-        .select([col("Postcode")])
-        .collect()
-        .context("Failed to validate missing postcode coordinates")?;
-
-    let postcode_column = offending_df
-        .column("Postcode")
-        .context("Joined frame missing 'Postcode' during coordinate validation")?
-        .str()
-        .context("'Postcode' column is not a string during coordinate validation")?;
-    let summary = summarize_missing_coordinate_postcodes(postcode_column);
-
-    if summary.row_count > 0 {
-        bail!(
-            "{}",
-            missing_england_coordinates_error(&summary, &country_column)
-        );
-    }
-
-    Ok(())
-}
-
-#[derive(Clone, Debug)]
-struct AddressTermGroup {
-    alternatives: Vec<String>,
-}
-
-#[derive(Debug)]
-struct AddressQuery {
-    full_postcode: Option<String>,
-    /// Compact uppercase outward code (optionally with a sector digit) recovered when the
-    /// user appended a partial postcode like "NW1" or "NW1 6". Used as an additive ranking
-    /// bias, never as a hard filter — so the disambiguating hint is honoured without
-    /// excluding the same road in other areas.
-    postcode_area: Option<String>,
-    text_groups: Vec<AddressTermGroup>,
-    numeric_terms: Vec<String>,
-    candidate_terms: Vec<String>,
-}
-
-fn tokenize_address_text(text: &str) -> Vec<String> {
-    let mut tokens = Vec::new();
-    let mut current = String::new();
-
-    for ch in text.chars() {
-        if ch.is_ascii_alphanumeric() {
-            current.push(ch.to_ascii_lowercase());
-        } else if matches!(ch, '\'' | '’' | '`') {
-            continue;
-        } else if !current.is_empty() {
-            tokens.push(std::mem::take(&mut current));
-        }
-    }
-
-    if !current.is_empty() {
-        tokens.push(current);
-    }
-
-    tokens
-}
-
-fn is_full_postcode_compact(compact: &str) -> bool {
-    let bytes = compact.as_bytes();
-    let len = bytes.len();
-    if !(5..=7).contains(&len) {
-        return false;
-    }
-
-    let inward = &bytes[len - 3..];
-    if !inward[0].is_ascii_digit()
-        || !inward[1].is_ascii_alphabetic()
-        || !inward[2].is_ascii_alphabetic()
-    {
-        return false;
-    }
-
-    let outward = &bytes[..len - 3];
-    if !(2..=4).contains(&outward.len()) {
-        return false;
-    }
-
-    outward[0].is_ascii_alphabetic()
-        && outward.iter().all(u8::is_ascii_alphanumeric)
-        && outward.iter().any(u8::is_ascii_digit)
-}
-
-fn canonical_postcode_from_compact(compact: &str) -> String {
-    let upper = compact.to_ascii_uppercase();
-    let split = upper.len() - 3;
-    format!("{} {}", &upper[..split], &upper[split..])
-}
-
-fn extract_full_postcode(tokens: &[String]) -> Option<(String, Vec<usize>)> {
-    for (idx, token) in tokens.iter().enumerate() {
-        let compact = token.to_ascii_uppercase();
-        if is_full_postcode_compact(&compact) {
-            return Some((canonical_postcode_from_compact(&compact), vec![idx]));
-        }
-    }
-
-    for idx in 0..tokens.len().saturating_sub(1) {
-        let compact = format!(
-            "{}{}",
-            tokens[idx].to_ascii_uppercase(),
-            tokens[idx + 1].to_ascii_uppercase()
-        );
-        if is_full_postcode_compact(&compact) {
-            return Some((
-                canonical_postcode_from_compact(&compact),
-                vec![idx, idx + 1],
-            ));
-        }
-    }
-
-    None
-}
-
-fn looks_like_postcode_fragment(token: &str) -> bool {
-    (2..=4).contains(&token.len())
-        && token
-            .chars()
-            .next()
-            .is_some_and(|ch| ch.is_ascii_alphabetic())
-        && token.chars().any(|ch| ch.is_ascii_digit())
-        && token.chars().all(|ch| ch.is_ascii_alphanumeric())
-}
-
-fn is_numeric_address_token(token: &str) -> bool {
-    token.chars().all(|ch| ch.is_ascii_digit())
-}
-
-fn address_token_aliases(token: &str) -> Vec<&'static str> {
-    match token {
-        "apt" => vec!["apt", "apartment"],
-        "apartment" => vec!["apartment", "apt"],
-        "ave" => vec!["ave", "avenue"],
-        "avenue" => vec!["avenue", "ave"],
-        "blvd" => vec!["blvd", "boulevard"],
-        "boulevard" => vec!["boulevard", "blvd"],
-        "cl" => vec!["cl", "close"],
-        "close" => vec!["close", "cl"],
-        "ct" => vec!["ct", "court"],
-        "court" => vec!["court", "ct"],
-        "cres" => vec!["cres", "crescent"],
-        "crescent" => vec!["crescent", "cres"],
-        "dr" => vec!["dr", "drive"],
-        "drive" => vec!["drive", "dr"],
-        "fl" => vec!["fl", "flat"],
-        "flat" => vec!["flat", "fl"],
-        "gdns" => vec!["gdns", "gardens", "garden"],
-        "garden" => vec!["garden", "gardens", "gdns"],
-        "gardens" => vec!["gardens", "garden", "gdns"],
-        "hse" => vec!["hse", "house"],
-        "house" => vec!["house", "hse"],
-        "ln" => vec!["ln", "lane"],
-        "lane" => vec!["lane", "ln"],
-        "rd" => vec!["rd", "road"],
-        "road" => vec!["road", "rd"],
-        "sq" => vec!["sq", "square"],
-        "square" => vec!["square", "sq"],
-        "st" => vec!["st", "street", "saint"],
-        "street" => vec!["street", "st"],
-        "saint" => vec!["saint", "st"],
-        "terr" => vec!["terr", "terrace"],
-        "terrace" => vec!["terrace", "terr"],
-        _ => Vec::new(),
-    }
-}
-
-fn is_address_stop_token(token: &str) -> bool {
-    matches!(
-        token,
-        "a" | "an"
-            | "and"
-            | "apartment"
-            | "apt"
-            | "avenue"
-            | "ave"
-            | "block"
-            | "building"
-            | "bungalow"
-            | "close"
-            | "cl"
-            | "court"
-            | "ct"
-            | "cres"
-            | "crescent"
-            | "drive"
-            | "dr"
-            | "estate"
-            | "flat"
-            | "fl"
-            | "floor"
-            | "garden"
-            | "gardens"
-            | "gdns"
-            | "grove"
-            | "house"
-            | "hse"
-            | "lane"
-            | "ln"
-            | "lodge"
-            | "mansions"
-            | "mews"
-            | "of"
-            | "park"
-            | "place"
-            | "road"
-            | "rd"
-            | "room"
-            | "row"
-            | "saint"
-            | "sq"
-            | "square"
-            | "st"
-            | "street"
-            | "terr"
-            | "terrace"
-            | "the"
-            | "unit"
-            | "view"
-            | "villas"
-            | "walk"
-            | "way"
-            | "yard"
-    )
-}
-
-fn address_term_group(token: &str) -> Option<AddressTermGroup> {
-    if token.len() < 3 || is_numeric_address_token(token) || looks_like_postcode_fragment(token) {
-        return None;
-    }
-
-    let mut alternatives = Vec::new();
-    alternatives.push(token.to_string());
-    for alias in address_token_aliases(token) {
-        if !alternatives.iter().any(|existing| existing == alias) {
-            alternatives.push(alias.to_string());
-        }
-    }
-
-    if alternatives
-        .iter()
-        .all(|alternative| is_address_stop_token(alternative))
-    {
-        return None;
-    }
-
-    Some(AddressTermGroup { alternatives })
-}
-
-fn address_search_tokens(text: &str) -> Vec<String> {
-    let mut tokens: Vec<String> = tokenize_address_text(text)
-        .into_iter()
-        .filter(|token| is_address_search_token(token))
-        .collect();
-    tokens.sort_unstable();
-    tokens.dedup();
-    tokens
-}
-
-fn is_address_search_token(token: &str) -> bool {
-    if looks_like_postcode_fragment(token) {
-        return false;
-    }
-
-    if is_numeric_address_token(token) {
-        return true;
-    }
-
-    if token.chars().any(|ch| ch.is_ascii_digit()) {
-        return token.len() >= 2;
-    }
-
-    token.len() >= 3
-}
-
-fn is_address_candidate_token(token: &str) -> bool {
-    !is_numeric_address_token(token)
-        && !looks_like_postcode_fragment(token)
-        && (token.chars().any(|ch| ch.is_ascii_digit())
-            || (token.len() >= 3 && !is_address_stop_token(token)))
-}
-
-fn address_prefix_key(term: &str) -> &str {
-    if term.len() > ADDRESS_SEARCH_PREFIX_MAX_LEN {
-        &term[..ADDRESS_SEARCH_PREFIX_MAX_LEN]
-    } else {
-        term
-    }
-}
-
-fn build_address_prefix_index(
-    address_token_index: &FxHashMap<String, Vec<u32>>,
-) -> FxHashMap<String, Vec<String>> {
-    let mut prefix_index: FxHashMap<String, Vec<String>> = FxHashMap::default();
-
-    for token in address_token_index.keys() {
-        let max_prefix_len = token.len().min(ADDRESS_SEARCH_PREFIX_MAX_LEN);
-        for prefix_len in ADDRESS_SEARCH_PREFIX_MIN_LEN..=max_prefix_len {
-            prefix_index
-                .entry(token[..prefix_len].to_string())
-                .or_default()
-                .push(token.clone());
-        }
-    }
-
-    for tokens in prefix_index.values_mut() {
-        tokens.sort_unstable();
-        tokens.dedup();
-    }
-
-    prefix_index
-}
-
-/// Intersect two ascending-sorted row-id slices.
-fn intersect_sorted(left: &[u32], right: &[u32]) -> Vec<u32> {
-    let mut out = Vec::new();
-    let (mut i, mut j) = (0, 0);
-    while i < left.len() && j < right.len() {
-        match left[i].cmp(&right[j]) {
-            std::cmp::Ordering::Less => i += 1,
-            std::cmp::Ordering::Greater => j += 1,
-            std::cmp::Ordering::Equal => {
-                out.push(left[i]);
-                i += 1;
-                j += 1;
-            }
-        }
-    }
-    out
-}
-
-/// Union two ascending-sorted row-id slices (deduplicated, stays sorted).
-fn union_sorted(left: &[u32], right: &[u32]) -> Vec<u32> {
-    let mut out = Vec::with_capacity(left.len() + right.len());
-    let (mut i, mut j) = (0, 0);
-    while i < left.len() && j < right.len() {
-        match left[i].cmp(&right[j]) {
-            std::cmp::Ordering::Less => {
-                out.push(left[i]);
-                i += 1;
-            }
-            std::cmp::Ordering::Greater => {
-                out.push(right[j]);
-                j += 1;
-            }
-            std::cmp::Ordering::Equal => {
-                out.push(left[i]);
-                i += 1;
-                j += 1;
-            }
-        }
-    }
-    out.extend_from_slice(&left[i..]);
-    out.extend_from_slice(&right[j..]);
-    out
-}
-
-/// An ordinal like "1st", "2nd", "3rd", "21st" — part of the street name ("2nd Avenue"), not a
-/// house-number prefix.
-fn is_ordinal_token(token: &str) -> bool {
-    let split = token.len().saturating_sub(2);
-    let (digits, suffix) = token.split_at(split);
-    !digits.is_empty()
-        && digits.chars().all(|ch| ch.is_ascii_digit())
-        && matches!(suffix, "st" | "nd" | "rd" | "th")
-}
-
-/// Leading address tokens that denote a unit/house number rather than the street itself.
-fn is_house_prefix_token(token: &str) -> bool {
-    if is_ordinal_token(token) {
-        return false;
-    }
-    matches!(
-        token,
-        "flat" | "fl" | "apartment" | "apt" | "unit" | "no" | "block" | "floor" | "room"
-    ) || token.len() == 1
-        || token.chars().all(|ch| ch.is_ascii_digit())
-        || (token.chars().next().is_some_and(|ch| ch.is_ascii_digit())
-            && token.chars().any(|ch| ch.is_ascii_alphabetic()))
-}
-
-/// Street-level key for an address: drops the leading house-number / flat prefix so that
-/// "12 Baker Street" and "5 Baker Street" collapse to a single street entry.
-fn street_key(address: &str) -> String {
-    let tokens = tokenize_address_text(address);
-    let mut start = 0;
-    while start < tokens.len() && is_house_prefix_token(&tokens[start]) {
-        start += 1;
-    }
-    if start >= tokens.len() {
-        return tokens.join(" ");
-    }
-    tokens[start..].join(" ")
-}
-
-/// Road-type words. Their presence (with no house number) marks a road browse, which we
-/// collapse to one result per street.
-const ROAD_TYPE_TOKENS: &[&str] = &[
-    "street",
-    "st",
-    "road",
-    "rd",
-    "lane",
-    "ln",
-    "avenue",
-    "ave",
-    "close",
-    "cl",
-    "drive",
-    "dr",
-    "way",
-    "court",
-    "ct",
-    "crescent",
-    "cres",
-    "place",
-    "terrace",
-    "terr",
-    "grove",
-    "gardens",
-    "gdns",
-    "walk",
-    "row",
-    "square",
-    "sq",
-    "hill",
-    "parade",
-    "mews",
-    "embankment",
-    "broadway",
-    "boulevard",
-    "blvd",
-];
-
-fn query_has_road_type(query: &str) -> bool {
-    tokenize_address_text(query)
-        .iter()
-        .any(|token| ROAD_TYPE_TOKENS.contains(&token.as_str()))
-}
-
-/// The outward code (everything before the space) of a canonical postcode.
-fn outcode_of(postcode: &str) -> &str {
-    postcode.split(' ').next().unwrap_or(postcode)
-}
-
-fn parse_address_query(query: &str) -> AddressQuery {
-    let tokens = tokenize_address_text(query);
-    let (full_postcode, postcode_token_indices) = extract_full_postcode(&tokens)
-        .map(|(postcode, indices)| (Some(postcode), indices))
-        .unwrap_or((None, Vec::new()));
-
-    let skip_postcode_tokens: FxHashSet<usize> = postcode_token_indices.into_iter().collect();
-
-    // Recover an appended partial postcode (outcode, or outcode + sector digit) as a ranking
-    // bias rather than discarding it — but only from the TRAILING position, so a leading road
-    // designation like "A4 Great West Road" is not mistaken for an area refinement.
-    let mut postcode_area: Option<String> = None;
-    let mut consumed_partial_tokens: FxHashSet<usize> = FxHashSet::default();
-    if full_postcode.is_none() && !tokens.is_empty() {
-        let last = tokens.len() - 1;
-        if !skip_postcode_tokens.contains(&last) {
-            let sector_digit =
-                tokens[last].len() == 1 && tokens[last].chars().all(|ch| ch.is_ascii_digit());
-            if last >= 1
-                && sector_digit
-                && !skip_postcode_tokens.contains(&(last - 1))
-                && looks_like_postcode_fragment(&tokens[last - 1])
-            {
-                postcode_area = Some(format!(
-                    "{}{}",
-                    tokens[last - 1].to_ascii_uppercase(),
-                    tokens[last]
-                ));
-                consumed_partial_tokens.insert(last);
-                consumed_partial_tokens.insert(last - 1);
-            } else if looks_like_postcode_fragment(&tokens[last]) {
-                postcode_area = Some(tokens[last].to_ascii_uppercase());
-                consumed_partial_tokens.insert(last);
-            }
-        }
-    }
-
-    let mut text_groups = Vec::new();
-    let mut numeric_terms = Vec::new();
-    let mut candidate_terms = Vec::new();
-
-    for (idx, token) in tokens.iter().enumerate() {
-        if skip_postcode_tokens.contains(&idx)
-            || consumed_partial_tokens.contains(&idx)
-            || looks_like_postcode_fragment(token)
-        {
-            continue;
-        }
-
-        if is_numeric_address_token(token) {
-            numeric_terms.push(token.clone());
-            continue;
-        }
-
-        if let Some(group) = address_term_group(token) {
-            for alternative in &group.alternatives {
-                if !is_address_stop_token(alternative)
-                    && !candidate_terms.iter().any(|term| term == alternative)
-                {
-                    candidate_terms.push(alternative.clone());
-                }
-            }
-            text_groups.push(group);
-        } else if token.chars().any(|ch| ch.is_ascii_digit()) && token.len() >= 2 {
-            numeric_terms.push(token.clone());
-            if !candidate_terms.iter().any(|term| term == token) {
-                candidate_terms.push(token.clone());
-            }
-        }
-    }
-
-    text_groups.dedup_by(|left, right| left.alternatives == right.alternatives);
-    numeric_terms.sort_unstable();
-    numeric_terms.dedup();
-
-    AddressQuery {
-        full_postcode,
-        postcode_area,
-        text_groups,
-        numeric_terms,
-        candidate_terms,
-    }
-}
-
-fn token_matches_query_term(token: &str, query_term: &str) -> bool {
-    token == query_term || (query_term.len() >= 3 && token.starts_with(query_term))
-}
-
-fn token_matches_numeric_term(token: &str, query_term: &str) -> bool {
-    token == query_term || token.starts_with(query_term)
-}
-
-#[cfg(test)]
-fn address_tokens_match_group(tokens: &[String], group: &AddressTermGroup) -> bool {
-    group.alternatives.iter().any(|alternative| {
-        tokens
-            .iter()
-            .any(|token| token_matches_query_term(token, alternative))
-    })
-}
-
-/// Histogram with outlier buckets at the edges.
-/// - Bin 0: [min, p1) — low outliers
-/// - Bins 1 to n-2: [p1, p99) — main distribution, evenly divided
-/// - Bin n-1: [p99, max] — high outliers
-#[derive(Serialize, Clone)]
-pub struct Histogram {
-    pub min: f32,
-    pub max: f32,
-    /// 1st percentile (left edge of main distribution)
-    pub p1: f32,
-    /// 99th percentile (right edge of main distribution)
-    pub p99: f32,
-    pub counts: Vec<u64>,
-}
-
-impl Histogram {
-    /// Return the bin index for a given value using the outlier-bracket layout.
-    #[cfg(test)]
-    pub fn bin_for_value(&self, value: f32) -> usize {
-        let num_bins = self.counts.len();
-        if value < self.p1 {
-            0
-        } else if value >= self.p99 {
-            num_bins - 1
-        } else {
-            let middle_bins = num_bins.saturating_sub(2);
-            if middle_bins > 0 && self.p99 > self.p1 {
-                let width = (self.p99 - self.p1) / middle_bins as f32;
-                let middle_bin = ((value - self.p1) / width) as usize;
-                (1 + middle_bin).min(num_bins - 2)
-            } else {
-                num_bins / 2
-            }
-        }
-    }
-
-    /// Width of a single middle bin (bins 1..n-2).
-    #[cfg(test)]
-    pub fn middle_bin_width(&self) -> f32 {
-        let middle_bins = self.counts.len().saturating_sub(2);
-        if middle_bins > 0 && self.p99 > self.p1 {
-            (self.p99 - self.p1) / middle_bins as f32
-        } else {
-            0.0
-        }
-    }
-}
-
-pub struct FeatureStats {
-    pub slider_min: f32,
-    pub slider_max: f32,
-    pub histogram: Histogram,
-}
-
-#[derive(Serialize, Clone)]
-pub struct RenovationEvent {
-    pub year: i32,
-    pub event: String,
-}
-
-#[derive(Serialize, Clone)]
-pub struct HistoricalPrice {
-    pub year: i32,
-    pub month: u8,
-    pub price: i64,
-}
-
-/// Lightweight reference to quantization parameters for decoding u16 feature data.
-pub struct QuantRef<'a> {
-    pub dequant_a: &'a [f32],
-    pub quant_min: &'a [f32],
-    pub quant_range: &'a [f32],
-    pub num_numeric: usize,
-}
-
-impl QuantRef<'_> {
-    /// Decode a raw u16 value back to f32.
-    #[inline]
-    pub fn decode(&self, feat_idx: usize, raw: u16) -> f32 {
-        if raw == NAN_U16 {
-            return f32::NAN;
-        }
-        if feat_idx >= self.num_numeric {
-            raw as f32
-        } else {
-            raw as f32 * self.dequant_a[feat_idx] + self.quant_min[feat_idx]
-        }
-    }
-
-    /// Encode a filter minimum bound to u16 (floors to include boundary values).
-    #[inline]
-    pub fn encode_min(&self, feat_idx: usize, value: f32) -> u16 {
-        if !value.is_finite() || self.quant_range[feat_idx] == 0.0 {
-            return 0;
-        }
-        let norm = (value - self.quant_min[feat_idx]) / self.quant_range[feat_idx];
-        (norm * QUANT_SCALE).floor().clamp(0.0, QUANT_SCALE) as u16
-    }
-
-    /// Encode a filter maximum bound to u16 (ceils to include boundary values).
-    #[inline]
-    pub fn encode_max(&self, feat_idx: usize, value: f32) -> u16 {
-        if !value.is_finite() || self.quant_range[feat_idx] == 0.0 {
-            return QUANT_SCALE as u16;
-        }
-        let norm = (value - self.quant_min[feat_idx]) / self.quant_range[feat_idx];
-        (norm * QUANT_SCALE).ceil().clamp(0.0, QUANT_SCALE) as u16
-    }
-}
-
-pub struct PostcodePoiMetrics {
-    pub feature_names: Vec<String>,
-    pub name_to_index: FxHashMap<String, usize>,
-    /// Metric-major storage: columns[metric_idx][postcode_metric_idx].
-    pub columns: Vec<Vec<u16>>,
-    pub feature_stats: Vec<FeatureStats>,
-    /// Per-property row lookup into the postcode metric table.
-    row_to_metric_idx: Vec<u32>,
-    dequant_a: Vec<f32>,
-    quant_min: Vec<f32>,
-    quant_range: Vec<f32>,
-}
-
-impl PostcodePoiMetrics {
-    fn empty(row_count: usize) -> Self {
-        Self {
-            feature_names: Vec::new(),
-            name_to_index: FxHashMap::default(),
-            columns: Vec::new(),
-            feature_stats: Vec::new(),
-            row_to_metric_idx: vec![NO_POI_METRIC_ROW; row_count],
-            dequant_a: Vec::new(),
-            quant_min: Vec::new(),
-            quant_range: Vec::new(),
-        }
-    }
-
-    fn from_postcode_df(df: &DataFrame, feature_names: Vec<String>) -> anyhow::Result<Self> {
-        if feature_names.is_empty() {
-            return Ok(Self::empty(0));
-        }
-
-        tracing::info!(
-            metrics = feature_names.len(),
-            postcodes = df.height(),
-            "Building postcode POI metric side table"
-        );
-
-        let col_major: Vec<Vec<f32>> = feature_names
-            .par_iter()
-            .map(|name| {
-                let column = df
-                    .column(name.as_str())
-                    .with_context(|| format!("Missing POI metric column '{name}'"))?;
-                column_to_f32_vec(column)
-            })
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        let feature_stats: Vec<FeatureStats> = col_major
-            .par_iter()
-            .enumerate()
-            .map(|(metric_idx, vals)| {
-                let name = feature_names[metric_idx].as_str();
-                let bounds = features::bounds_for(name)
-                    .with_context(|| format!("No bounds config for POI metric '{name}'"))?;
-                Ok(compute_feature_stats(
-                    vals,
-                    &bounds,
-                    features::has_integer_bins(name),
-                ))
-            })
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        let mut quant_min = Vec::with_capacity(feature_names.len());
-        let mut quant_range = Vec::with_capacity(feature_names.len());
-        for (metric_idx, stats) in feature_stats.iter().enumerate() {
-            let (min, max) = match features::bounds_for(feature_names[metric_idx].as_str()) {
-                Some(Bounds::Fixed { min, max }) => (min, max),
-                _ => (stats.histogram.min, stats.histogram.max),
-            };
-            quant_min.push(min);
-            quant_range.push(if max > min { max - min } else { 0.0 });
-        }
-        let dequant_a: Vec<f32> = quant_range
-            .iter()
-            .map(|&range| {
-                if range > 0.0 {
-                    range / QUANT_SCALE
-                } else {
-                    0.0
-                }
-            })
-            .collect();
-
-        let columns: Vec<Vec<u16>> = col_major
-            .par_iter()
-            .enumerate()
-            .map(|(metric_idx, vals)| {
-                let range = quant_range[metric_idx];
-                let min = quant_min[metric_idx];
-                vals.iter()
-                    .map(|&value| {
-                        if !value.is_finite() {
-                            NAN_U16
-                        } else if range > 0.0 {
-                            let normalized = (value - min) / range;
-                            (normalized * QUANT_SCALE).round().clamp(0.0, QUANT_SCALE) as u16
-                        } else {
-                            0
-                        }
-                    })
-                    .collect()
-            })
-            .collect();
-
-        let name_to_index = feature_names
-            .iter()
-            .enumerate()
-            .map(|(idx, name)| (name.clone(), idx))
-            .collect();
-
-        Ok(Self {
-            feature_names,
-            name_to_index,
-            columns,
-            feature_stats,
-            row_to_metric_idx: Vec::new(),
-            dequant_a,
-            quant_min,
-            quant_range,
-        })
-    }
-
-    fn set_row_mapping(&mut self, row_to_metric_idx: Vec<u32>) {
-        self.row_to_metric_idx = row_to_metric_idx;
-    }
-
-    pub fn is_empty(&self) -> bool {
-        self.feature_names.is_empty()
-    }
-
-    pub fn num_features(&self) -> usize {
-        self.feature_names.len()
-    }
-
-    pub fn quant_ref(&self) -> QuantRef<'_> {
-        QuantRef {
-            dequant_a: &self.dequant_a,
-            quant_min: &self.quant_min,
-            quant_range: &self.quant_range,
-            num_numeric: self.feature_names.len(),
-        }
-    }
-
-    #[inline]
-    pub fn metric_row_for_property(&self, row: usize) -> Option<usize> {
-        self.row_to_metric_idx
-            .get(row)
-            .copied()
-            .filter(|&idx| idx != NO_POI_METRIC_ROW)
-            .map(|idx| idx as usize)
-    }
-
-    #[inline]
-    pub fn raw_for_metric_row(&self, metric_row: usize, metric_idx: usize) -> u16 {
-        self.columns[metric_idx][metric_row]
-    }
-
-    #[inline]
-    pub fn raw_for_property_row(&self, row: usize, metric_idx: usize) -> u16 {
-        let Some(metric_row) = self.metric_row_for_property(row) else {
-            return NAN_U16;
-        };
-        self.raw_for_metric_row(metric_row, metric_idx)
-    }
-
-    #[inline]
-    pub fn decode_raw(&self, metric_idx: usize, raw: u16) -> f32 {
-        if raw == NAN_U16 {
-            f32::NAN
-        } else {
-            raw as f32 * self.dequant_a[metric_idx] + self.quant_min[metric_idx]
-        }
-    }
-
-    #[inline]
-    pub fn get_for_property_row(&self, row: usize, metric_idx: usize) -> f32 {
-        self.decode_raw(metric_idx, self.raw_for_property_row(row, metric_idx))
-    }
-}
-
-pub struct PropertyData {
-    pub lat: Vec<f32>,
-    pub lon: Vec<f32>,
-    pub feature_names: Vec<String>,
-    pub num_features: usize,
-    /// Number of numeric features (enum features start at this index).
-    pub num_numeric: usize,
-    /// Row-major flat array: feature_data[row * num_features + feat_idx].
-    /// Quantized to u16. NaN sentinel = u16::MAX (65535).
-    /// Numeric features: encoded via (val - min) / range * 65534.
-    /// Enum features: stored directly as u16 cast of the f32 index.
-    pub feature_data: Vec<u16>,
-    /// Per-feature: range / QUANT_SCALE for fast decode.
-    dequant_a: Vec<f32>,
-    /// Per-feature: minimum value (offset for dequantization).
-    quant_min: Vec<f32>,
-    /// Per-feature: max - min (for encoding filter bounds).
-    quant_range: Vec<f32>,
-    pub feature_stats: Vec<FeatureStats>,
-    pub poi_metrics: PostcodePoiMetrics,
-    /// Unquantized last sale price used by the price-history chart.
-    last_known_price_raw: Vec<f32>,
-    /// Contiguous buffer holding all address strings end-to-end.
-    address_buffer: String,
-    /// Byte offset into `address_buffer` where each row's address starts.
-    address_offsets: Vec<u32>,
-    /// Length in bytes of each row's address.
-    address_lengths: Vec<u16>,
-    /// Interned postcodes: reader is thread-safe, keys index into it.
-    postcode_interner: lasso::RodeoReader,
-    postcode_keys: Vec<lasso::Spur>,
-    /// Rows for each postcode, keyed by the interned postcode key.
-    postcode_row_index: FxHashMap<lasso::Spur, Vec<u32>>,
-    /// Inverted index from address tokens to property rows.
-    address_token_index: FxHashMap<String, Vec<u32>>,
-    /// Prefix lookup from typed address-token prefix to indexed full address tokens.
-    address_prefix_index: FxHashMap<String, Vec<String>>,
-    /// Interned normalized address-search tokens used for per-row scoring.
-    address_search_interner: lasso::RodeoReader,
-    /// Flat per-row normalized address-search token keys.
-    address_search_token_keys: Vec<lasso::Spur>,
-    /// Offset into `address_search_token_keys` for each row.
-    address_search_token_offsets: Vec<u32>,
-    /// Number of normalized address-search token keys for each row.
-    address_search_token_lengths: Vec<u16>,
-    /// For enum features: maps feature index to list of possible string values.
-    /// Index in values list corresponds to the u16 value stored in feature_data.
-    pub enum_values: rustc_hash::FxHashMap<usize, Vec<String>>,
-    /// For enum features: maps feature index to per-value global counts (same order as enum_values).
-    pub enum_counts: rustc_hash::FxHashMap<usize, Vec<u64>>,
-    /// Per-row flag: true = construction date is approximate (from EPC band),
-    /// false = exact (from new-build transaction date).
-    /// Bit-packed: byte `row / 8`, bit `row % 8`. 8x smaller than Vec<bool>.
-    approx_build_date_bits: Vec<u8>,
-    /// Per-row renovation events. Keyed by (permuted) row index.
-    /// Only rows with events are present in the map.
-    renovation_history: FxHashMap<u32, Vec<RenovationEvent>>,
-    /// Per-row historical sale transactions (Land Registry price-paid).
-    /// Keyed by (permuted) row index. Only rows with prices are present.
-    historical_prices: FxHashMap<u32, Vec<HistoricalPrice>>,
-    property_sub_type: FxHashMap<u32, String>,
-    price_qualifier: FxHashMap<u32, String>,
-}
-
-impl PropertyData {
-    /// Get the address string for a given row.
-    pub fn address(&self, row: usize) -> &str {
-        let offset = self.address_offsets[row] as usize;
-        let length = self.address_lengths[row] as usize;
-        &self.address_buffer[offset..offset + length]
-    }
-
-    /// Get the postcode string for a given row.
-    pub fn postcode(&self, row: usize) -> &str {
-        self.postcode_interner.resolve(&self.postcode_keys[row])
-    }
-
-    /// Get postcode components for field-level borrowing (avoids conflicting borrows with feature_data).
-    pub fn postcode_parts(&self) -> (&lasso::RodeoReader, &[lasso::Spur]) {
-        (&self.postcode_interner, &self.postcode_keys)
-    }
-
-    /// Property rows for a given postcode string, or empty if unknown.
-    pub fn rows_for_postcode(&self, postcode: &str) -> &[u32] {
-        self.postcode_interner
-            .get(postcode)
-            .and_then(|key| self.postcode_row_index.get(&key))
-            .map(Vec::as_slice)
-            .unwrap_or(&[])
-    }
-
-    fn row_address_search_tokens(&self, row: usize) -> &[lasso::Spur] {
-        let offset = self.address_search_token_offsets[row] as usize;
-        let length = self.address_search_token_lengths[row] as usize;
-        &self.address_search_token_keys[offset..offset + length]
-    }
-
-    /// Search individual property addresses, returning `(row, score)` ranked best-first.
-    ///
-    /// Candidate rows come from intersecting the posting lists of the distinctive words the
-    /// user typed in full (so "Cherry Hinton Road" narrows to rows containing both), unioned
-    /// with the exact-postcode rows when a complete postcode is present (so a postcode is a
-    /// boost, not an all-or-nothing gate). An appended partial postcode keeps in-area rows
-    /// ahead of the candidate cut and adds a scoring bias. With a road-type word and no house
-    /// number, results collapse to one row per street.
-    pub fn search_addresses(&self, query: &str, limit: usize) -> Vec<(usize, i32)> {
-        if limit == 0 {
-            return Vec::new();
-        }
-
-        let parsed = parse_address_query(query);
-        if parsed.full_postcode.is_none()
-            && parsed.text_groups.is_empty()
-            && parsed.numeric_terms.is_empty()
-        {
-            return Vec::new();
-        }
-
-        let mut candidate_rows = self.address_candidate_rows(&parsed.candidate_terms);
-
-        // A complete postcode contributes its rows too, instead of replacing the road match.
-        if let Some(postcode) = parsed.full_postcode.as_deref() {
-            if let Some(rows) = self
-                .postcode_interner
-                .get(postcode)
-                .and_then(|key| self.postcode_row_index.get(&key))
-            {
-                candidate_rows = if candidate_rows.is_empty() {
-                    rows.clone()
-                } else {
-                    union_sorted(&candidate_rows, rows)
-                };
-            }
-        }
-
-        if candidate_rows.is_empty() {
-            return Vec::new();
-        }
-
-        // When the user appended a partial postcode, keep in-area rows ahead of the cut so the
-        // refinement still surfaces even for very common roads. Single pass (stable partition) so
-        // the postcode check — which allocates — runs exactly once per candidate.
-        if let Some(area) = parsed.postcode_area.as_deref() {
-            let mut in_area = Vec::new();
-            let mut others = Vec::new();
-            for &row in &candidate_rows {
-                if self.row_postcode_in_area(row as usize, area) {
-                    in_area.push(row);
-                } else {
-                    others.push(row);
-                }
-            }
-            in_area.extend(others);
-            candidate_rows = in_area;
-        }
-        candidate_rows.truncate(ADDRESS_SEARCH_CANDIDATE_LIMIT);
-
-        let mut scored: Vec<(i32, usize, usize)> = candidate_rows
-            .into_iter()
-            .filter_map(|row| {
-                let row = row as usize;
-                self.address_match_score(row, &parsed)
-                    .map(|score| (score, self.address(row).len(), row))
-            })
-            .collect();
-
-        scored.sort_unstable_by(|left, right| {
-            right
-                .0
-                .cmp(&left.0)
-                .then(left.1.cmp(&right.1))
-                .then(left.2.cmp(&right.2))
-        });
-
-        // Collapse a road browse (road-type word, no house number) to one row per street.
-        let collapse_streets = parsed.numeric_terms.is_empty() && query_has_road_type(query);
-
-        let mut seen = FxHashSet::default();
-        let mut results = Vec::with_capacity(limit);
-        for (score, _, row) in scored {
-            let address = self.address(row).trim();
-            if address.is_empty() {
-                continue;
-            }
-            let key = if collapse_streets {
-                format!(
-                    "{}\n{}",
-                    street_key(address),
-                    outcode_of(self.postcode(row))
-                )
-            } else {
-                format!("{}\n{}", address.to_ascii_lowercase(), self.postcode(row))
-            };
-            if !seen.insert(key) {
-                continue;
-            }
-            results.push((row, score));
-            if results.len() == limit {
-                break;
-            }
-        }
-
-        results
-    }
-
-    /// True when the row's postcode begins with the compact partial-postcode `area`
-    /// (e.g. "NW1" or "NW16" matches "NW1 6XE").
-    fn row_postcode_in_area(&self, row: usize, area: &str) -> bool {
-        let mut compact = String::new();
-        for ch in self.postcode(row).chars() {
-            if !ch.is_whitespace() {
-                compact.push(ch.to_ascii_uppercase());
-            }
-        }
-        compact.starts_with(area)
-    }
-
-    /// Candidate rows for the distinctive query words. Words typed in full intersect by their
-    /// exact posting lists (precise); a still-being-typed final word with no exact match seeds
-    /// from the smallest prefix-expanded posting list (so partial typing keeps working).
-    fn address_candidate_rows(&self, terms: &[String]) -> Vec<u32> {
-        let mut exact: Vec<&[u32]> = terms
-            .iter()
-            .filter_map(|term| self.address_token_index.get(term).map(Vec::as_slice))
-            .collect();
-
-        if !exact.is_empty() {
-            exact.sort_by_key(|rows| rows.len());
-            let mut acc = exact[0].to_vec();
-            for rows in &exact[1..] {
-                if acc.is_empty() {
-                    break;
-                }
-                acc = intersect_sorted(&acc, rows);
-            }
-            return acc;
-        }
-
-        self.prefix_seed_rows(terms)
-    }
-
-    /// Seed rows from the smallest prefix-expanded term — used only when no word matched an
-    /// indexed token exactly (i.e. the user is still typing the final word).
-    fn prefix_seed_rows(&self, terms: &[String]) -> Vec<u32> {
-        let mut best: Option<Vec<u32>> = None;
-        for term in terms {
-            if term.len() < ADDRESS_SEARCH_PREFIX_MIN_LEN {
-                continue;
-            }
-            let Some(tokens) = self.address_prefix_index.get(address_prefix_key(term)) else {
-                continue;
-            };
-            let mut union: Vec<u32> = Vec::new();
-            for token in tokens {
-                if !token.starts_with(term) {
-                    continue;
-                }
-                if let Some(rows) = self.address_token_index.get(token) {
-                    union = if union.is_empty() {
-                        rows.clone()
-                    } else {
-                        union_sorted(&union, rows)
-                    };
-                }
-            }
-            if !union.is_empty()
-                && best
-                    .as_ref()
-                    .is_none_or(|current| union.len() < current.len())
-            {
-                best = Some(union);
-            }
-        }
-        best.unwrap_or_default()
-    }
-
-    fn address_match_score(&self, row: usize, parsed: &AddressQuery) -> Option<i32> {
-        if self.address(row).trim().is_empty() {
-            return None;
-        }
-
-        let tokens = self.row_address_search_tokens(row);
-        if parsed
-            .text_groups
-            .iter()
-            .any(|group| !self.address_tokens_match_group(tokens, group))
-        {
-            return None;
-        }
-
-        let numeric_matches = parsed
-            .numeric_terms
-            .iter()
-            .filter(|term| {
-                tokens.iter().any(|token| {
-                    token_matches_numeric_term(self.address_search_interner.resolve(token), term)
-                })
-            })
-            .count();
-
-        if !parsed.numeric_terms.is_empty() && numeric_matches == 0 {
-            return None;
-        }
-
-        let mut score = 0;
-        if parsed.full_postcode.is_some() {
-            score += 1_000;
-        }
-        score += (parsed.text_groups.len() as i32) * 200;
-        score += (numeric_matches as i32) * 90;
-        if numeric_matches == parsed.numeric_terms.len() && numeric_matches > 0 {
-            score += 50;
-        }
-        // Additive bias (never a filter) when the row sits in the appended partial postcode.
-        if let Some(area) = parsed.postcode_area.as_deref() {
-            if self.row_postcode_in_area(row, area) {
-                score += 400;
-            }
-        }
-
-        Some(score)
-    }
-
-    fn address_tokens_match_group(&self, tokens: &[lasso::Spur], group: &AddressTermGroup) -> bool {
-        group.alternatives.iter().any(|alternative| {
-            tokens.iter().any(|token| {
-                token_matches_query_term(self.address_search_interner.resolve(token), alternative)
-            })
-        })
-    }
-
-    /// Get the is_approx_build_date flag for a given row (bit-packed).
-    pub fn is_approx_build_date(&self, row: usize) -> bool {
-        let byte = self.approx_build_date_bits[row / 8];
-        byte & (1 << (row % 8)) != 0
-    }
-
-    /// Get renovation events for a given row (empty slice if none).
-    pub fn renovation_history(&self, row: usize) -> &[RenovationEvent] {
-        self.renovation_history
-            .get(&(row as u32))
-            .map(|v| v.as_slice())
-            .unwrap_or(&[])
-    }
-
-    /// Get historical sale transactions for a given row (empty slice if none).
-    pub fn historical_prices(&self, row: usize) -> &[HistoricalPrice] {
-        self.historical_prices
-            .get(&(row as u32))
-            .map(|v| v.as_slice())
-            .unwrap_or(&[])
-    }
-
-    /// Get property sub-type for a given row.
-    pub fn property_sub_type(&self, row: usize) -> Option<&str> {
-        self.property_sub_type
-            .get(&(row as u32))
-            .map(String::as_str)
-    }
-
-    /// Get price qualifier for a given row.
-    pub fn price_qualifier(&self, row: usize) -> Option<&str> {
-        self.price_qualifier.get(&(row as u32)).map(String::as_str)
-    }
-
-    /// Get the unquantized last sale price for charting.
-    #[inline]
-    pub fn last_known_price_raw(&self, row: usize) -> f32 {
-        self.last_known_price_raw[row]
-    }
-
-    /// Decode a single feature value from quantized u16 storage.
-    #[inline]
-    pub fn get_feature(&self, row: usize, feat_idx: usize) -> f32 {
-        let raw = self.feature_data[row * self.num_features + feat_idx];
-        if raw == NAN_U16 {
-            return f32::NAN;
-        }
-        if feat_idx >= self.num_numeric {
-            raw as f32
-        } else {
-            raw as f32 * self.dequant_a[feat_idx] + self.quant_min[feat_idx]
-        }
-    }
-
-    /// Get a QuantRef for passing to aggregation/filter functions.
-    pub fn quant_ref(&self) -> QuantRef<'_> {
-        QuantRef {
-            dequant_a: &self.dequant_a,
-            quant_min: &self.quant_min,
-            quant_range: &self.quant_range,
-            num_numeric: self.num_numeric,
-        }
-    }
-}
-
-/// Compute a percentile from a uniformly-binned histogram.
-/// `prelim_counts` are uniform bins over [min, max].
-fn percentile_from_uniform_histogram(
-    count: usize,
-    min: f32,
-    max: f32,
-    prelim_counts: &[u64],
-    percentile: f32,
-) -> f32 {
-    if count == 0 || prelim_counts.is_empty() {
-        return min;
-    }
-    let target = (count as f64 * percentile as f64 / 100.0).floor() as u64;
-    let bin_width = (max - min) / prelim_counts.len() as f32;
-    let mut cumulative = 0u64;
-    for (i, &bin_count) in prelim_counts.iter().enumerate() {
-        let prev_cumulative = cumulative;
-        cumulative += bin_count;
-        if cumulative > target {
-            // Interpolate within this bin
-            let bin_start = min + i as f32 * bin_width;
-            let fraction = if bin_count > 0 {
-                (target - prev_cumulative) as f32 / bin_count as f32
-            } else {
-                0.0
-            };
-            return bin_start + fraction * bin_width;
-        }
-    }
-    max
-}
-
-/// Build a histogram and compute slider bounds based on the feature's Bounds config.
-pub fn compute_feature_stats(vals: &[f32], bounds: &Bounds, integer_bins: bool) -> FeatureStats {
-    // Single pass: min, max, count (skipping NaN and infinity)
-    let mut min = f32::INFINITY;
-    let mut max = f32::NEG_INFINITY;
-    let mut count = 0usize;
-    for &value in vals {
-        if value.is_finite() {
-            if value < min {
-                min = value;
-            }
-            if value > max {
-                max = value;
-            }
-            count += 1;
-        }
-    }
-
-    if count == 0 {
-        let (slider_min, slider_max) = match bounds {
-            Bounds::Fixed {
-                min: fmin,
-                max: fmax,
-            } => (*fmin, *fmax),
-            Bounds::Percentile { .. } => (0.0, 0.0),
-        };
-        return FeatureStats {
-            slider_min,
-            slider_max,
-            histogram: Histogram {
-                min: 0.0,
-                max: 0.0,
-                p1: 0.0,
-                p99: 0.0,
-                counts: vec![0; HISTOGRAM_BINS],
-            },
-        };
-    }
-
-    // Build preliminary histogram with uniform bins to compute percentiles
-    // Use full HISTOGRAM_BINS for percentile precision
-    let range = if max == min { 1.0 } else { max - min };
-    let prelim_max = min + range * (1.0 + 1e-6);
-    let prelim_bin_width = (prelim_max - min) / HISTOGRAM_BINS as f32;
-
-    let mut prelim_counts = vec![0u64; HISTOGRAM_BINS];
-    for &value in vals {
-        if value.is_finite() {
-            let bin = ((value - min) / prelim_bin_width) as usize;
-            prelim_counts[bin.min(HISTOGRAM_BINS - 1)] += 1;
-        }
-    }
-
-    // Compute p1 and p99 from preliminary histogram
-    let mut p1 = percentile_from_uniform_histogram(count, min, max, &prelim_counts, 1.0);
-    let mut p99 = percentile_from_uniform_histogram(count, min, max, &prelim_counts, 99.0);
-
-    // Iterative refinement for outlier-dominated distributions.
-    // When extreme outliers (e.g. 317M sqm from web scraping) dominate the range,
-    // the uniform histogram puts all real data in one bin, making percentile
-    // estimation useless. Zoom into the estimated data region and recompute.
-    let mut refined_counts = prelim_counts;
-    let mut refined_count = count;
-    let mut refined_min = min;
-    let mut refined_max = max;
-    for _ in 0..3 {
-        let iqr = p99 - p1;
-        if iqr <= 0.0 || (refined_max - refined_min) <= 5.0 * iqr {
-            break;
-        }
-        let new_min = (p1 - iqr).max(min);
-        let new_max = p99 + iqr;
-        if new_max <= new_min {
-            break;
-        }
-        let bin_width = (new_max - new_min) / HISTOGRAM_BINS as f32;
-        let mut counts = vec![0u64; HISTOGRAM_BINS];
-        let mut cnt = 0usize;
-        for &value in vals {
-            if value.is_finite() && value >= new_min && value <= new_max {
-                let bin = ((value - new_min) / bin_width) as usize;
-                counts[bin.min(HISTOGRAM_BINS - 1)] += 1;
-                cnt += 1;
-            }
-        }
-        if cnt == 0 {
-            break;
-        }
-        p1 = percentile_from_uniform_histogram(cnt, new_min, new_max, &counts, 1.0);
-        p99 = percentile_from_uniform_histogram(cnt, new_min, new_max, &counts, 99.0);
-        refined_counts = counts;
-        refined_count = cnt;
-        refined_min = new_min;
-        refined_max = new_max;
-    }
-
-    // For integer-binned features, snap p1/p99 to integer boundaries
-    // so each middle bin is exactly 1 unit wide.
-    if integer_bins {
-        p1 = p1.floor();
-        p99 = p99.ceil();
-    }
-
-    // Determine number of histogram bins
-    let num_bins = if integer_bins && p99 > p1 {
-        // One middle bin per integer + 2 outlier bins
-        (p99 - p1) as usize + 2
-    } else {
-        // Count unique values within the p1–p99 range to cap histogram bins.
-        // Using the full-range cardinality would over-allocate bins when outliers
-        // inflate it (e.g. bedrooms: 1–137 unique values but only ~10 within p1–p99).
-        let cardinality = {
-            let mut unique_set = rustc_hash::FxHashSet::default();
-            for &val in vals {
-                if val.is_finite() && val >= p1 && val <= p99 {
-                    unique_set.insert(val.to_bits());
-                }
-            }
-            unique_set.len()
-        };
-        HISTOGRAM_BINS.min(cardinality).max(3)
-    };
-
-    // Build final histogram with outlier bins at edges:
-    // - Bin 0: [min, p1) — low outliers
-    // - Bins 1 to n-2: [p1, p99) — main distribution, evenly divided
-    // - Bin n-1: [p99, max] — high outliers
-    let mut counts = vec![0u64; num_bins];
-    let middle_bins = num_bins.saturating_sub(2);
-    let middle_width = if middle_bins > 0 && p99 > p1 {
-        (p99 - p1) / middle_bins as f32
-    } else {
-        0.0
-    };
-
-    for &value in vals {
-        if value.is_finite() {
-            let bin = if value < p1 {
-                0 // Low outlier bin
-            } else if value >= p99 {
-                num_bins - 1 // High outlier bin
-            } else if middle_width > 0.0 {
-                // Middle bins (1 to n-2)
-                let middle_bin = ((value - p1) / middle_width) as usize;
-                (1 + middle_bin).min(num_bins - 2)
-            } else {
-                num_bins / 2 // Fallback if p1 == p99
-            };
-            counts[bin] += 1;
-        }
-    }
-
-    let histogram = Histogram {
-        min: refined_min,
-        max: refined_max,
-        p1,
-        p99,
-        counts,
-    };
-
-    // Compute slider bounds (use refined histogram for accurate percentiles)
-    let (slider_min, slider_max) = match bounds {
-        Bounds::Fixed {
-            min: fmin,
-            max: fmax,
-        } => (*fmin, *fmax),
-        Bounds::Percentile { low, high } => {
-            let p_low = percentile_from_uniform_histogram(
-                refined_count,
-                refined_min,
-                refined_max,
-                &refined_counts,
-                *low as f32,
-            );
-            let p_high = percentile_from_uniform_histogram(
-                refined_count,
-                refined_min,
-                refined_max,
-                &refined_counts,
-                *high as f32,
-            );
-            (p_low, p_high)
-        }
-    };
-
-    FeatureStats {
-        slider_min,
-        slider_max,
-        histogram,
-    }
-}
-
-fn column_to_f32_vec(column: &Column) -> anyhow::Result<Vec<f32>> {
-    let float_series = column
-        .cast(&DataType::Float32)
-        .context("Failed to cast column to Float32")?;
-    let chunked = float_series
-        .f32()
-        .context("Failed to get f32 chunked array")?;
-    Ok(chunked
-        .into_iter()
-        .map(|value| value.unwrap_or(f32::NAN))
-        .collect())
-}
-
-/// Precompute H3 cell IDs for all rows at the maximum resolution only.
-/// Parent cells for lower resolutions are derived on the fly via `CellIndex::parent()`.
-pub fn precompute_h3(lat: &[f32], lon: &[f32]) -> anyhow::Result<Vec<u64>> {
-    let res = H3_PRECOMPUTE_MAX;
-    tracing::info!("Precomputing H3 cells at resolution {}", res);
-
-    let h3_res =
-        h3o::Resolution::try_from(res).with_context(|| format!("Invalid H3 resolution: {res}"))?;
-
-    let cells: Vec<u64> = lat
-        .par_iter()
-        .zip(lon.par_iter())
-        .enumerate()
-        .map(|(i, (&latitude, &longitude))| {
-            let coord = h3o::LatLng::new(latitude as f64, longitude as f64).unwrap_or_else(|err| {
-                panic!(
-                    "Invalid coordinates at row {}: lat={}, lon={}: {}",
-                    i, latitude, longitude, err
-                )
-            });
-            u64::from(coord.to_cell(h3_res))
-        })
-        .collect();
-
-    tracing::info!("H3 precomputation complete ({} cells)", cells.len());
-    Ok(cells)
-}
-
-impl PropertyData {
-    pub fn load(properties_path: &Path, postcode_features_path: &Path) -> anyhow::Result<Self> {
-        super::run_polars_io(|| Self::load_inner(properties_path, postcode_features_path))
-    }
-
-    fn load_inner(properties_path: &Path, postcode_features_path: &Path) -> anyhow::Result<Self> {
-        // Load postcode.parquet
-        tracing::info!(
-            "Loading postcode features from {:?}",
-            postcode_features_path
-        );
-        let postcode_features_path = PlRefPath::try_from_path(postcode_features_path)
-            .context("Failed to normalize postcode parquet path")?;
-        let postcode_df = LazyFrame::scan_parquet(postcode_features_path, Default::default())
-            .context("Failed to scan postcode parquet")?
-            .collect()
-            .context("Failed to read postcode parquet")?;
-        tracing::info!(rows = postcode_df.height(), "Postcode features loaded");
-
-        let mut poi_metric_names: Vec<String> = postcode_df
-            .get_column_names()
-            .iter()
-            .map(|name| name.as_str())
-            .filter(|&name| features::is_dynamic_poi_feature(name))
-            .map(str::to_string)
-            .collect();
-        poi_metric_names.sort_by_key(|name| features::dynamic_poi_feature_sort_key(name));
-
-        let poi_metric_by_postcode: FxHashMap<String, u32> = if poi_metric_names.is_empty() {
-            FxHashMap::default()
-        } else {
-            let postcode_column = postcode_df
-                .column("Postcode")
-                .context("Postcode feature parquet missing 'Postcode' column")?
-                .str()
-                .context("'Postcode' column in postcode feature parquet is not a string")?;
-            postcode_column
-                .into_iter()
-                .enumerate()
-                .filter_map(|(idx, postcode)| {
-                    postcode.map(|postcode| (postcode.to_string(), idx as u32))
-                })
-                .collect()
-        };
-        let mut poi_metrics = PostcodePoiMetrics::from_postcode_df(&postcode_df, poi_metric_names)?;
-
-        // Load properties.parquet and join with postcode data lazily so the
-        // wide combined frame is never fully materialized — projection is
-        // pushed down into the join, keeping peak memory bounded.
-        tracing::info!("Loading properties from {:?}", properties_path);
-        let properties_path = PlRefPath::try_from_path(properties_path)
-            .context("Failed to normalize properties parquet path")?;
-        let properties_lf = LazyFrame::scan_parquet(properties_path, Default::default())
-            .context("Failed to scan properties parquet")?;
-        let combined_lf = properties_lf.join(
-            postcode_df.lazy(),
-            [col("Postcode")],
-            [col("Postcode")],
-            JoinArgs::new(JoinType::Left),
-        );
-
-        // Get configured feature/enum names in config order. Dynamic POI
-        // metrics live in a postcode-level side table so they do not widen the
-        // hot row-major property feature matrix.
-        let configured_numeric_names = features::all_numeric_feature_names();
-        let enum_names = features::all_enum_feature_names();
-
-        let schema = combined_lf
-            .clone()
-            .collect_schema()
-            .context("Failed to collect joined schema")?;
-        validate_no_england_rows_missing_coordinates(&combined_lf, &schema)?;
-        let numeric_names: Vec<String> = configured_numeric_names
-            .iter()
-            .map(|name| (*name).to_string())
-            .collect();
-
-        for name in &numeric_names {
-            match schema.get(name.as_str()) {
-                Some(dtype) if is_numeric_dtype(dtype) => {}
-                Some(dtype) => bail!(
-                    "Configured numeric feature '{}' has non-numeric type {:?}",
-                    name,
-                    dtype
-                ),
-                None => bail!(
-                    "Configured numeric feature '{}' not found in combined schema",
-                    name
-                ),
-            }
-        }
-        for name in &enum_names {
-            match schema.get(name) {
-                Some(dtype) if matches!(dtype, DataType::String) || dtype.is_categorical() => {}
-                Some(dtype) => bail!(
-                    "Configured enum feature '{}' has unexpected type {:?}",
-                    name,
-                    dtype
-                ),
-                None => bail!(
-                    "Configured enum feature '{}' not found in combined schema",
-                    name
-                ),
-            }
-        }
-
-        // Combine numeric and enum feature names (numeric first, then enum)
-        let feature_names: Vec<String> = numeric_names
-            .iter()
-            .map(|name| name.to_string())
-            .chain(enum_names.iter().map(|name| name.to_string()))
-            .collect();
-        let num_features = feature_names.len();
-        let num_numeric = numeric_names.len();
-        tracing::info!(
-            numeric = num_numeric,
-            enums = enum_names.len(),
-            total = num_features,
-            "Feature columns from config"
-        );
-
-        // Build select expressions for the combined DataFrame
-        let mut select_exprs: Vec<polars::prelude::Expr> = vec![];
-        select_exprs.push(col("lat").cast(DataType::Float32));
-        select_exprs.push(col("lon").cast(DataType::Float32));
-
-        // Select numeric features as Float32 (datetime columns → fractional year)
-        for name in &numeric_names {
-            if is_datetime_dtype(schema.get(name.as_str()).unwrap()) {
-                select_exprs.push(
-                    (col(name.as_str()).dt().year().cast(DataType::Float32)
-                        + (col(name.as_str()).dt().month().cast(DataType::Float32) - lit(1.0f32))
-                            / lit(12.0f32))
-                    .alias(name.as_str()),
-                );
-            } else {
-                select_exprs.push(col(name.as_str()).cast(DataType::Float32));
-            }
-        }
-
-        // String columns for address/postcode and property metadata
-        for &string_col_name in &[
-            "Address per Property Register",
-            "Address per EPC",
-            "Postcode",
-            "Property sub-type",
-            "Price qualifier",
-        ] {
-            if schema.get(string_col_name).is_some() {
-                select_exprs.push(col(string_col_name).cast(DataType::String));
-            }
-        }
-
-        // Enum features as String
-        for &name in &enum_names {
-            select_exprs.push(col(name).cast(DataType::String));
-        }
-
-        // Optional columns
-        let has_approx_col = schema.get("Is construction date approximate").is_some();
-        if has_approx_col {
-            select_exprs.push(col("Is construction date approximate").cast(DataType::Float32));
-        }
-        let has_renovation_history = schema.get("renovation_history").is_some();
-        if has_renovation_history {
-            select_exprs.push(col("renovation_history"));
-        }
-        let has_historical_prices = schema.get("historical_prices").is_some();
-        if has_historical_prices {
-            select_exprs.push(col("historical_prices"));
-        }
-        let df = combined_lf
-            .filter(col("lat").is_not_null().and(col("lon").is_not_null()))
-            .select(select_exprs)
-            .collect()
-            .context("Failed to select columns from joined frame")?;
-
-        let row_count = df.height();
-        if row_count == 0 {
-            bail!("No property rows have usable coordinates after joining postcode data");
-        }
-        tracing::info!(rows = row_count, "Combined data selected");
-
-        let lat_series = df
-            .column("lat")
-            .context("Missing 'lat' column")?
-            .cast(&DataType::Float32)
-            .context("Failed to cast 'lat' to Float32")?;
-        let lat: Vec<f32> = lat_series
-            .f32()
-            .context("Failed to read 'lat' as f32")?
-            .into_iter()
-            .map(|value| value.context("Missing 'lat' value after coordinate filter"))
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        let lon_series = df
-            .column("lon")
-            .context("Missing 'lon' column")?
-            .cast(&DataType::Float32)
-            .context("Failed to cast 'lon' to Float32")?;
-        let lon: Vec<f32> = lon_series
-            .f32()
-            .context("Failed to read 'lon' as f32")?
-            .into_iter()
-            .map(|value| value.context("Missing 'lon' value after coordinate filter"))
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        for (row, (&latitude, &longitude)) in lat.iter().zip(&lon).enumerate() {
-            if !(-90.0..=90.0).contains(&latitude) || !(-180.0..=180.0).contains(&longitude) {
-                bail!("Invalid coordinates at row {row}: lat={latitude}, lon={longitude}");
-            }
-        }
-
-        tracing::info!("Extracting numeric feature columns");
-        let numeric_col_major: Vec<Vec<f32>> = numeric_names
-            .par_iter()
-            .map(|name| {
-                let column = df
-                    .column(name.as_str())
-                    .with_context(|| format!("Missing feature column '{name}'"))?;
-                column_to_f32_vec(column)
-            })
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        tracing::info!("Computing histograms for numeric features");
-        let numeric_feature_stats: Vec<FeatureStats> = numeric_col_major
-            .par_iter()
-            .enumerate()
-            .map(|(feat_index, vals)| {
-                let name = numeric_names[feat_index].as_str();
-                let bounds = features::bounds_for(name)
-                    .with_context(|| format!("No bounds config for feature '{}'", name))?;
-                let stats = compute_feature_stats(vals, &bounds, features::has_integer_bins(name));
-                tracing::debug!(
-                    feature = %name,
-                    slider_min = format_args!("{:.2}", stats.slider_min),
-                    slider_max = format_args!("{:.2}", stats.slider_max),
-                    bins = stats.histogram.counts.len(),
-                    "Feature stats"
-                );
-                Ok(stats)
-            })
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        // Compute quantization parameters from feature stats (numeric features).
-        // For features with Fixed bounds, use those bounds so the full configured range
-        // is representable — the histogram refinement can narrow min/max to exclude
-        // "outliers" that are actually valid data (e.g. ethnicity percentages).
-        // For Percentile-bounded features, use the (possibly refined) histogram range
-        // so extreme outliers don't destroy precision for the main distribution.
-        let mut quant_min = Vec::with_capacity(num_features);
-        let mut quant_range = Vec::with_capacity(num_features);
-        for (feat_idx, stats) in numeric_feature_stats.iter().enumerate() {
-            let (min, max) = match features::bounds_for(numeric_names[feat_idx].as_str()) {
-                Some(Bounds::Fixed { min, max }) => (min, max),
-                _ => (stats.histogram.min, stats.histogram.max),
-            };
-            quant_min.push(min);
-            quant_range.push(if max > min { max - min } else { 0.0 });
-        }
-
-        tracing::info!("Extracting string columns");
-        let extract_string_col = |df: &DataFrame, name: &str| -> anyhow::Result<Vec<String>> {
-            let column = df
-                .column(name)
-                .with_context(|| format!("Required column '{name}' not found in parquet"))?;
-            let string_column = column
-                .str()
-                .with_context(|| format!("Column '{name}' is not a string column"))?;
-            string_column
-                .into_iter()
-                .enumerate()
-                .map(|(row, value)| {
-                    value
-                        .map(ToString::to_string)
-                        .with_context(|| format!("Required column '{name}' has null at row {row}"))
-                })
-                .collect()
-        };
-
-        let address_raw = extract_string_col(&df, "Address per Property Register")?;
-        let postcode_raw = extract_string_col(&df, "Postcode")?;
-
-        // Extract optional string columns
-        let extract_optional_string_col =
-            |df: &DataFrame, name: &str| -> anyhow::Result<Vec<Option<String>>> {
-                if let Ok(column) = df.column(name) {
-                    let string_column = column
-                        .str()
-                        .with_context(|| format!("Column '{name}' is not a string column"))?;
-                    Ok(string_column
-                        .into_iter()
-                        .map(|value| {
-                            value.and_then(|s| {
-                                let trimmed = s.trim();
-                                if trimmed.is_empty() {
-                                    None
-                                } else {
-                                    Some(trimmed.to_string())
-                                }
-                            })
-                        })
-                        .collect())
-                } else {
-                    Ok(vec![None; row_count])
-                }
-            };
-        let property_sub_type_raw = extract_optional_string_col(&df, "Property sub-type")?;
-        let price_qualifier_raw = extract_optional_string_col(&df, "Price qualifier")?;
-
-        tracing::info!("Building enum features");
-        // enum_col_major: Vec<(values_list, encoded_as_f32)>
-        let enum_col_major: Vec<(Vec<String>, Vec<f32>)> = enum_names
-            .par_iter()
-            .map(|&name| -> anyhow::Result<(Vec<String>, Vec<f32>)> {
-                let column_data = df
-                    .column(name)
-                    .with_context(|| format!("Required enum column '{name}' not found"))?;
-                let string_column = column_data
-                    .str()
-                    .with_context(|| format!("Enum column '{name}' is not a string column"))?;
-                let unique_set: std::collections::HashSet<String> = string_column
-                    .into_iter()
-                    .filter_map(|value| {
-                        let text = value?.trim();
-                        (!text.is_empty()).then(|| text.to_string())
-                    })
-                    .collect();
-
-                // Use configured order if available, otherwise alphabetical
-                let unique: Vec<String> = if let Some(order) = features::order_for(name) {
-                    let mut ordered: Vec<String> = Vec::new();
-                    for &ordered_value in order {
-                        if unique_set.contains(ordered_value) {
-                            ordered.push(ordered_value.to_string());
-                        }
-                    }
-                    // Append any values not in the configured order, alphabetically
-                    // Use HashSet for O(1) contains instead of O(n) slice search
-                    let order_set: rustc_hash::FxHashSet<&str> = order.iter().copied().collect();
-                    let mut remainder: Vec<String> = unique_set
-                        .iter()
-                        .filter(|value| !order_set.contains(value.as_str()))
-                        .cloned()
-                        .collect();
-                    remainder.sort();
-                    ordered.extend(remainder);
-                    ordered
-                } else {
-                    let mut sorted: Vec<String> = unique_set.into_iter().collect();
-                    sorted.sort();
-                    sorted
-                };
-
-                let value_to_idx: std::collections::HashMap<&str, f32> = unique
-                    .iter()
-                    .enumerate()
-                    .map(|(index, value)| (value.as_str(), index as f32))
-                    .collect();
-
-                let encoded: Vec<f32> = string_column
-                    .into_iter()
-                    .enumerate()
-                    .map(|(row, value)| {
-                        let Some(text) = value.map(str::trim).filter(|text| !text.is_empty())
-                        else {
-                            return Ok(f32::NAN);
-                        };
-                        value_to_idx.get(text).copied().with_context(|| {
-                            format!("Enum column '{name}' has unknown value '{text}' at row {row}")
-                        })
-                    })
-                    .collect::<anyhow::Result<Vec<_>>>()?;
-
-                tracing::debug!(column = %name, unique_values = unique.len(), "Enum feature encoded as f32");
-                Ok((unique, encoded))
-            })
-            .collect::<anyhow::Result<Vec<_>>>()?;
-
-        // Extract is_approx_build_date: 0.0 = exact, anything else (1.0/NaN) = approximate
-        let is_approx_build_date_raw: Vec<bool> = if has_approx_col {
-            let column_data = df
-                .column("Is construction date approximate")
-                .context("Missing 'Is construction date approximate' column")?;
-            let float_series = column_data
-                .cast(&DataType::Float32)
-                .context("Failed to cast 'Is construction date approximate' to Float32")?;
-            let chunked = float_series
-                .f32()
-                .context("Failed to read 'Is construction date approximate' as f32")?;
-            chunked
-                .into_iter()
-                .map(|value| match value {
-                    Some(0.0) => false,
-                    _ => true, // 1.0 or NaN → approximate
-                })
-                .collect()
-        } else {
-            vec![true; row_count] // default: all approximate
-        };
-
-        // Extract renovation_history: List<Struct{year: i32, event: str}>
-        let mut renovation_raw: FxHashMap<u32, Vec<RenovationEvent>> = if has_renovation_history {
-            tracing::info!("Extracting renovation history");
-            let reno_col = df
-                .column("renovation_history")
-                .context("Missing renovation_history column")?;
-            let list_ca = reno_col
-                .list()
-                .context("renovation_history is not a list column")?;
-
-            let mut history: FxHashMap<u32, Vec<RenovationEvent>> = FxHashMap::default();
-            for old_row in 0..row_count {
-                if let Some(inner) = list_ca.get_as_series(old_row) {
-                    if inner.is_empty() {
-                        continue;
-                    }
-                    let structs = inner
-                        .struct_()
-                        .context("renovation_history inner is not a struct")?;
-                    let years = structs
-                        .field_by_name("year")
-                        .context("Missing 'year' field in renovation_history struct")?;
-                    let events = structs
-                        .field_by_name("event")
-                        .context("Missing 'event' field in renovation_history struct")?;
-
-                    let mut row_events = Vec::new();
-                    for idx in 0..inner.len() {
-                        let year = years.get(idx).context("Failed to get year value")?;
-                        let event = events.get(idx).context("Failed to get event value")?;
-                        if let (AnyValue::Int32(yr), AnyValue::String(ev)) = (&year, &event) {
-                            row_events.push(RenovationEvent {
-                                year: *yr,
-                                event: ev.to_string(),
-                            });
-                        }
-                    }
-                    if !row_events.is_empty() {
-                        history.insert(old_row as u32, row_events);
-                    }
-                }
-            }
-            tracing::info!(
-                properties_with_events = history.len(),
-                "Renovation history extracted"
-            );
-            history
-        } else {
-            FxHashMap::default()
-        };
-
-        // Extract historical_prices: List<Struct{year: i32, month: u8, price: i64}>
-        let mut historical_prices_raw: FxHashMap<u32, Vec<HistoricalPrice>> =
-            if has_historical_prices {
-                tracing::info!("Extracting historical prices");
-                let prices_col = df
-                    .column("historical_prices")
-                    .context("Missing historical_prices column")?;
-                let list_ca = prices_col
-                    .list()
-                    .context("historical_prices is not a list column")?;
-
-                let mut history: FxHashMap<u32, Vec<HistoricalPrice>> = FxHashMap::default();
-                for old_row in 0..row_count {
-                    if let Some(inner) = list_ca.get_as_series(old_row) {
-                        if inner.is_empty() {
-                            continue;
-                        }
-                        let structs = inner
-                            .struct_()
-                            .context("historical_prices inner is not a struct")?;
-                        let years = structs
-                            .field_by_name("year")
-                            .context("Missing 'year' field in historical_prices struct")?;
-                        let months = structs
-                            .field_by_name("month")
-                            .context("Missing 'month' field in historical_prices struct")?;
-                        let prices = structs
-                            .field_by_name("price")
-                            .context("Missing 'price' field in historical_prices struct")?;
-
-                        let mut row_prices = Vec::new();
-                        for idx in 0..inner.len() {
-                            let year = years.get(idx).context("Failed to get year value")?;
-                            let month = months.get(idx).context("Failed to get month value")?;
-                            let price = prices.get(idx).context("Failed to get price value")?;
-                            let AnyValue::Int32(year_i32) = year else {
-                                bail!("historical_prices.year is not Int32 at row {old_row}, got {year:?}");
-                            };
-                            let AnyValue::UInt8(month_u8) = month else {
-                                bail!("historical_prices.month is not UInt8 at row {old_row}, got {month:?}");
-                            };
-                            let AnyValue::Int64(price_i64) = price else {
-                                bail!("historical_prices.price is not Int64 at row {old_row}, got {price:?}");
-                            };
-                            row_prices.push(HistoricalPrice {
-                                year: year_i32,
-                                month: month_u8,
-                                price: price_i64,
-                            });
-                        }
-                        if !row_prices.is_empty() {
-                            history.insert(old_row as u32, row_prices);
-                        }
-                    }
-                }
-                tracing::info!(
-                    properties_with_prices = history.len(),
-                    "Historical prices extracted"
-                );
-                history
-            } else {
-                FxHashMap::default()
-            };
-
-        // Free the projected joined frame before building the row-major matrix.
-        drop(df);
-
-        // Sort all rows by spatial locality so that grid queries access
-        // contiguous memory (sequential reads instead of random DRAM accesses).
-        tracing::info!("Sorting rows by spatial locality");
-        let grid_cell_size = 0.01_f32;
-        let min_lat_val = lat.iter().cloned().fold(f32::INFINITY, f32::min) - grid_cell_size;
-        let min_lon_val = lon.iter().cloned().fold(f32::INFINITY, f32::min) - grid_cell_size;
-        let max_lon_val = lon.iter().cloned().fold(f32::NEG_INFINITY, f32::max) + grid_cell_size;
-        let grid_cols = ((max_lon_val - min_lon_val) / grid_cell_size).ceil() as u64 + 1;
-
-        let mut perm: Vec<u32> = (0..row_count as u32).collect();
-        perm.par_sort_unstable_by_key(|&perm_index| {
-            let grid_row = ((lat[perm_index as usize] - min_lat_val) / grid_cell_size) as u64;
-            let grid_col = ((lon[perm_index as usize] - min_lon_val) / grid_cell_size) as u64;
-            grid_row * grid_cols + grid_col
-        });
-
-        let lat: Vec<f32> = perm
-            .iter()
-            .map(|&perm_index| lat[perm_index as usize])
-            .collect();
-        let lon: Vec<f32> = perm
-            .iter()
-            .map(|&perm_index| lon[perm_index as usize])
-            .collect();
-        let last_known_price_raw: Vec<f32> = numeric_names
-            .iter()
-            .position(|name| name == "Last known price")
-            .map(|price_idx| {
-                perm.iter()
-                    .map(|&perm_index| numeric_col_major[price_idx][perm_index as usize])
-                    .collect()
-            })
-            .context("Required numeric column 'Last known price' not configured")?;
-
-        // Build contiguous address buffer and address search index (permuted)
-        tracing::info!("Building interned strings");
-        let total_addr_bytes: usize = address_raw.iter().map(|text| text.len()).sum();
-        let mut address_buffer = String::with_capacity(total_addr_bytes);
-        let mut address_offsets = Vec::with_capacity(row_count);
-        let mut address_lengths = Vec::with_capacity(row_count);
-        let mut address_token_index: FxHashMap<String, Vec<u32>> = FxHashMap::default();
-        let mut address_search_rodeo = lasso::Rodeo::default();
-        let mut address_search_token_keys: Vec<lasso::Spur> = Vec::new();
-        let mut address_search_token_offsets = Vec::with_capacity(row_count);
-        let mut address_search_token_lengths = Vec::with_capacity(row_count);
-        for (new_row, &perm_index) in perm.iter().enumerate() {
-            let addr = &address_raw[perm_index as usize];
-            let offset = address_buffer.len() as u32;
-            let length = addr.len().min(u16::MAX as usize) as u16;
-            address_offsets.push(offset);
-            address_lengths.push(length);
-            address_buffer.push_str(&addr[..length as usize]);
-
-            let search_tokens = address_search_tokens(addr);
-            let token_offset = address_search_token_keys.len() as u32;
-            let token_length = search_tokens.len().min(u16::MAX as usize) as u16;
-            address_search_token_offsets.push(token_offset);
-            address_search_token_lengths.push(token_length);
-
-            for token in search_tokens.iter().take(token_length as usize) {
-                let key = address_search_rodeo.get_or_intern(token);
-                address_search_token_keys.push(key);
-
-                if is_address_candidate_token(token) {
-                    address_token_index
-                        .entry(token.clone())
-                        .or_default()
-                        .push(new_row as u32);
-                }
-            }
-        }
-        // Keep every distinctive token: common road words ("high", "church", "station") are
-        // exactly what people search, and dropping them made those roads unsearchable while a
-        // prefix fallback surfaced the wrong street ("Highbury" for "High"). The candidate scan
-        // is bounded per query instead (ADDRESS_SEARCH_CANDIDATE_LIMIT), and stop words are
-        // already excluded from the index, so the largest posting lists stay modest.
-        let max_postings = address_token_index
-            .values()
-            .map(Vec::len)
-            .max()
-            .unwrap_or(0);
-        let address_prefix_index = build_address_prefix_index(&address_token_index);
-        let address_search_interner = address_search_rodeo.into_reader();
-        let address_postings_count: usize = address_token_index.values().map(Vec::len).sum();
-        tracing::info!(
-            tokens = address_token_index.len(),
-            prefixes = address_prefix_index.len(),
-            max_postings_per_token = max_postings,
-            postings = address_postings_count,
-            row_tokens = address_search_token_keys.len(),
-            "Address search index built"
-        );
-
-        // Intern postcodes (permuted)
-        let mut postcode_rodeo = lasso::Rodeo::default();
-        let mut postcode_keys: Vec<lasso::Spur> = Vec::with_capacity(row_count);
-        let mut postcode_row_index: FxHashMap<lasso::Spur, Vec<u32>> = FxHashMap::default();
-        for (new_row, &perm_index) in perm.iter().enumerate() {
-            let key = postcode_rodeo.get_or_intern(&postcode_raw[perm_index as usize]);
-            postcode_keys.push(key);
-            postcode_row_index
-                .entry(key)
-                .or_default()
-                .push(new_row as u32);
-        }
-        let postcode_interner = postcode_rodeo.into_reader();
-
-        let row_to_poi_metric_idx: Vec<u32> = if poi_metrics.is_empty() {
-            vec![NO_POI_METRIC_ROW; row_count]
-        } else {
-            perm.iter()
-                .map(|&old_row| {
-                    poi_metric_by_postcode
-                        .get(postcode_raw[old_row as usize].as_str())
-                        .copied()
-                        .unwrap_or(NO_POI_METRIC_ROW)
-                })
-                .collect()
-        };
-        poi_metrics.set_row_mapping(row_to_poi_metric_idx);
-
-        // Pack is_approx_build_date into a bitvec (8 bools per byte)
-        let num_bytes = row_count.div_ceil(8);
-        let mut approx_build_date_bits = vec![0u8; num_bytes];
-        for (new_row, &old_row) in perm.iter().enumerate() {
-            if is_approx_build_date_raw[old_row as usize] {
-                approx_build_date_bits[new_row / 8] |= 1 << (new_row % 8);
-            }
-        }
-
-        // Re-key renovation_history by permuted row index
-        let renovation_history: FxHashMap<u32, Vec<RenovationEvent>> = {
-            let mut map =
-                FxHashMap::with_capacity_and_hasher(renovation_raw.len(), Default::default());
-            for (new_row, &old_row) in perm.iter().enumerate() {
-                if let Some(events) = renovation_raw.remove(&old_row) {
-                    map.insert(new_row as u32, events);
-                }
-            }
-            map
-        };
-
-        // Re-key historical_prices by permuted row index
-        let historical_prices: FxHashMap<u32, Vec<HistoricalPrice>> = {
-            let mut map = FxHashMap::with_capacity_and_hasher(
-                historical_prices_raw.len(),
-                Default::default(),
-            );
-            for (new_row, &old_row) in perm.iter().enumerate() {
-                if let Some(prices) = historical_prices_raw.remove(&old_row) {
-                    map.insert(new_row as u32, prices);
-                }
-            }
-            map
-        };
-
-        // Permute optional string columns into sparse HashMaps
-        let property_sub_type: FxHashMap<u32, String> = {
-            let mut map = FxHashMap::default();
-            for (new_row, &old_row) in perm.iter().enumerate() {
-                if let Some(ref s) = property_sub_type_raw[old_row as usize] {
-                    map.insert(new_row as u32, s.clone());
-                }
-            }
-            map
-        };
-        let price_qualifier: FxHashMap<u32, String> = {
-            let mut map = FxHashMap::default();
-            for (new_row, &old_row) in perm.iter().enumerate() {
-                if let Some(ref s) = price_qualifier_raw[old_row as usize] {
-                    map.insert(new_row as u32, s.clone());
-                }
-            }
-            map
-        };
-
-        // Build enum_values map: feature_index -> list of string values
-        // and enum_counts map: feature_index -> per-value global counts
-        let mut enum_values: rustc_hash::FxHashMap<usize, Vec<String>> =
-            rustc_hash::FxHashMap::default();
-        let mut enum_counts: rustc_hash::FxHashMap<usize, Vec<u64>> =
-            rustc_hash::FxHashMap::default();
-        for (enum_idx, (values, encoded)) in enum_col_major.iter().enumerate() {
-            let feature_idx = num_numeric + enum_idx;
-            enum_values.insert(feature_idx, values.clone());
-            let mut counts = vec![0u64; values.len()];
-            for &val in encoded {
-                if val.is_finite() {
-                    let idx = val as usize;
-                    if idx < counts.len() {
-                        counts[idx] += 1;
-                    }
-                }
-            }
-            enum_counts.insert(feature_idx, counts);
-        }
-
-        // Build feature_stats: numeric stats + placeholder stats for enums
-        let mut feature_stats = numeric_feature_stats;
-        for (values, _) in &enum_col_major {
-            // For enum features, slider range is 0 to num_values-1
-            let num_values = values.len();
-            let max_val = num_values as f32;
-            feature_stats.push(FeatureStats {
-                slider_min: 0.0,
-                slider_max: (num_values.saturating_sub(1)) as f32,
-                histogram: Histogram {
-                    min: 0.0,
-                    max: max_val,
-                    p1: 0.0,
-                    p99: max_val,
-                    counts: vec![0; num_values.max(1)],
-                },
-            });
-            // Enum features: not quantized, stored directly as u16
-            quant_min.push(0.0);
-            quant_range.push(0.0);
-        }
-        let dequant_a: Vec<f32> = quant_range
-            .iter()
-            .map(|&r| if r > 0.0 { r / QUANT_SCALE } else { 0.0 })
-            .collect();
-
-        // Transpose to row-major AND apply spatial permutation in one pass.
-        // Combines numeric and enum features into a single feature_data array, quantized to u16.
-        tracing::info!("Transposing to row-major layout (spatially sorted, quantized to u16)");
-        let mut feature_data = vec![NAN_U16; row_count * num_features];
-        feature_data
-            .par_chunks_mut(num_features)
-            .enumerate()
-            .for_each(|(new_row, row_slice)| {
-                let old_index = perm[new_row] as usize;
-                // Numeric features: quantize to u16
-                for (feat_idx, col_vec) in numeric_col_major.iter().enumerate() {
-                    let value = col_vec[old_index];
-                    row_slice[feat_idx] = if value.is_finite() {
-                        let range = quant_range[feat_idx];
-                        if range > 0.0 {
-                            let normalized = (value - quant_min[feat_idx]) / range;
-                            (normalized * QUANT_SCALE).round().clamp(0.0, QUANT_SCALE) as u16
-                        } else {
-                            0
-                        }
-                    } else {
-                        NAN_U16
-                    };
-                }
-                // Enum features: store as u16 directly
-                for (enum_idx, (_, encoded)) in enum_col_major.iter().enumerate() {
-                    let value = encoded[old_index];
-                    row_slice[num_numeric + enum_idx] = if value.is_finite() {
-                        value as u16
-                    } else {
-                        NAN_U16
-                    };
-                }
-            });
-
-        tracing::info!("Data loading complete");
-
-        Ok(PropertyData {
-            lat,
-            lon,
-            feature_names,
-            num_features,
-            num_numeric,
-            feature_data,
-            dequant_a,
-            quant_min,
-            quant_range,
-            feature_stats,
-            poi_metrics,
-            last_known_price_raw,
-            address_buffer,
-            address_offsets,
-            address_lengths,
-            postcode_interner,
-            postcode_keys,
-            postcode_row_index,
-            address_token_index,
-            address_prefix_index,
-            address_search_interner,
-            address_search_token_keys,
-            address_search_token_offsets,
-            address_search_token_lengths,
-            enum_values,
-            enum_counts,
-            approx_build_date_bits,
-            renovation_history,
-            historical_prices,
-            property_sub_type,
-            price_qualifier,
-        })
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::features::Bounds;
-
-    fn make_fixed_bounds(min: f32, max: f32) -> Bounds {
-        Bounds::Fixed { min, max }
-    }
-
-    fn make_percentile_bounds(low: f64, high: f64) -> Bounds {
-        Bounds::Percentile { low, high }
-    }
-
-    #[test]
-    fn country_column_detection_prefers_reliable_country_code() {
-        let df = df!(
-            "Postcode" => &["SW1A 1AA"],
-            "ctry25cd" => &["E92000001"],
-            "lat" => &[Some(51.501_f64)],
-            "lon" => &[Some(-0.141_f64)],
-        )
-        .expect("test dataframe should build");
-
-        assert_eq!(
-            find_country_column(df.schema()).as_deref(),
-            Some("ctry25cd")
-        );
-    }
-
-    #[test]
-    fn missing_coordinate_summary_counts_rows_and_distinct_samples() {
-        let summary = summarize_missing_coordinate_postcodes([
-            Some("SW1A 1AA"),
-            Some("SW1A 1AA"),
-            Some("E14 2DG"),
-            Some(""),
-            None,
-        ]);
-
-        assert_eq!(
-            summary,
-            MissingCoordinateSummary {
-                row_count: 5,
-                unique_postcode_count: 2,
-                sample_postcodes: vec!["E14 2DG".to_string(), "SW1A 1AA".to_string()],
-            }
-        );
-    }
-
-    #[test]
-    fn missing_england_coordinates_error_includes_counts_and_samples() {
-        let summary = MissingCoordinateSummary {
-            row_count: 3,
-            unique_postcode_count: 2,
-            sample_postcodes: vec!["E14 2DG".to_string(), "SW1A 1AA".to_string()],
-        };
-
-        let message = missing_england_coordinates_error(&summary, "ctry25cd");
-
-        assert!(message.contains("3 rows across 2 postcodes"));
-        assert!(message.contains("country column 'ctry25cd'"));
-        assert!(message.contains("E14 2DG, SW1A 1AA"));
-    }
-
-    #[test]
-    fn coordinate_validation_errors_for_england_rows() {
-        let lf = df!(
-            "Postcode" => &["SW1A 1AA", "E14 2DG"],
-            "ctry25cd" => &["E92000001", "E92000001"],
-            "lat" => &[Some(51.501_f64), None],
-            "lon" => &[Some(-0.141_f64), Some(-0.001_f64)],
-        )
-        .expect("test dataframe should build")
-        .lazy();
-        let schema = lf.clone().collect_schema().expect("schema should collect");
-
-        let err = validate_no_england_rows_missing_coordinates(&lf, &schema)
-            .expect_err("England row with missing coordinate should error");
-        let message = err.to_string();
-
-        assert!(message.contains("1 rows across 1 postcodes"));
-        assert!(message.contains("E14 2DG"));
-    }
-
-    #[test]
-    fn coordinate_validation_allows_non_england_rows() {
-        let lf = df!(
-            "Postcode" => &["CF10 1AA", "SW1A 1AA"],
-            "ctry25cd" => &["W92000004", "E92000001"],
-            "lat" => &[None, Some(51.501_f64)],
-            "lon" => &[Some(-3.179_f64), Some(-0.141_f64)],
-        )
-        .expect("test dataframe should build")
-        .lazy();
-        let schema = lf.clone().collect_schema().expect("schema should collect");
-
-        validate_no_england_rows_missing_coordinates(&lf, &schema)
-            .expect("non-England row with missing coordinate should be skipped");
-    }
-
-    #[test]
-    fn coordinate_validation_requires_country_column() {
-        let lf = df!(
-            "Postcode" => &["SW1A 1AA"],
-            "lat" => &[None::<f64>],
-            "lon" => &[Some(-0.141_f64)],
-        )
-        .expect("test dataframe should build")
-        .lazy();
-        let schema = lf.clone().collect_schema().expect("schema should collect");
-
-        let err = validate_no_england_rows_missing_coordinates(&lf, &schema)
-            .expect_err("missing country provenance should error");
-
-        assert!(err.to_string().contains("no reliable country column"));
-    }
-
-    #[test]
-    fn full_postcode_detection_accepts_common_formats() {
-        assert!(is_full_postcode_compact("SW1A1AA"));
-        assert!(is_full_postcode_compact("E142DG"));
-        assert!(is_full_postcode_compact("M11AE"));
-        assert!(!is_full_postcode_compact("E14"));
-        assert!(!is_full_postcode_compact("DOWNING"));
-        assert!(!is_full_postcode_compact("10A"));
-    }
-
-    #[test]
-    fn address_query_parsing_skips_postcodes_and_street_suffixes() {
-        let parsed = parse_address_query("Flat 2, 10 Downing St, SW1A 2AA");
-
-        assert_eq!(parsed.full_postcode.as_deref(), Some("SW1A 2AA"));
-        assert_eq!(
-            parsed.numeric_terms,
-            vec!["10".to_string(), "2".to_string()]
-        );
-        assert_eq!(parsed.candidate_terms, vec!["downing".to_string()]);
-        assert_eq!(parsed.text_groups.len(), 1);
-        assert_eq!(
-            parsed.text_groups[0].alternatives,
-            vec!["downing".to_string()]
-        );
-    }
-
-    #[test]
-    fn address_query_parsing_handles_compact_postcodes() {
-        let parsed = parse_address_query("10 downing street sw1a1aa");
-
-        assert_eq!(parsed.full_postcode.as_deref(), Some("SW1A 1AA"));
-        assert_eq!(parsed.numeric_terms, vec!["10".to_string()]);
-        assert_eq!(parsed.candidate_terms, vec!["downing".to_string()]);
-    }
-
-    #[test]
-    fn address_query_recovers_appended_partial_postcode_as_bias() {
-        let parsed = parse_address_query("Baker Street NW1");
-        assert_eq!(parsed.full_postcode, None);
-        assert_eq!(parsed.postcode_area.as_deref(), Some("NW1"));
-        // The road words are still searchable; the postcode fragment did not consume them.
-        assert_eq!(parsed.candidate_terms, vec!["baker".to_string()]);
-        assert!(parsed.numeric_terms.is_empty());
-    }
-
-    #[test]
-    fn address_query_recovers_outcode_plus_sector_without_a_phantom_house_number() {
-        let parsed = parse_address_query("High Street CR0 2");
-        assert_eq!(parsed.postcode_area.as_deref(), Some("CR02"));
-        // The lone sector digit must not be treated as a house number.
-        assert!(parsed.numeric_terms.is_empty());
-        assert_eq!(parsed.candidate_terms, vec!["high".to_string()]);
-    }
-
-    #[test]
-    fn full_postcode_takes_precedence_over_partial_bias() {
-        let parsed = parse_address_query("Baker Street NW1 6XE");
-        assert_eq!(parsed.full_postcode.as_deref(), Some("NW1 6XE"));
-        assert_eq!(parsed.postcode_area, None);
-    }
-
-    #[test]
-    fn intersect_and_union_sorted_row_ids() {
-        assert_eq!(
-            intersect_sorted(&[1, 2, 3, 5], &[2, 3, 4, 5]),
-            vec![2, 3, 5]
-        );
-        assert_eq!(intersect_sorted(&[1, 2], &[3, 4]), Vec::<u32>::new());
-        assert_eq!(union_sorted(&[1, 3, 5], &[2, 3, 4]), vec![1, 2, 3, 4, 5]);
-        assert_eq!(union_sorted(&[], &[2, 4]), vec![2, 4]);
-    }
-
-    #[test]
-    fn street_key_collapses_house_numbers_and_flats() {
-        assert_eq!(street_key("12 Baker Street"), "baker street");
-        assert_eq!(street_key("5 Baker Street"), "baker street");
-        assert_eq!(street_key("Flat 2, 10 Downing Street"), "downing street");
-        assert_eq!(street_key("221B Baker Street"), "baker street");
-    }
-
-    #[test]
-    fn street_key_keeps_ordinal_street_names() {
-        // Ordinals are part of the street name, not a house-number prefix.
-        assert_eq!(street_key("2nd Avenue"), "2nd avenue");
-        assert_eq!(street_key("12 3rd Avenue"), "3rd avenue");
-        assert!(is_ordinal_token("21st"));
-        assert!(!is_ordinal_token("21"));
-        assert!(!is_ordinal_token("221b"));
-    }
-
-    #[test]
-    fn postcode_area_recovered_only_from_the_trailing_position() {
-        // A leading road designation must NOT be taken as an area refinement.
-        let parsed = parse_address_query("A4 Great West Road");
-        assert_eq!(parsed.postcode_area, None);
-        // A genuine trailing outcode still is.
-        let trailing = parse_address_query("Great West Road W4");
-        assert_eq!(trailing.postcode_area.as_deref(), Some("W4"));
-    }
-
-    #[test]
-    fn road_type_detection() {
-        assert!(query_has_road_type("high street"));
-        assert!(query_has_road_type("acacia avenue"));
-        assert!(!query_has_road_type("acacia"));
-        assert!(!query_has_road_type("london"));
-    }
-
-    #[test]
-    fn address_query_parsing_keeps_partial_terms_for_row_matching() {
-        let parsed = parse_address_query("settlers cour");
-
-        assert_eq!(parsed.full_postcode, None);
-        assert_eq!(parsed.numeric_terms, Vec::<String>::new());
-        assert_eq!(
-            parsed.candidate_terms,
-            vec!["settlers".to_string(), "cour".to_string()]
-        );
-        assert_eq!(parsed.text_groups.len(), 2);
-        assert_eq!(
-            parsed.text_groups[0].alternatives,
-            vec!["settlers".to_string()]
-        );
-        assert_eq!(parsed.text_groups[1].alternatives, vec!["cour".to_string()]);
-    }
-
-    #[test]
-    fn address_search_tokens_keep_actual_address_terms_for_scoring() {
-        let tokens = address_search_tokens("Flat 2, 10 Downing Cour");
-
-        assert_eq!(
-            tokens,
-            vec![
-                "10".to_string(),
-                "2".to_string(),
-                "cour".to_string(),
-                "downing".to_string(),
-                "flat".to_string()
-            ]
-        );
-    }
-
-    #[test]
-    fn address_prefix_index_finds_partial_address_terms() {
-        let mut token_index: FxHashMap<String, Vec<u32>> = FxHashMap::default();
-        token_index.insert("downing".to_string(), vec![1]);
-        token_index.insert("downton".to_string(), vec![2]);
-        token_index.insert("market".to_string(), vec![3]);
-
-        let prefix_index = build_address_prefix_index(&token_index);
-
-        assert_eq!(
-            prefix_index.get("down").cloned().unwrap_or_default(),
-            vec!["downing".to_string(), "downton".to_string()]
-        );
-        assert_eq!(
-            prefix_index.get("downi").cloned().unwrap_or_default(),
-            vec!["downing".to_string()]
-        );
-        assert_eq!(
-            prefix_index.get("downt").cloned().unwrap_or_default(),
-            vec!["downton".to_string()]
-        );
-        assert!(!prefix_index.contains_key("do"));
-    }
-
-    #[test]
-    fn address_term_matching_allows_prefixes_and_aliases() {
-        let tokens = tokenize_address_text("10 Downing Street");
-        let prefix_group = address_term_group("down").expect("prefix term should be searchable");
-        let alias_group = AddressTermGroup {
-            alternatives: vec!["st".to_string(), "street".to_string()],
-        };
-
-        assert!(address_tokens_match_group(&tokens, &prefix_group));
-        assert!(address_tokens_match_group(&tokens, &alias_group));
-    }
-
-    #[test]
-    fn address_term_matching_uses_actual_token_prefixes() {
-        let tokens = tokenize_address_text("12 Settlers Court");
-        let prefix_group = address_term_group("cou").expect("partial term should be searchable");
-
-        assert!(address_tokens_match_group(&tokens, &prefix_group));
-    }
-
-    #[test]
-    fn histogram_empty_data() {
-        let data: Vec<f32> = vec![];
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.slider_min, 0.0);
-        assert_eq!(stats.slider_max, 100.0);
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 0);
-    }
-
-    #[test]
-    fn histogram_single_value() {
-        let data = vec![50.0_f32];
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.min, 50.0);
-        assert_eq!(stats.histogram.max, 50.0);
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 1);
-    }
-
-    #[test]
-    fn histogram_uniform_distribution() {
-        let data: Vec<f32> = (0..100).map(|i| i as f32).collect();
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.min, 0.0);
-        assert_eq!(stats.histogram.max, 99.0);
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 100);
-    }
-
-    #[test]
-    fn histogram_with_nan_values() {
-        let data = vec![10.0_f32, f32::NAN, 20.0, f32::NAN, 30.0];
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 3);
-        assert_eq!(stats.histogram.min, 10.0);
-        assert_eq!(stats.histogram.max, 30.0);
-    }
-
-    #[test]
-    fn histogram_all_nan() {
-        let data = vec![f32::NAN, f32::NAN, f32::NAN];
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 0);
-    }
-
-    #[test]
-    fn histogram_all_same_value() {
-        let data = vec![42.0_f32; 1000];
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.min, 42.0);
-        assert_eq!(stats.histogram.max, 42.0);
-        assert_eq!(stats.histogram.p1, 42.0);
-        assert_eq!(stats.histogram.p99, 42.0);
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 1000);
-    }
-
-    #[test]
-    fn histogram_percentile_bounds() {
-        let mut data: Vec<f32> = vec![0.0]; // Low outlier
-        data.extend((1..99).map(|i| 50.0 + i as f32 * 0.01));
-        data.push(1000.0); // High outlier
-
-        let bounds = make_percentile_bounds(2.0, 98.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert!(stats.slider_min > 0.0);
-        assert!(stats.slider_max < 1000.0);
-    }
-
-    #[test]
-    fn fixed_price_bounds_keep_slider_cap() {
-        let data = vec![400_000.0_f32, 2_500_000.0, 3_750_000.0];
-        let bounds = make_fixed_bounds(0.0, 2_500_000.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.slider_min, 0.0);
-        assert_eq!(stats.slider_max, 2_500_000.0);
-    }
-
-    #[test]
-    fn histogram_bin_for_value() {
-        let hist = Histogram {
-            min: 0.0,
-            max: 100.0,
-            p1: 10.0,
-            p99: 90.0,
-            counts: vec![0; 10],
-        };
-
-        assert_eq!(hist.bin_for_value(5.0), 0); // Low outlier bin
-        assert_eq!(hist.bin_for_value(95.0), 9); // High outlier bin
-
-        let mid_value = 50.0;
-        let bin = hist.bin_for_value(mid_value);
-        assert!((1..=8).contains(&bin));
-    }
-
-    #[test]
-    fn histogram_middle_bin_width() {
-        let hist = Histogram {
-            min: 0.0,
-            max: 100.0,
-            p1: 10.0,
-            p99: 90.0,
-            counts: vec![0; 10],
-        };
-
-        let expected_width = (90.0 - 10.0) / 8.0;
-        assert!((hist.middle_bin_width() - expected_width).abs() < 0.001);
-    }
-
-    #[test]
-    fn histogram_cardinality_caps_bins() {
-        let data = vec![1.0_f32, 1.0, 2.0, 2.0, 3.0, 3.0];
-        let bounds = make_fixed_bounds(0.0, 100.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.counts.len(), 3);
-    }
-
-    #[test]
-    fn min_max_skips_nan() {
-        let values = vec![10.0_f32, f32::NAN, 20.0, f32::NAN, 5.0];
-
-        let mut min = f32::INFINITY;
-        let mut max = f32::NEG_INFINITY;
-        for &v in &values {
-            if v.is_finite() {
-                if v < min {
-                    min = v;
-                }
-                if v > max {
-                    max = v;
-                }
-            }
-        }
-
-        assert_eq!(min, 5.0);
-        assert_eq!(max, 20.0);
-    }
-
-    #[test]
-    fn count_skips_nan() {
-        let values = [1.0_f32, f32::NAN, 2.0, f32::NAN, 3.0];
-        let count = values.iter().filter(|v| v.is_finite()).count();
-        assert_eq!(count, 3);
-    }
-
-    #[test]
-    fn enum_value_counting() {
-        let values = vec![0.0_f32, 1.0, 1.0, 2.0, f32::NAN, 3.0, 1.0];
-        let enum_count = 4;
-
-        let mut counts = vec![0u64; enum_count];
-        for &v in &values {
-            if v.is_finite() {
-                let idx = v as usize;
-                if idx < enum_count {
-                    counts[idx] += 1;
-                }
-            }
-        }
-
-        assert_eq!(counts[0], 1);
-        assert_eq!(counts[1], 3);
-        assert_eq!(counts[2], 1);
-        assert_eq!(counts[3], 1);
-    }
-
-    #[test]
-    fn infinity_values_excluded() {
-        let data = vec![f32::INFINITY, f32::NEG_INFINITY, 50.0];
-        let bounds = Bounds::Fixed {
-            min: 0.0,
-            max: 100.0,
-        };
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.min, 50.0);
-        assert_eq!(stats.histogram.max, 50.0);
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 1);
-    }
-
-    #[test]
-    fn only_finite_values() {
-        let data = vec![10.0_f32, 20.0, 30.0];
-        let bounds = Bounds::Fixed {
-            min: 0.0,
-            max: 100.0,
-        };
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        assert_eq!(stats.histogram.min, 10.0);
-        assert_eq!(stats.histogram.max, 30.0);
-        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 3);
-    }
-
-    #[test]
-    fn extreme_outlier_does_not_destroy_quantization() {
-        // Simulate floor area: 10k normal values (50-200 sqm) + one 317M outlier
-        let mut data: Vec<f32> = (0..10_000).map(|i| 50.0 + (i % 150) as f32).collect();
-        data.push(317_000_000.0); // Extreme outlier from web scraping
-
-        let bounds = make_percentile_bounds(0.0, 98.0);
-        let stats = compute_feature_stats(&data, &bounds, false);
-
-        // After refinement, histogram range should be much tighter than 317M
-        assert!(
-            stats.histogram.max < 1_000_000.0,
-            "histogram.max should be refined, got {}",
-            stats.histogram.max,
-        );
-        // p1 should be near 50, not millions
-        assert!(
-            stats.histogram.p1 < 100.0,
-            "p1 should be near real data, got {}",
-            stats.histogram.p1,
-        );
-        // Slider min should reflect actual data range
-        assert!(
-            stats.slider_min < 100.0,
-            "slider_min should be near real data, got {}",
-            stats.slider_min,
-        );
-
-        // Quantization using histogram.min/max should give usable range
-        let qmin = stats.histogram.min;
-        let qrange = stats.histogram.max - stats.histogram.min;
-        assert!(qrange > 0.0 && qrange < 1_000_000.0);
-
-        // A typical floor area (100 sqm) should be distinguishable from min
-        let normalized = (100.0 - qmin) / qrange;
-        let encoded = (normalized * QUANT_SCALE).round() as u16;
-        assert!(
-            encoded > 100,
-            "100 sqm should encode to a meaningful u16 value, got {}",
-            encoded,
-        );
-    }
-}
diff --git a/server-rs/src/data/property/address_search.rs b/server-rs/src/data/property/address_search.rs
new file mode 100644
index 0000000..0a2c4d1
--- /dev/null
+++ b/server-rs/src/data/property/address_search.rs
@@ -0,0 +1,973 @@
+//! Address search: tokenization, query parsing, inverted/prefix indexes and the
+//! ranked per-row search over property addresses.
+
+use rustc_hash::{FxHashMap, FxHashSet};
+
+use super::PropertyData;
+
+/// Upper bound on rows scored per query. Intersection keeps most candidate sets far below
+/// this; only a single very common road word (e.g. "high") approaches it, and the in-area
+/// priority sort keeps a refined query's matches ahead of the cut.
+const ADDRESS_SEARCH_CANDIDATE_LIMIT: usize = 150_000;
+const ADDRESS_SEARCH_PREFIX_MIN_LEN: usize = 4;
+const ADDRESS_SEARCH_PREFIX_MAX_LEN: usize = 8;
+
+#[derive(Clone, Debug)]
+pub(super) struct AddressTermGroup {
+    alternatives: Vec<String>,
+}
+
+#[derive(Debug)]
+pub(super) struct AddressQuery {
+    full_postcode: Option<String>,
+    /// Compact uppercase outward code (optionally with a sector digit) recovered when the
+    /// user appended a partial postcode like "NW1" or "NW1 6". Used as an additive ranking
+    /// bias, never as a hard filter — so the disambiguating hint is honoured without
+    /// excluding the same road in other areas.
+    postcode_area: Option<String>,
+    text_groups: Vec<AddressTermGroup>,
+    numeric_terms: Vec<String>,
+    candidate_terms: Vec<String>,
+}
+
+fn tokenize_address_text(text: &str) -> Vec<String> {
+    let mut tokens = Vec::new();
+    let mut current = String::new();
+
+    for ch in text.chars() {
+        if ch.is_ascii_alphanumeric() {
+            current.push(ch.to_ascii_lowercase());
+        } else if matches!(ch, '\'' | '’' | '`') {
+            continue;
+        } else if !current.is_empty() {
+            tokens.push(std::mem::take(&mut current));
+        }
+    }
+
+    if !current.is_empty() {
+        tokens.push(current);
+    }
+
+    tokens
+}
+
+fn is_full_postcode_compact(compact: &str) -> bool {
+    let bytes = compact.as_bytes();
+    let len = bytes.len();
+    if !(5..=7).contains(&len) {
+        return false;
+    }
+
+    let inward = &bytes[len - 3..];
+    if !inward[0].is_ascii_digit()
+        || !inward[1].is_ascii_alphabetic()
+        || !inward[2].is_ascii_alphabetic()
+    {
+        return false;
+    }
+
+    let outward = &bytes[..len - 3];
+    if !(2..=4).contains(&outward.len()) {
+        return false;
+    }
+
+    outward[0].is_ascii_alphabetic()
+        && outward.iter().all(u8::is_ascii_alphanumeric)
+        && outward.iter().any(u8::is_ascii_digit)
+}
+
+fn canonical_postcode_from_compact(compact: &str) -> String {
+    let upper = compact.to_ascii_uppercase();
+    let split = upper.len() - 3;
+    format!("{} {}", &upper[..split], &upper[split..])
+}
+
+fn extract_full_postcode(tokens: &[String]) -> Option<(String, Vec<usize>)> {
+    for (idx, token) in tokens.iter().enumerate() {
+        let compact = token.to_ascii_uppercase();
+        if is_full_postcode_compact(&compact) {
+            return Some((canonical_postcode_from_compact(&compact), vec![idx]));
+        }
+    }
+
+    for idx in 0..tokens.len().saturating_sub(1) {
+        let compact = format!(
+            "{}{}",
+            tokens[idx].to_ascii_uppercase(),
+            tokens[idx + 1].to_ascii_uppercase()
+        );
+        if is_full_postcode_compact(&compact) {
+            return Some((
+                canonical_postcode_from_compact(&compact),
+                vec![idx, idx + 1],
+            ));
+        }
+    }
+
+    None
+}
+
+fn looks_like_postcode_fragment(token: &str) -> bool {
+    (2..=4).contains(&token.len())
+        && token
+            .chars()
+            .next()
+            .is_some_and(|ch| ch.is_ascii_alphabetic())
+        && token.chars().any(|ch| ch.is_ascii_digit())
+        && token.chars().all(|ch| ch.is_ascii_alphanumeric())
+}
+
+fn is_numeric_address_token(token: &str) -> bool {
+    token.chars().all(|ch| ch.is_ascii_digit())
+}
+
+fn address_token_aliases(token: &str) -> Vec<&'static str> {
+    match token {
+        "apt" => vec!["apt", "apartment"],
+        "apartment" => vec!["apartment", "apt"],
+        "ave" => vec!["ave", "avenue"],
+        "avenue" => vec!["avenue", "ave"],
+        "blvd" => vec!["blvd", "boulevard"],
+        "boulevard" => vec!["boulevard", "blvd"],
+        "cl" => vec!["cl", "close"],
+        "close" => vec!["close", "cl"],
+        "ct" => vec!["ct", "court"],
+        "court" => vec!["court", "ct"],
+        "cres" => vec!["cres", "crescent"],
+        "crescent" => vec!["crescent", "cres"],
+        "dr" => vec!["dr", "drive"],
+        "drive" => vec!["drive", "dr"],
+        "fl" => vec!["fl", "flat"],
+        "flat" => vec!["flat", "fl"],
+        "gdns" => vec!["gdns", "gardens", "garden"],
+        "garden" => vec!["garden", "gardens", "gdns"],
+        "gardens" => vec!["gardens", "garden", "gdns"],
+        "hse" => vec!["hse", "house"],
+        "house" => vec!["house", "hse"],
+        "ln" => vec!["ln", "lane"],
+        "lane" => vec!["lane", "ln"],
+        "rd" => vec!["rd", "road"],
+        "road" => vec!["road", "rd"],
+        "sq" => vec!["sq", "square"],
+        "square" => vec!["square", "sq"],
+        "st" => vec!["st", "street", "saint"],
+        "street" => vec!["street", "st"],
+        "saint" => vec!["saint", "st"],
+        "terr" => vec!["terr", "terrace"],
+        "terrace" => vec!["terrace", "terr"],
+        _ => Vec::new(),
+    }
+}
+
+fn is_address_stop_token(token: &str) -> bool {
+    matches!(
+        token,
+        "a" | "an"
+            | "and"
+            | "apartment"
+            | "apt"
+            | "avenue"
+            | "ave"
+            | "block"
+            | "building"
+            | "bungalow"
+            | "close"
+            | "cl"
+            | "court"
+            | "ct"
+            | "cres"
+            | "crescent"
+            | "drive"
+            | "dr"
+            | "estate"
+            | "flat"
+            | "fl"
+            | "floor"
+            | "garden"
+            | "gardens"
+            | "gdns"
+            | "grove"
+            | "house"
+            | "hse"
+            | "lane"
+            | "ln"
+            | "lodge"
+            | "mansions"
+            | "mews"
+            | "of"
+            | "park"
+            | "place"
+            | "road"
+            | "rd"
+            | "room"
+            | "row"
+            | "saint"
+            | "sq"
+            | "square"
+            | "st"
+            | "street"
+            | "terr"
+            | "terrace"
+            | "the"
+            | "unit"
+            | "view"
+            | "villas"
+            | "walk"
+            | "way"
+            | "yard"
+    )
+}
+
+fn address_term_group(token: &str) -> Option<AddressTermGroup> {
+    if token.len() < 3 || is_numeric_address_token(token) || looks_like_postcode_fragment(token) {
+        return None;
+    }
+
+    let mut alternatives = Vec::new();
+    alternatives.push(token.to_string());
+    for alias in address_token_aliases(token) {
+        if !alternatives.iter().any(|existing| existing == alias) {
+            alternatives.push(alias.to_string());
+        }
+    }
+
+    if alternatives
+        .iter()
+        .all(|alternative| is_address_stop_token(alternative))
+    {
+        return None;
+    }
+
+    Some(AddressTermGroup { alternatives })
+}
+
+pub(super) fn address_search_tokens(text: &str) -> Vec<String> {
+    let mut tokens: Vec<String> = tokenize_address_text(text)
+        .into_iter()
+        .filter(|token| is_address_search_token(token))
+        .collect();
+    tokens.sort_unstable();
+    tokens.dedup();
+    tokens
+}
+
+fn is_address_search_token(token: &str) -> bool {
+    if looks_like_postcode_fragment(token) {
+        return false;
+    }
+
+    if is_numeric_address_token(token) {
+        return true;
+    }
+
+    if token.chars().any(|ch| ch.is_ascii_digit()) {
+        return token.len() >= 2;
+    }
+
+    token.len() >= 3
+}
+
+pub(super) fn is_address_candidate_token(token: &str) -> bool {
+    !is_numeric_address_token(token)
+        && !looks_like_postcode_fragment(token)
+        && (token.chars().any(|ch| ch.is_ascii_digit())
+            || (token.len() >= 3 && !is_address_stop_token(token)))
+}
+
+fn address_prefix_key(term: &str) -> &str {
+    if term.len() > ADDRESS_SEARCH_PREFIX_MAX_LEN {
+        &term[..ADDRESS_SEARCH_PREFIX_MAX_LEN]
+    } else {
+        term
+    }
+}
+
+pub(super) fn build_address_prefix_index(
+    address_token_index: &FxHashMap<String, Vec<u32>>,
+) -> FxHashMap<String, Vec<String>> {
+    let mut prefix_index: FxHashMap<String, Vec<String>> = FxHashMap::default();
+
+    for token in address_token_index.keys() {
+        let max_prefix_len = token.len().min(ADDRESS_SEARCH_PREFIX_MAX_LEN);
+        for prefix_len in ADDRESS_SEARCH_PREFIX_MIN_LEN..=max_prefix_len {
+            prefix_index
+                .entry(token[..prefix_len].to_string())
+                .or_default()
+                .push(token.clone());
+        }
+    }
+
+    for tokens in prefix_index.values_mut() {
+        tokens.sort_unstable();
+        tokens.dedup();
+    }
+
+    prefix_index
+}
+
+/// Intersect two ascending-sorted row-id slices.
+fn intersect_sorted(left: &[u32], right: &[u32]) -> Vec<u32> {
+    let mut out = Vec::new();
+    let (mut i, mut j) = (0, 0);
+    while i < left.len() && j < right.len() {
+        match left[i].cmp(&right[j]) {
+            std::cmp::Ordering::Less => i += 1,
+            std::cmp::Ordering::Greater => j += 1,
+            std::cmp::Ordering::Equal => {
+                out.push(left[i]);
+                i += 1;
+                j += 1;
+            }
+        }
+    }
+    out
+}
+
+/// Union two ascending-sorted row-id slices (deduplicated, stays sorted).
+fn union_sorted(left: &[u32], right: &[u32]) -> Vec<u32> {
+    let mut out = Vec::with_capacity(left.len() + right.len());
+    let (mut i, mut j) = (0, 0);
+    while i < left.len() && j < right.len() {
+        match left[i].cmp(&right[j]) {
+            std::cmp::Ordering::Less => {
+                out.push(left[i]);
+                i += 1;
+            }
+            std::cmp::Ordering::Greater => {
+                out.push(right[j]);
+                j += 1;
+            }
+            std::cmp::Ordering::Equal => {
+                out.push(left[i]);
+                i += 1;
+                j += 1;
+            }
+        }
+    }
+    out.extend_from_slice(&left[i..]);
+    out.extend_from_slice(&right[j..]);
+    out
+}
+
+/// An ordinal like "1st", "2nd", "3rd", "21st" — part of the street name ("2nd Avenue"), not a
+/// house-number prefix.
+fn is_ordinal_token(token: &str) -> bool {
+    let split = token.len().saturating_sub(2);
+    let (digits, suffix) = token.split_at(split);
+    !digits.is_empty()
+        && digits.chars().all(|ch| ch.is_ascii_digit())
+        && matches!(suffix, "st" | "nd" | "rd" | "th")
+}
+
+/// Leading address tokens that denote a unit/house number rather than the street itself.
+fn is_house_prefix_token(token: &str) -> bool {
+    if is_ordinal_token(token) {
+        return false;
+    }
+    matches!(
+        token,
+        "flat" | "fl" | "apartment" | "apt" | "unit" | "no" | "block" | "floor" | "room"
+    ) || token.len() == 1
+        || token.chars().all(|ch| ch.is_ascii_digit())
+        || (token.chars().next().is_some_and(|ch| ch.is_ascii_digit())
+            && token.chars().any(|ch| ch.is_ascii_alphabetic()))
+}
+
+/// Street-level key for an address: drops the leading house-number / flat prefix so that
+/// "12 Baker Street" and "5 Baker Street" collapse to a single street entry.
+fn street_key(address: &str) -> String {
+    let tokens = tokenize_address_text(address);
+    let mut start = 0;
+    while start < tokens.len() && is_house_prefix_token(&tokens[start]) {
+        start += 1;
+    }
+    if start >= tokens.len() {
+        return tokens.join(" ");
+    }
+    tokens[start..].join(" ")
+}
+
+/// Road-type words. Their presence (with no house number) marks a road browse, which we
+/// collapse to one result per street.
+const ROAD_TYPE_TOKENS: &[&str] = &[
+    "street",
+    "st",
+    "road",
+    "rd",
+    "lane",
+    "ln",
+    "avenue",
+    "ave",
+    "close",
+    "cl",
+    "drive",
+    "dr",
+    "way",
+    "court",
+    "ct",
+    "crescent",
+    "cres",
+    "place",
+    "terrace",
+    "terr",
+    "grove",
+    "gardens",
+    "gdns",
+    "walk",
+    "row",
+    "square",
+    "sq",
+    "hill",
+    "parade",
+    "mews",
+    "embankment",
+    "broadway",
+    "boulevard",
+    "blvd",
+];
+
+fn query_has_road_type(query: &str) -> bool {
+    tokenize_address_text(query)
+        .iter()
+        .any(|token| ROAD_TYPE_TOKENS.contains(&token.as_str()))
+}
+
+/// The outward code (everything before the space) of a canonical postcode.
+fn outcode_of(postcode: &str) -> &str {
+    postcode.split(' ').next().unwrap_or(postcode)
+}
+
+fn parse_address_query(query: &str) -> AddressQuery {
+    let tokens = tokenize_address_text(query);
+    let (full_postcode, postcode_token_indices) = extract_full_postcode(&tokens)
+        .map(|(postcode, indices)| (Some(postcode), indices))
+        .unwrap_or((None, Vec::new()));
+
+    let skip_postcode_tokens: FxHashSet<usize> = postcode_token_indices.into_iter().collect();
+
+    // Recover an appended partial postcode (outcode, or outcode + sector digit) as a ranking
+    // bias rather than discarding it — but only from the TRAILING position, so a leading road
+    // designation like "A4 Great West Road" is not mistaken for an area refinement.
+    let mut postcode_area: Option<String> = None;
+    let mut consumed_partial_tokens: FxHashSet<usize> = FxHashSet::default();
+    if full_postcode.is_none() && !tokens.is_empty() {
+        let last = tokens.len() - 1;
+        if !skip_postcode_tokens.contains(&last) {
+            let sector_digit =
+                tokens[last].len() == 1 && tokens[last].chars().all(|ch| ch.is_ascii_digit());
+            if last >= 1
+                && sector_digit
+                && !skip_postcode_tokens.contains(&(last - 1))
+                && looks_like_postcode_fragment(&tokens[last - 1])
+            {
+                postcode_area = Some(format!(
+                    "{}{}",
+                    tokens[last - 1].to_ascii_uppercase(),
+                    tokens[last]
+                ));
+                consumed_partial_tokens.insert(last);
+                consumed_partial_tokens.insert(last - 1);
+            } else if looks_like_postcode_fragment(&tokens[last]) {
+                postcode_area = Some(tokens[last].to_ascii_uppercase());
+                consumed_partial_tokens.insert(last);
+            }
+        }
+    }
+
+    let mut text_groups = Vec::new();
+    let mut numeric_terms = Vec::new();
+    let mut candidate_terms = Vec::new();
+
+    for (idx, token) in tokens.iter().enumerate() {
+        if skip_postcode_tokens.contains(&idx)
+            || consumed_partial_tokens.contains(&idx)
+            || looks_like_postcode_fragment(token)
+        {
+            continue;
+        }
+
+        if is_numeric_address_token(token) {
+            numeric_terms.push(token.clone());
+            continue;
+        }
+
+        if let Some(group) = address_term_group(token) {
+            for alternative in &group.alternatives {
+                if !is_address_stop_token(alternative)
+                    && !candidate_terms.iter().any(|term| term == alternative)
+                {
+                    candidate_terms.push(alternative.clone());
+                }
+            }
+            text_groups.push(group);
+        } else if token.chars().any(|ch| ch.is_ascii_digit()) && token.len() >= 2 {
+            numeric_terms.push(token.clone());
+            if !candidate_terms.iter().any(|term| term == token) {
+                candidate_terms.push(token.clone());
+            }
+        }
+    }
+
+    text_groups.dedup_by(|left, right| left.alternatives == right.alternatives);
+    numeric_terms.sort_unstable();
+    numeric_terms.dedup();
+
+    AddressQuery {
+        full_postcode,
+        postcode_area,
+        text_groups,
+        numeric_terms,
+        candidate_terms,
+    }
+}
+
+fn token_matches_query_term(token: &str, query_term: &str) -> bool {
+    token == query_term || (query_term.len() >= 3 && token.starts_with(query_term))
+}
+
+fn token_matches_numeric_term(token: &str, query_term: &str) -> bool {
+    token == query_term || token.starts_with(query_term)
+}
+
+#[cfg(test)]
+fn address_tokens_match_group(tokens: &[String], group: &AddressTermGroup) -> bool {
+    group.alternatives.iter().any(|alternative| {
+        tokens
+            .iter()
+            .any(|token| token_matches_query_term(token, alternative))
+    })
+}
+
+impl PropertyData {
+    fn row_address_search_tokens(&self, row: usize) -> &[lasso::Spur] {
+        let offset = self.address_search_token_offsets[row] as usize;
+        let length = self.address_search_token_lengths[row] as usize;
+        &self.address_search_token_keys[offset..offset + length]
+    }
+
+    /// Search individual property addresses, returning `(row, score)` ranked best-first.
+    ///
+    /// Candidate rows come from intersecting the posting lists of the distinctive words the
+    /// user typed in full (so "Cherry Hinton Road" narrows to rows containing both), unioned
+    /// with the exact-postcode rows when a complete postcode is present (so a postcode is a
+    /// boost, not an all-or-nothing gate). An appended partial postcode keeps in-area rows
+    /// ahead of the candidate cut and adds a scoring bias. With a road-type word and no house
+    /// number, results collapse to one row per street.
+    pub fn search_addresses(&self, query: &str, limit: usize) -> Vec<(usize, i32)> {
+        if limit == 0 {
+            return Vec::new();
+        }
+
+        let parsed = parse_address_query(query);
+        if parsed.full_postcode.is_none()
+            && parsed.text_groups.is_empty()
+            && parsed.numeric_terms.is_empty()
+        {
+            return Vec::new();
+        }
+
+        let mut candidate_rows = self.address_candidate_rows(&parsed.candidate_terms);
+
+        // A complete postcode contributes its rows too, instead of replacing the road match.
+        if let Some(postcode) = parsed.full_postcode.as_deref() {
+            if let Some(rows) = self
+                .postcode_interner
+                .get(postcode)
+                .and_then(|key| self.postcode_row_index.get(&key))
+            {
+                candidate_rows = if candidate_rows.is_empty() {
+                    rows.clone()
+                } else {
+                    union_sorted(&candidate_rows, rows)
+                };
+            }
+        }
+
+        if candidate_rows.is_empty() {
+            return Vec::new();
+        }
+
+        // When the user appended a partial postcode, keep in-area rows ahead of the cut so the
+        // refinement still surfaces even for very common roads. Single pass (stable partition) so
+        // the postcode check — which allocates — runs exactly once per candidate.
+        if let Some(area) = parsed.postcode_area.as_deref() {
+            let mut in_area = Vec::new();
+            let mut others = Vec::new();
+            for &row in &candidate_rows {
+                if self.row_postcode_in_area(row as usize, area) {
+                    in_area.push(row);
+                } else {
+                    others.push(row);
+                }
+            }
+            in_area.extend(others);
+            candidate_rows = in_area;
+        }
+        candidate_rows.truncate(ADDRESS_SEARCH_CANDIDATE_LIMIT);
+
+        let mut scored: Vec<(i32, usize, usize)> = candidate_rows
+            .into_iter()
+            .filter_map(|row| {
+                let row = row as usize;
+                self.address_match_score(row, &parsed)
+                    .map(|score| (score, self.address(row).len(), row))
+            })
+            .collect();
+
+        scored.sort_unstable_by(|left, right| {
+            right
+                .0
+                .cmp(&left.0)
+                .then(left.1.cmp(&right.1))
+                .then(left.2.cmp(&right.2))
+        });
+
+        // Collapse a road browse (road-type word, no house number) to one row per street.
+        let collapse_streets = parsed.numeric_terms.is_empty() && query_has_road_type(query);
+
+        let mut seen = FxHashSet::default();
+        let mut results = Vec::with_capacity(limit);
+        for (score, _, row) in scored {
+            let address = self.address(row).trim();
+            if address.is_empty() {
+                continue;
+            }
+            let key = if collapse_streets {
+                format!(
+                    "{}\n{}",
+                    street_key(address),
+                    outcode_of(self.postcode(row))
+                )
+            } else {
+                format!("{}\n{}", address.to_ascii_lowercase(), self.postcode(row))
+            };
+            if !seen.insert(key) {
+                continue;
+            }
+            results.push((row, score));
+            if results.len() == limit {
+                break;
+            }
+        }
+
+        results
+    }
+
+    /// True when the row's postcode begins with the compact partial-postcode `area`
+    /// (e.g. "NW1" or "NW16" matches "NW1 6XE").
+    fn row_postcode_in_area(&self, row: usize, area: &str) -> bool {
+        let mut compact = String::new();
+        for ch in self.postcode(row).chars() {
+            if !ch.is_whitespace() {
+                compact.push(ch.to_ascii_uppercase());
+            }
+        }
+        compact.starts_with(area)
+    }
+
+    /// Candidate rows for the distinctive query words. Words typed in full intersect by their
+    /// exact posting lists (precise); a still-being-typed final word with no exact match seeds
+    /// from the smallest prefix-expanded posting list (so partial typing keeps working).
+    fn address_candidate_rows(&self, terms: &[String]) -> Vec<u32> {
+        let mut exact: Vec<&[u32]> = terms
+            .iter()
+            .filter_map(|term| self.address_token_index.get(term).map(Vec::as_slice))
+            .collect();
+
+        if !exact.is_empty() {
+            exact.sort_by_key(|rows| rows.len());
+            let mut acc = exact[0].to_vec();
+            for rows in &exact[1..] {
+                if acc.is_empty() {
+                    break;
+                }
+                acc = intersect_sorted(&acc, rows);
+            }
+            return acc;
+        }
+
+        self.prefix_seed_rows(terms)
+    }
+
+    /// Seed rows from the smallest prefix-expanded term — used only when no word matched an
+    /// indexed token exactly (i.e. the user is still typing the final word).
+    fn prefix_seed_rows(&self, terms: &[String]) -> Vec<u32> {
+        let mut best: Option<Vec<u32>> = None;
+        for term in terms {
+            if term.len() < ADDRESS_SEARCH_PREFIX_MIN_LEN {
+                continue;
+            }
+            let Some(tokens) = self.address_prefix_index.get(address_prefix_key(term)) else {
+                continue;
+            };
+            let mut union: Vec<u32> = Vec::new();
+            for token in tokens {
+                if !token.starts_with(term) {
+                    continue;
+                }
+                if let Some(rows) = self.address_token_index.get(token) {
+                    union = if union.is_empty() {
+                        rows.clone()
+                    } else {
+                        union_sorted(&union, rows)
+                    };
+                }
+            }
+            if !union.is_empty()
+                && best
+                    .as_ref()
+                    .is_none_or(|current| union.len() < current.len())
+            {
+                best = Some(union);
+            }
+        }
+        best.unwrap_or_default()
+    }
+
+    fn address_match_score(&self, row: usize, parsed: &AddressQuery) -> Option<i32> {
+        if self.address(row).trim().is_empty() {
+            return None;
+        }
+
+        let tokens = self.row_address_search_tokens(row);
+        if parsed
+            .text_groups
+            .iter()
+            .any(|group| !self.address_tokens_match_group(tokens, group))
+        {
+            return None;
+        }
+
+        let numeric_matches = parsed
+            .numeric_terms
+            .iter()
+            .filter(|term| {
+                tokens.iter().any(|token| {
+                    token_matches_numeric_term(self.address_search_interner.resolve(token), term)
+                })
+            })
+            .count();
+
+        if !parsed.numeric_terms.is_empty() && numeric_matches == 0 {
+            return None;
+        }
+
+        let mut score = 0;
+        if parsed.full_postcode.is_some() {
+            score += 1_000;
+        }
+        score += (parsed.text_groups.len() as i32) * 200;
+        score += (numeric_matches as i32) * 90;
+        if numeric_matches == parsed.numeric_terms.len() && numeric_matches > 0 {
+            score += 50;
+        }
+        // Additive bias (never a filter) when the row sits in the appended partial postcode.
+        if let Some(area) = parsed.postcode_area.as_deref() {
+            if self.row_postcode_in_area(row, area) {
+                score += 400;
+            }
+        }
+
+        Some(score)
+    }
+
+    fn address_tokens_match_group(&self, tokens: &[lasso::Spur], group: &AddressTermGroup) -> bool {
+        group.alternatives.iter().any(|alternative| {
+            tokens.iter().any(|token| {
+                token_matches_query_term(self.address_search_interner.resolve(token), alternative)
+            })
+        })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn full_postcode_detection_accepts_common_formats() {
+        assert!(is_full_postcode_compact("SW1A1AA"));
+        assert!(is_full_postcode_compact("E142DG"));
+        assert!(is_full_postcode_compact("M11AE"));
+        assert!(!is_full_postcode_compact("E14"));
+        assert!(!is_full_postcode_compact("DOWNING"));
+        assert!(!is_full_postcode_compact("10A"));
+    }
+
+    #[test]
+    fn address_query_parsing_skips_postcodes_and_street_suffixes() {
+        let parsed = parse_address_query("Flat 2, 10 Downing St, SW1A 2AA");
+
+        assert_eq!(parsed.full_postcode.as_deref(), Some("SW1A 2AA"));
+        assert_eq!(
+            parsed.numeric_terms,
+            vec!["10".to_string(), "2".to_string()]
+        );
+        assert_eq!(parsed.candidate_terms, vec!["downing".to_string()]);
+        assert_eq!(parsed.text_groups.len(), 1);
+        assert_eq!(
+            parsed.text_groups[0].alternatives,
+            vec!["downing".to_string()]
+        );
+    }
+
+    #[test]
+    fn address_query_parsing_handles_compact_postcodes() {
+        let parsed = parse_address_query("10 downing street sw1a1aa");
+
+        assert_eq!(parsed.full_postcode.as_deref(), Some("SW1A 1AA"));
+        assert_eq!(parsed.numeric_terms, vec!["10".to_string()]);
+        assert_eq!(parsed.candidate_terms, vec!["downing".to_string()]);
+    }
+
+    #[test]
+    fn address_query_recovers_appended_partial_postcode_as_bias() {
+        let parsed = parse_address_query("Baker Street NW1");
+        assert_eq!(parsed.full_postcode, None);
+        assert_eq!(parsed.postcode_area.as_deref(), Some("NW1"));
+        // The road words are still searchable; the postcode fragment did not consume them.
+        assert_eq!(parsed.candidate_terms, vec!["baker".to_string()]);
+        assert!(parsed.numeric_terms.is_empty());
+    }
+
+    #[test]
+    fn address_query_recovers_outcode_plus_sector_without_a_phantom_house_number() {
+        let parsed = parse_address_query("High Street CR0 2");
+        assert_eq!(parsed.postcode_area.as_deref(), Some("CR02"));
+        // The lone sector digit must not be treated as a house number.
+        assert!(parsed.numeric_terms.is_empty());
+        assert_eq!(parsed.candidate_terms, vec!["high".to_string()]);
+    }
+
+    #[test]
+    fn full_postcode_takes_precedence_over_partial_bias() {
+        let parsed = parse_address_query("Baker Street NW1 6XE");
+        assert_eq!(parsed.full_postcode.as_deref(), Some("NW1 6XE"));
+        assert_eq!(parsed.postcode_area, None);
+    }
+
+    #[test]
+    fn intersect_and_union_sorted_row_ids() {
+        assert_eq!(
+            intersect_sorted(&[1, 2, 3, 5], &[2, 3, 4, 5]),
+            vec![2, 3, 5]
+        );
+        assert_eq!(intersect_sorted(&[1, 2], &[3, 4]), Vec::<u32>::new());
+        assert_eq!(union_sorted(&[1, 3, 5], &[2, 3, 4]), vec![1, 2, 3, 4, 5]);
+        assert_eq!(union_sorted(&[], &[2, 4]), vec![2, 4]);
+    }
+
+    #[test]
+    fn street_key_collapses_house_numbers_and_flats() {
+        assert_eq!(street_key("12 Baker Street"), "baker street");
+        assert_eq!(street_key("5 Baker Street"), "baker street");
+        assert_eq!(street_key("Flat 2, 10 Downing Street"), "downing street");
+        assert_eq!(street_key("221B Baker Street"), "baker street");
+    }
+
+    #[test]
+    fn street_key_keeps_ordinal_street_names() {
+        // Ordinals are part of the street name, not a house-number prefix.
+        assert_eq!(street_key("2nd Avenue"), "2nd avenue");
+        assert_eq!(street_key("12 3rd Avenue"), "3rd avenue");
+        assert!(is_ordinal_token("21st"));
+        assert!(!is_ordinal_token("21"));
+        assert!(!is_ordinal_token("221b"));
+    }
+
+    #[test]
+    fn postcode_area_recovered_only_from_the_trailing_position() {
+        // A leading road designation must NOT be taken as an area refinement.
+        let parsed = parse_address_query("A4 Great West Road");
+        assert_eq!(parsed.postcode_area, None);
+        // A genuine trailing outcode still is.
+        let trailing = parse_address_query("Great West Road W4");
+        assert_eq!(trailing.postcode_area.as_deref(), Some("W4"));
+    }
+
+    #[test]
+    fn road_type_detection() {
+        assert!(query_has_road_type("high street"));
+        assert!(query_has_road_type("acacia avenue"));
+        assert!(!query_has_road_type("acacia"));
+        assert!(!query_has_road_type("london"));
+    }
+
+    #[test]
+    fn address_query_parsing_keeps_partial_terms_for_row_matching() {
+        let parsed = parse_address_query("settlers cour");
+
+        assert_eq!(parsed.full_postcode, None);
+        assert_eq!(parsed.numeric_terms, Vec::<String>::new());
+        assert_eq!(
+            parsed.candidate_terms,
+            vec!["settlers".to_string(), "cour".to_string()]
+        );
+        assert_eq!(parsed.text_groups.len(), 2);
+        assert_eq!(
+            parsed.text_groups[0].alternatives,
+            vec!["settlers".to_string()]
+        );
+        assert_eq!(parsed.text_groups[1].alternatives, vec!["cour".to_string()]);
+    }
+
+    #[test]
+    fn address_search_tokens_keep_actual_address_terms_for_scoring() {
+        let tokens = address_search_tokens("Flat 2, 10 Downing Cour");
+
+        assert_eq!(
+            tokens,
+            vec![
+                "10".to_string(),
+                "2".to_string(),
+                "cour".to_string(),
+                "downing".to_string(),
+                "flat".to_string()
+            ]
+        );
+    }
+
+    #[test]
+    fn address_prefix_index_finds_partial_address_terms() {
+        let mut token_index: FxHashMap<String, Vec<u32>> = FxHashMap::default();
+        token_index.insert("downing".to_string(), vec![1]);
+        token_index.insert("downton".to_string(), vec![2]);
+        token_index.insert("market".to_string(), vec![3]);
+
+        let prefix_index = build_address_prefix_index(&token_index);
+
+        assert_eq!(
+            prefix_index.get("down").cloned().unwrap_or_default(),
+            vec!["downing".to_string(), "downton".to_string()]
+        );
+        assert_eq!(
+            prefix_index.get("downi").cloned().unwrap_or_default(),
+            vec!["downing".to_string()]
+        );
+        assert_eq!(
+            prefix_index.get("downt").cloned().unwrap_or_default(),
+            vec!["downton".to_string()]
+        );
+        assert!(!prefix_index.contains_key("do"));
+    }
+
+    #[test]
+    fn address_term_matching_allows_prefixes_and_aliases() {
+        let tokens = tokenize_address_text("10 Downing Street");
+        let prefix_group = address_term_group("down").expect("prefix term should be searchable");
+        let alias_group = AddressTermGroup {
+            alternatives: vec!["st".to_string(), "street".to_string()],
+        };
+
+        assert!(address_tokens_match_group(&tokens, &prefix_group));
+        assert!(address_tokens_match_group(&tokens, &alias_group));
+    }
+
+    #[test]
+    fn address_term_matching_uses_actual_token_prefixes() {
+        let tokens = tokenize_address_text("12 Settlers Court");
+        let prefix_group = address_term_group("cou").expect("partial term should be searchable");
+
+        assert!(address_tokens_match_group(&tokens, &prefix_group));
+    }
+}
diff --git a/server-rs/src/data/property/h3.rs b/server-rs/src/data/property/h3.rs
new file mode 100644
index 0000000..fc185e9
--- /dev/null
+++ b/server-rs/src/data/property/h3.rs
@@ -0,0 +1,34 @@
+//! H3 spatial cell precomputation for property rows.
+
+use anyhow::Context;
+use rayon::prelude::*;
+
+use crate::consts::H3_PRECOMPUTE_MAX;
+
+/// Precompute H3 cell IDs for all rows at the maximum resolution only.
+/// Parent cells for lower resolutions are derived on the fly via `CellIndex::parent()`.
+pub fn precompute_h3(lat: &[f32], lon: &[f32]) -> anyhow::Result<Vec<u64>> {
+    let res = H3_PRECOMPUTE_MAX;
+    tracing::info!("Precomputing H3 cells at resolution {}", res);
+
+    let h3_res =
+        h3o::Resolution::try_from(res).with_context(|| format!("Invalid H3 resolution: {res}"))?;
+
+    let cells: Vec<u64> = lat
+        .par_iter()
+        .zip(lon.par_iter())
+        .enumerate()
+        .map(|(i, (&latitude, &longitude))| {
+            let coord = h3o::LatLng::new(latitude as f64, longitude as f64).unwrap_or_else(|err| {
+                panic!(
+                    "Invalid coordinates at row {}: lat={}, lon={}: {}",
+                    i, latitude, longitude, err
+                )
+            });
+            u64::from(coord.to_cell(h3_res))
+        })
+        .collect();
+
+    tracing::info!("H3 precomputation complete ({} cells)", cells.len());
+    Ok(cells)
+}
diff --git a/server-rs/src/data/property/loading.rs b/server-rs/src/data/property/loading.rs
new file mode 100644
index 0000000..4081e0b
--- /dev/null
+++ b/server-rs/src/data/property/loading.rs
@@ -0,0 +1,1105 @@
+//! Parquet ingestion: loading and joining the properties + postcode parquet
+//! files, validating coordinates, spatially sorting rows and building the
+//! quantized row-major feature matrix plus all side tables.
+
+use anyhow::{bail, Context};
+use polars::lazy::frame::LazyFrame;
+use polars::prelude::*;
+use rayon::prelude::*;
+use std::path::Path;
+
+use rustc_hash::{FxHashMap, FxHashSet};
+
+use crate::consts::{NAN_U16, QUANT_SCALE};
+use crate::features::{self, Bounds};
+
+use super::address_search::{
+    address_search_tokens, build_address_prefix_index, is_address_candidate_token,
+};
+use super::poi_metrics::{PostcodePoiMetrics, NO_POI_METRIC_ROW};
+use super::stats::{column_to_f32_vec, compute_feature_stats, FeatureStats, Histogram};
+use super::{HistoricalPrice, PropertyData, RenovationEvent};
+
+const MISSING_COORDINATE_SAMPLE_LIMIT: usize = 10;
+const COUNTRY_COLUMN_CANDIDATES: &[&str] = &[
+    "ctry25cd",
+    "ctry24cd",
+    "ctry23cd",
+    "ctry22cd",
+    "country_code",
+    "Country code",
+    "country",
+    "Country",
+];
+const ENGLAND_COUNTRY_VALUES: &[&str] = &["E92000001", "England", "ENGLAND", "england"];
+
+fn is_numeric_dtype(dtype: &DataType) -> bool {
+    matches!(
+        dtype,
+        DataType::Int8
+            | DataType::Int16
+            | DataType::Int32
+            | DataType::Int64
+            | DataType::UInt8
+            | DataType::UInt16
+            | DataType::UInt32
+            | DataType::UInt64
+            | DataType::Float32
+            | DataType::Float64
+            | DataType::Datetime(_, _)
+            | DataType::Date
+    )
+}
+
+fn is_datetime_dtype(dtype: &DataType) -> bool {
+    matches!(dtype, DataType::Datetime(_, _) | DataType::Date)
+}
+
+fn find_country_column(schema: &Schema) -> Option<String> {
+    COUNTRY_COLUMN_CANDIDATES
+        .iter()
+        .find_map(|&name| match schema.get(name) {
+            Some(dtype) if matches!(dtype, DataType::String) || dtype.is_categorical() => {
+                Some(name.to_string())
+            }
+            _ => None,
+        })
+}
+
+fn england_country_expr(country_column: &str) -> Expr {
+    ENGLAND_COUNTRY_VALUES.iter().skip(1).fold(
+        col(country_column)
+            .cast(DataType::String)
+            .eq(lit(ENGLAND_COUNTRY_VALUES[0])),
+        |expr, value| expr.or(col(country_column).cast(DataType::String).eq(lit(*value))),
+    )
+}
+
+#[derive(Debug, PartialEq, Eq)]
+struct MissingCoordinateSummary {
+    row_count: usize,
+    unique_postcode_count: usize,
+    sample_postcodes: Vec<String>,
+}
+
+fn summarize_missing_coordinate_postcodes<'a>(
+    postcodes: impl IntoIterator<Item = Option<&'a str>>,
+) -> MissingCoordinateSummary {
+    let mut row_count = 0usize;
+    let mut unique_postcodes = FxHashSet::default();
+
+    for postcode in postcodes {
+        row_count += 1;
+        if let Some(postcode) = postcode {
+            let trimmed = postcode.trim();
+            if !trimmed.is_empty() {
+                unique_postcodes.insert(trimmed.to_string());
+            }
+        }
+    }
+
+    let mut sample_postcodes: Vec<String> = unique_postcodes.iter().cloned().collect();
+    sample_postcodes.sort_unstable();
+    sample_postcodes.truncate(MISSING_COORDINATE_SAMPLE_LIMIT);
+
+    MissingCoordinateSummary {
+        row_count,
+        unique_postcode_count: unique_postcodes.len(),
+        sample_postcodes,
+    }
+}
+
+fn missing_england_coordinates_error(
+    summary: &MissingCoordinateSummary,
+    country_column: &str,
+) -> String {
+    let samples = if summary.sample_postcodes.is_empty() {
+        "none".to_string()
+    } else {
+        summary.sample_postcodes.join(", ")
+    };
+    format!(
+        "England property rows missing postcode coordinates after joining postcode data: {} rows across {} postcodes (country column '{}'). Sample postcodes: {}",
+        summary.row_count, summary.unique_postcode_count, country_column, samples
+    )
+}
+
+fn validate_no_england_rows_missing_coordinates(
+    combined_lf: &LazyFrame,
+    schema: &Schema,
+) -> anyhow::Result<()> {
+    let Some(country_column) = find_country_column(schema) else {
+        bail!(
+            "Postcode feature parquet has no reliable country column; cannot verify that rows with missing coordinates are outside England. Regenerate postcode.parquet with pipeline.transform.merge so it includes ctry25cd."
+        );
+    };
+
+    let missing_coordinates = col("lat").is_null().or(col("lon").is_null());
+    let offending_df = combined_lf
+        .clone()
+        .filter(missing_coordinates.and(england_country_expr(&country_column)))
+        .select([col("Postcode")])
+        .collect()
+        .context("Failed to validate missing postcode coordinates")?;
+
+    let postcode_column = offending_df
+        .column("Postcode")
+        .context("Joined frame missing 'Postcode' during coordinate validation")?
+        .str()
+        .context("'Postcode' column is not a string during coordinate validation")?;
+    let summary = summarize_missing_coordinate_postcodes(postcode_column);
+
+    if summary.row_count > 0 {
+        bail!(
+            "{}",
+            missing_england_coordinates_error(&summary, &country_column)
+        );
+    }
+
+    Ok(())
+}
+
+impl PropertyData {
+    pub fn load(properties_path: &Path, postcode_features_path: &Path) -> anyhow::Result<Self> {
+        crate::data::run_polars_io(|| Self::load_inner(properties_path, postcode_features_path))
+    }
+
+    fn load_inner(properties_path: &Path, postcode_features_path: &Path) -> anyhow::Result<Self> {
+        // Load postcode.parquet
+        tracing::info!(
+            "Loading postcode features from {:?}",
+            postcode_features_path
+        );
+        let postcode_features_path = PlRefPath::try_from_path(postcode_features_path)
+            .context("Failed to normalize postcode parquet path")?;
+        let postcode_df = LazyFrame::scan_parquet(postcode_features_path, Default::default())
+            .context("Failed to scan postcode parquet")?
+            .collect()
+            .context("Failed to read postcode parquet")?;
+        tracing::info!(rows = postcode_df.height(), "Postcode features loaded");
+
+        let mut poi_metric_names: Vec<String> = postcode_df
+            .get_column_names()
+            .iter()
+            .map(|name| name.as_str())
+            .filter(|&name| features::is_dynamic_poi_feature(name))
+            .map(str::to_string)
+            .collect();
+        poi_metric_names.sort_by_key(|name| features::dynamic_poi_feature_sort_key(name));
+
+        let poi_metric_by_postcode: FxHashMap<String, u32> = if poi_metric_names.is_empty() {
+            FxHashMap::default()
+        } else {
+            let postcode_column = postcode_df
+                .column("Postcode")
+                .context("Postcode feature parquet missing 'Postcode' column")?
+                .str()
+                .context("'Postcode' column in postcode feature parquet is not a string")?;
+            postcode_column
+                .into_iter()
+                .enumerate()
+                .filter_map(|(idx, postcode)| {
+                    postcode.map(|postcode| (postcode.to_string(), idx as u32))
+                })
+                .collect()
+        };
+        let mut poi_metrics = PostcodePoiMetrics::from_postcode_df(&postcode_df, poi_metric_names)?;
+
+        // Load properties.parquet and join with postcode data lazily so the
+        // wide combined frame is never fully materialized — projection is
+        // pushed down into the join, keeping peak memory bounded.
+        tracing::info!("Loading properties from {:?}", properties_path);
+        let properties_path = PlRefPath::try_from_path(properties_path)
+            .context("Failed to normalize properties parquet path")?;
+        let properties_lf = LazyFrame::scan_parquet(properties_path, Default::default())
+            .context("Failed to scan properties parquet")?;
+        let combined_lf = properties_lf.join(
+            postcode_df.lazy(),
+            [col("Postcode")],
+            [col("Postcode")],
+            JoinArgs::new(JoinType::Left),
+        );
+
+        // Get configured feature/enum names in config order. Dynamic POI
+        // metrics live in a postcode-level side table so they do not widen the
+        // hot row-major property feature matrix.
+        let configured_numeric_names = features::all_numeric_feature_names();
+        let enum_names = features::all_enum_feature_names();
+
+        let schema = combined_lf
+            .clone()
+            .collect_schema()
+            .context("Failed to collect joined schema")?;
+        validate_no_england_rows_missing_coordinates(&combined_lf, &schema)?;
+        let numeric_names: Vec<String> = configured_numeric_names
+            .iter()
+            .map(|name| (*name).to_string())
+            .collect();
+
+        for name in &numeric_names {
+            match schema.get(name.as_str()) {
+                Some(dtype) if is_numeric_dtype(dtype) => {}
+                Some(dtype) => bail!(
+                    "Configured numeric feature '{}' has non-numeric type {:?}",
+                    name,
+                    dtype
+                ),
+                None => bail!(
+                    "Configured numeric feature '{}' not found in combined schema",
+                    name
+                ),
+            }
+        }
+        for name in &enum_names {
+            match schema.get(name) {
+                Some(dtype) if matches!(dtype, DataType::String) || dtype.is_categorical() => {}
+                Some(dtype) => bail!(
+                    "Configured enum feature '{}' has unexpected type {:?}",
+                    name,
+                    dtype
+                ),
+                None => bail!(
+                    "Configured enum feature '{}' not found in combined schema",
+                    name
+                ),
+            }
+        }
+
+        // Combine numeric and enum feature names (numeric first, then enum)
+        let feature_names: Vec<String> = numeric_names
+            .iter()
+            .map(|name| name.to_string())
+            .chain(enum_names.iter().map(|name| name.to_string()))
+            .collect();
+        let num_features = feature_names.len();
+        let num_numeric = numeric_names.len();
+        tracing::info!(
+            numeric = num_numeric,
+            enums = enum_names.len(),
+            total = num_features,
+            "Feature columns from config"
+        );
+
+        // Build select expressions for the combined DataFrame
+        let mut select_exprs: Vec<polars::prelude::Expr> = vec![];
+        select_exprs.push(col("lat").cast(DataType::Float32));
+        select_exprs.push(col("lon").cast(DataType::Float32));
+
+        // Select numeric features as Float32 (datetime columns → fractional year)
+        for name in &numeric_names {
+            if is_datetime_dtype(schema.get(name.as_str()).unwrap()) {
+                select_exprs.push(
+                    (col(name.as_str()).dt().year().cast(DataType::Float32)
+                        + (col(name.as_str()).dt().month().cast(DataType::Float32) - lit(1.0f32))
+                            / lit(12.0f32))
+                    .alias(name.as_str()),
+                );
+            } else {
+                select_exprs.push(col(name.as_str()).cast(DataType::Float32));
+            }
+        }
+
+        // String columns for address/postcode and property metadata
+        for &string_col_name in &[
+            "Address per Property Register",
+            "Address per EPC",
+            "Postcode",
+            "Property sub-type",
+            "Price qualifier",
+        ] {
+            if schema.get(string_col_name).is_some() {
+                select_exprs.push(col(string_col_name).cast(DataType::String));
+            }
+        }
+
+        // Enum features as String
+        for &name in &enum_names {
+            select_exprs.push(col(name).cast(DataType::String));
+        }
+
+        // Optional columns
+        let has_approx_col = schema.get("Is construction date approximate").is_some();
+        if has_approx_col {
+            select_exprs.push(col("Is construction date approximate").cast(DataType::Float32));
+        }
+        let has_renovation_history = schema.get("renovation_history").is_some();
+        if has_renovation_history {
+            select_exprs.push(col("renovation_history"));
+        }
+        let has_historical_prices = schema.get("historical_prices").is_some();
+        if has_historical_prices {
+            select_exprs.push(col("historical_prices"));
+        }
+        let df = combined_lf
+            .filter(col("lat").is_not_null().and(col("lon").is_not_null()))
+            .select(select_exprs)
+            .collect()
+            .context("Failed to select columns from joined frame")?;
+
+        let row_count = df.height();
+        if row_count == 0 {
+            bail!("No property rows have usable coordinates after joining postcode data");
+        }
+        tracing::info!(rows = row_count, "Combined data selected");
+
+        let lat_series = df
+            .column("lat")
+            .context("Missing 'lat' column")?
+            .cast(&DataType::Float32)
+            .context("Failed to cast 'lat' to Float32")?;
+        let lat: Vec<f32> = lat_series
+            .f32()
+            .context("Failed to read 'lat' as f32")?
+            .into_iter()
+            .map(|value| value.context("Missing 'lat' value after coordinate filter"))
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        let lon_series = df
+            .column("lon")
+            .context("Missing 'lon' column")?
+            .cast(&DataType::Float32)
+            .context("Failed to cast 'lon' to Float32")?;
+        let lon: Vec<f32> = lon_series
+            .f32()
+            .context("Failed to read 'lon' as f32")?
+            .into_iter()
+            .map(|value| value.context("Missing 'lon' value after coordinate filter"))
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        for (row, (&latitude, &longitude)) in lat.iter().zip(&lon).enumerate() {
+            if !(-90.0..=90.0).contains(&latitude) || !(-180.0..=180.0).contains(&longitude) {
+                bail!("Invalid coordinates at row {row}: lat={latitude}, lon={longitude}");
+            }
+        }
+
+        tracing::info!("Extracting numeric feature columns");
+        let numeric_col_major: Vec<Vec<f32>> = numeric_names
+            .par_iter()
+            .map(|name| {
+                let column = df
+                    .column(name.as_str())
+                    .with_context(|| format!("Missing feature column '{name}'"))?;
+                column_to_f32_vec(column)
+            })
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        tracing::info!("Computing histograms for numeric features");
+        let numeric_feature_stats: Vec<FeatureStats> = numeric_col_major
+            .par_iter()
+            .enumerate()
+            .map(|(feat_index, vals)| {
+                let name = numeric_names[feat_index].as_str();
+                let bounds = features::bounds_for(name)
+                    .with_context(|| format!("No bounds config for feature '{}'", name))?;
+                let stats = compute_feature_stats(vals, &bounds, features::has_integer_bins(name));
+                tracing::debug!(
+                    feature = %name,
+                    slider_min = format_args!("{:.2}", stats.slider_min),
+                    slider_max = format_args!("{:.2}", stats.slider_max),
+                    bins = stats.histogram.counts.len(),
+                    "Feature stats"
+                );
+                Ok(stats)
+            })
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        // Compute quantization parameters from feature stats (numeric features).
+        // For features with Fixed bounds, use those bounds so the full configured range
+        // is representable — the histogram refinement can narrow min/max to exclude
+        // "outliers" that are actually valid data (e.g. ethnicity percentages).
+        // For Percentile-bounded features, use the (possibly refined) histogram range
+        // so extreme outliers don't destroy precision for the main distribution.
+        let mut quant_min = Vec::with_capacity(num_features);
+        let mut quant_range = Vec::with_capacity(num_features);
+        for (feat_idx, stats) in numeric_feature_stats.iter().enumerate() {
+            let (min, max) = match features::bounds_for(numeric_names[feat_idx].as_str()) {
+                Some(Bounds::Fixed { min, max }) => (min, max),
+                _ => (stats.histogram.min, stats.histogram.max),
+            };
+            quant_min.push(min);
+            quant_range.push(if max > min { max - min } else { 0.0 });
+        }
+
+        tracing::info!("Extracting string columns");
+        let extract_string_col = |df: &DataFrame, name: &str| -> anyhow::Result<Vec<String>> {
+            let column = df
+                .column(name)
+                .with_context(|| format!("Required column '{name}' not found in parquet"))?;
+            let string_column = column
+                .str()
+                .with_context(|| format!("Column '{name}' is not a string column"))?;
+            string_column
+                .into_iter()
+                .enumerate()
+                .map(|(row, value)| {
+                    value
+                        .map(ToString::to_string)
+                        .with_context(|| format!("Required column '{name}' has null at row {row}"))
+                })
+                .collect()
+        };
+
+        let address_raw = extract_string_col(&df, "Address per Property Register")?;
+        let postcode_raw = extract_string_col(&df, "Postcode")?;
+
+        // Extract optional string columns
+        let extract_optional_string_col =
+            |df: &DataFrame, name: &str| -> anyhow::Result<Vec<Option<String>>> {
+                if let Ok(column) = df.column(name) {
+                    let string_column = column
+                        .str()
+                        .with_context(|| format!("Column '{name}' is not a string column"))?;
+                    Ok(string_column
+                        .into_iter()
+                        .map(|value| {
+                            value.and_then(|s| {
+                                let trimmed = s.trim();
+                                if trimmed.is_empty() {
+                                    None
+                                } else {
+                                    Some(trimmed.to_string())
+                                }
+                            })
+                        })
+                        .collect())
+                } else {
+                    Ok(vec![None; row_count])
+                }
+            };
+        let property_sub_type_raw = extract_optional_string_col(&df, "Property sub-type")?;
+        let price_qualifier_raw = extract_optional_string_col(&df, "Price qualifier")?;
+
+        tracing::info!("Building enum features");
+        // enum_col_major: Vec<(values_list, encoded_as_f32)>
+        let enum_col_major: Vec<(Vec<String>, Vec<f32>)> = enum_names
+            .par_iter()
+            .map(|&name| -> anyhow::Result<(Vec<String>, Vec<f32>)> {
+                let column_data = df
+                    .column(name)
+                    .with_context(|| format!("Required enum column '{name}' not found"))?;
+                let string_column = column_data
+                    .str()
+                    .with_context(|| format!("Enum column '{name}' is not a string column"))?;
+                let unique_set: std::collections::HashSet<String> = string_column
+                    .into_iter()
+                    .filter_map(|value| {
+                        let text = value?.trim();
+                        (!text.is_empty()).then(|| text.to_string())
+                    })
+                    .collect();
+
+                // Use configured order if available, otherwise alphabetical
+                let unique: Vec<String> = if let Some(order) = features::order_for(name) {
+                    let mut ordered: Vec<String> = Vec::new();
+                    for &ordered_value in order {
+                        if unique_set.contains(ordered_value) {
+                            ordered.push(ordered_value.to_string());
+                        }
+                    }
+                    // Append any values not in the configured order, alphabetically
+                    // Use HashSet for O(1) contains instead of O(n) slice search
+                    let order_set: rustc_hash::FxHashSet<&str> = order.iter().copied().collect();
+                    let mut remainder: Vec<String> = unique_set
+                        .iter()
+                        .filter(|value| !order_set.contains(value.as_str()))
+                        .cloned()
+                        .collect();
+                    remainder.sort();
+                    ordered.extend(remainder);
+                    ordered
+                } else {
+                    let mut sorted: Vec<String> = unique_set.into_iter().collect();
+                    sorted.sort();
+                    sorted
+                };
+
+                let value_to_idx: std::collections::HashMap<&str, f32> = unique
+                    .iter()
+                    .enumerate()
+                    .map(|(index, value)| (value.as_str(), index as f32))
+                    .collect();
+
+                let encoded: Vec<f32> = string_column
+                    .into_iter()
+                    .enumerate()
+                    .map(|(row, value)| {
+                        let Some(text) = value.map(str::trim).filter(|text| !text.is_empty())
+                        else {
+                            return Ok(f32::NAN);
+                        };
+                        value_to_idx.get(text).copied().with_context(|| {
+                            format!("Enum column '{name}' has unknown value '{text}' at row {row}")
+                        })
+                    })
+                    .collect::<anyhow::Result<Vec<_>>>()?;
+
+                tracing::debug!(column = %name, unique_values = unique.len(), "Enum feature encoded as f32");
+                Ok((unique, encoded))
+            })
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        // Extract is_approx_build_date: 0.0 = exact, anything else (1.0/NaN) = approximate
+        let is_approx_build_date_raw: Vec<bool> = if has_approx_col {
+            let column_data = df
+                .column("Is construction date approximate")
+                .context("Missing 'Is construction date approximate' column")?;
+            let float_series = column_data
+                .cast(&DataType::Float32)
+                .context("Failed to cast 'Is construction date approximate' to Float32")?;
+            let chunked = float_series
+                .f32()
+                .context("Failed to read 'Is construction date approximate' as f32")?;
+            chunked
+                .into_iter()
+                .map(|value| match value {
+                    Some(0.0) => false,
+                    _ => true, // 1.0 or NaN → approximate
+                })
+                .collect()
+        } else {
+            vec![true; row_count] // default: all approximate
+        };
+
+        // Extract renovation_history: List<Struct{year: i32, event: str}>
+        let mut renovation_raw: FxHashMap<u32, Vec<RenovationEvent>> = if has_renovation_history {
+            tracing::info!("Extracting renovation history");
+            let reno_col = df
+                .column("renovation_history")
+                .context("Missing renovation_history column")?;
+            let list_ca = reno_col
+                .list()
+                .context("renovation_history is not a list column")?;
+
+            let mut history: FxHashMap<u32, Vec<RenovationEvent>> = FxHashMap::default();
+            for old_row in 0..row_count {
+                if let Some(inner) = list_ca.get_as_series(old_row) {
+                    if inner.is_empty() {
+                        continue;
+                    }
+                    let structs = inner
+                        .struct_()
+                        .context("renovation_history inner is not a struct")?;
+                    let years = structs
+                        .field_by_name("year")
+                        .context("Missing 'year' field in renovation_history struct")?;
+                    let events = structs
+                        .field_by_name("event")
+                        .context("Missing 'event' field in renovation_history struct")?;
+
+                    let mut row_events = Vec::new();
+                    for idx in 0..inner.len() {
+                        let year = years.get(idx).context("Failed to get year value")?;
+                        let event = events.get(idx).context("Failed to get event value")?;
+                        if let (AnyValue::Int32(yr), AnyValue::String(ev)) = (&year, &event) {
+                            row_events.push(RenovationEvent {
+                                year: *yr,
+                                event: ev.to_string(),
+                            });
+                        }
+                    }
+                    if !row_events.is_empty() {
+                        history.insert(old_row as u32, row_events);
+                    }
+                }
+            }
+            tracing::info!(
+                properties_with_events = history.len(),
+                "Renovation history extracted"
+            );
+            history
+        } else {
+            FxHashMap::default()
+        };
+
+        // Extract historical_prices: List<Struct{year: i32, month: u8, price: i64}>
+        let mut historical_prices_raw: FxHashMap<u32, Vec<HistoricalPrice>> =
+            if has_historical_prices {
+                tracing::info!("Extracting historical prices");
+                let prices_col = df
+                    .column("historical_prices")
+                    .context("Missing historical_prices column")?;
+                let list_ca = prices_col
+                    .list()
+                    .context("historical_prices is not a list column")?;
+
+                let mut history: FxHashMap<u32, Vec<HistoricalPrice>> = FxHashMap::default();
+                for old_row in 0..row_count {
+                    if let Some(inner) = list_ca.get_as_series(old_row) {
+                        if inner.is_empty() {
+                            continue;
+                        }
+                        let structs = inner
+                            .struct_()
+                            .context("historical_prices inner is not a struct")?;
+                        let years = structs
+                            .field_by_name("year")
+                            .context("Missing 'year' field in historical_prices struct")?;
+                        let months = structs
+                            .field_by_name("month")
+                            .context("Missing 'month' field in historical_prices struct")?;
+                        let prices = structs
+                            .field_by_name("price")
+                            .context("Missing 'price' field in historical_prices struct")?;
+
+                        let mut row_prices = Vec::new();
+                        for idx in 0..inner.len() {
+                            let year = years.get(idx).context("Failed to get year value")?;
+                            let month = months.get(idx).context("Failed to get month value")?;
+                            let price = prices.get(idx).context("Failed to get price value")?;
+                            let AnyValue::Int32(year_i32) = year else {
+                                bail!("historical_prices.year is not Int32 at row {old_row}, got {year:?}");
+                            };
+                            let AnyValue::UInt8(month_u8) = month else {
+                                bail!("historical_prices.month is not UInt8 at row {old_row}, got {month:?}");
+                            };
+                            let AnyValue::Int64(price_i64) = price else {
+                                bail!("historical_prices.price is not Int64 at row {old_row}, got {price:?}");
+                            };
+                            row_prices.push(HistoricalPrice {
+                                year: year_i32,
+                                month: month_u8,
+                                price: price_i64,
+                            });
+                        }
+                        if !row_prices.is_empty() {
+                            history.insert(old_row as u32, row_prices);
+                        }
+                    }
+                }
+                tracing::info!(
+                    properties_with_prices = history.len(),
+                    "Historical prices extracted"
+                );
+                history
+            } else {
+                FxHashMap::default()
+            };
+
+        // Free the projected joined frame before building the row-major matrix.
+        drop(df);
+
+        // Sort all rows by spatial locality so that grid queries access
+        // contiguous memory (sequential reads instead of random DRAM accesses).
+        tracing::info!("Sorting rows by spatial locality");
+        let grid_cell_size = 0.01_f32;
+        let min_lat_val = lat.iter().cloned().fold(f32::INFINITY, f32::min) - grid_cell_size;
+        let min_lon_val = lon.iter().cloned().fold(f32::INFINITY, f32::min) - grid_cell_size;
+        let max_lon_val = lon.iter().cloned().fold(f32::NEG_INFINITY, f32::max) + grid_cell_size;
+        let grid_cols = ((max_lon_val - min_lon_val) / grid_cell_size).ceil() as u64 + 1;
+
+        let mut perm: Vec<u32> = (0..row_count as u32).collect();
+        perm.par_sort_unstable_by_key(|&perm_index| {
+            let grid_row = ((lat[perm_index as usize] - min_lat_val) / grid_cell_size) as u64;
+            let grid_col = ((lon[perm_index as usize] - min_lon_val) / grid_cell_size) as u64;
+            grid_row * grid_cols + grid_col
+        });
+
+        let lat: Vec<f32> = perm
+            .iter()
+            .map(|&perm_index| lat[perm_index as usize])
+            .collect();
+        let lon: Vec<f32> = perm
+            .iter()
+            .map(|&perm_index| lon[perm_index as usize])
+            .collect();
+        let last_known_price_raw: Vec<f32> = numeric_names
+            .iter()
+            .position(|name| name == "Last known price")
+            .map(|price_idx| {
+                perm.iter()
+                    .map(|&perm_index| numeric_col_major[price_idx][perm_index as usize])
+                    .collect()
+            })
+            .context("Required numeric column 'Last known price' not configured")?;
+
+        // Build contiguous address buffer and address search index (permuted)
+        tracing::info!("Building interned strings");
+        let total_addr_bytes: usize = address_raw.iter().map(|text| text.len()).sum();
+        let mut address_buffer = String::with_capacity(total_addr_bytes);
+        let mut address_offsets = Vec::with_capacity(row_count);
+        let mut address_lengths = Vec::with_capacity(row_count);
+        let mut address_token_index: FxHashMap<String, Vec<u32>> = FxHashMap::default();
+        let mut address_search_rodeo = lasso::Rodeo::default();
+        let mut address_search_token_keys: Vec<lasso::Spur> = Vec::new();
+        let mut address_search_token_offsets = Vec::with_capacity(row_count);
+        let mut address_search_token_lengths = Vec::with_capacity(row_count);
+        for (new_row, &perm_index) in perm.iter().enumerate() {
+            let addr = &address_raw[perm_index as usize];
+            let offset = address_buffer.len() as u32;
+            let length = addr.len().min(u16::MAX as usize) as u16;
+            address_offsets.push(offset);
+            address_lengths.push(length);
+            address_buffer.push_str(&addr[..length as usize]);
+
+            let search_tokens = address_search_tokens(addr);
+            let token_offset = address_search_token_keys.len() as u32;
+            let token_length = search_tokens.len().min(u16::MAX as usize) as u16;
+            address_search_token_offsets.push(token_offset);
+            address_search_token_lengths.push(token_length);
+
+            for token in search_tokens.iter().take(token_length as usize) {
+                let key = address_search_rodeo.get_or_intern(token);
+                address_search_token_keys.push(key);
+
+                if is_address_candidate_token(token) {
+                    address_token_index
+                        .entry(token.clone())
+                        .or_default()
+                        .push(new_row as u32);
+                }
+            }
+        }
+        // Keep every distinctive token: common road words ("high", "church", "station") are
+        // exactly what people search, and dropping them made those roads unsearchable while a
+        // prefix fallback surfaced the wrong street ("Highbury" for "High"). The candidate scan
+        // is bounded per query instead (ADDRESS_SEARCH_CANDIDATE_LIMIT), and stop words are
+        // already excluded from the index, so the largest posting lists stay modest.
+        let max_postings = address_token_index
+            .values()
+            .map(Vec::len)
+            .max()
+            .unwrap_or(0);
+        let address_prefix_index = build_address_prefix_index(&address_token_index);
+        let address_search_interner = address_search_rodeo.into_reader();
+        let address_postings_count: usize = address_token_index.values().map(Vec::len).sum();
+        tracing::info!(
+            tokens = address_token_index.len(),
+            prefixes = address_prefix_index.len(),
+            max_postings_per_token = max_postings,
+            postings = address_postings_count,
+            row_tokens = address_search_token_keys.len(),
+            "Address search index built"
+        );
+
+        // Intern postcodes (permuted)
+        let mut postcode_rodeo = lasso::Rodeo::default();
+        let mut postcode_keys: Vec<lasso::Spur> = Vec::with_capacity(row_count);
+        let mut postcode_row_index: FxHashMap<lasso::Spur, Vec<u32>> = FxHashMap::default();
+        for (new_row, &perm_index) in perm.iter().enumerate() {
+            let key = postcode_rodeo.get_or_intern(&postcode_raw[perm_index as usize]);
+            postcode_keys.push(key);
+            postcode_row_index
+                .entry(key)
+                .or_default()
+                .push(new_row as u32);
+        }
+        let postcode_interner = postcode_rodeo.into_reader();
+
+        let row_to_poi_metric_idx: Vec<u32> = if poi_metrics.is_empty() {
+            vec![NO_POI_METRIC_ROW; row_count]
+        } else {
+            perm.iter()
+                .map(|&old_row| {
+                    poi_metric_by_postcode
+                        .get(postcode_raw[old_row as usize].as_str())
+                        .copied()
+                        .unwrap_or(NO_POI_METRIC_ROW)
+                })
+                .collect()
+        };
+        poi_metrics.set_row_mapping(row_to_poi_metric_idx);
+
+        // Pack is_approx_build_date into a bitvec (8 bools per byte)
+        let num_bytes = row_count.div_ceil(8);
+        let mut approx_build_date_bits = vec![0u8; num_bytes];
+        for (new_row, &old_row) in perm.iter().enumerate() {
+            if is_approx_build_date_raw[old_row as usize] {
+                approx_build_date_bits[new_row / 8] |= 1 << (new_row % 8);
+            }
+        }
+
+        // Re-key renovation_history by permuted row index
+        let renovation_history: FxHashMap<u32, Vec<RenovationEvent>> = {
+            let mut map =
+                FxHashMap::with_capacity_and_hasher(renovation_raw.len(), Default::default());
+            for (new_row, &old_row) in perm.iter().enumerate() {
+                if let Some(events) = renovation_raw.remove(&old_row) {
+                    map.insert(new_row as u32, events);
+                }
+            }
+            map
+        };
+
+        // Re-key historical_prices by permuted row index
+        let historical_prices: FxHashMap<u32, Vec<HistoricalPrice>> = {
+            let mut map = FxHashMap::with_capacity_and_hasher(
+                historical_prices_raw.len(),
+                Default::default(),
+            );
+            for (new_row, &old_row) in perm.iter().enumerate() {
+                if let Some(prices) = historical_prices_raw.remove(&old_row) {
+                    map.insert(new_row as u32, prices);
+                }
+            }
+            map
+        };
+
+        // Permute optional string columns into sparse HashMaps
+        let property_sub_type: FxHashMap<u32, String> = {
+            let mut map = FxHashMap::default();
+            for (new_row, &old_row) in perm.iter().enumerate() {
+                if let Some(ref s) = property_sub_type_raw[old_row as usize] {
+                    map.insert(new_row as u32, s.clone());
+                }
+            }
+            map
+        };
+        let price_qualifier: FxHashMap<u32, String> = {
+            let mut map = FxHashMap::default();
+            for (new_row, &old_row) in perm.iter().enumerate() {
+                if let Some(ref s) = price_qualifier_raw[old_row as usize] {
+                    map.insert(new_row as u32, s.clone());
+                }
+            }
+            map
+        };
+
+        // Build enum_values map: feature_index -> list of string values
+        // and enum_counts map: feature_index -> per-value global counts
+        let mut enum_values: rustc_hash::FxHashMap<usize, Vec<String>> =
+            rustc_hash::FxHashMap::default();
+        let mut enum_counts: rustc_hash::FxHashMap<usize, Vec<u64>> =
+            rustc_hash::FxHashMap::default();
+        for (enum_idx, (values, encoded)) in enum_col_major.iter().enumerate() {
+            let feature_idx = num_numeric + enum_idx;
+            enum_values.insert(feature_idx, values.clone());
+            let mut counts = vec![0u64; values.len()];
+            for &val in encoded {
+                if val.is_finite() {
+                    let idx = val as usize;
+                    if idx < counts.len() {
+                        counts[idx] += 1;
+                    }
+                }
+            }
+            enum_counts.insert(feature_idx, counts);
+        }
+
+        // Build feature_stats: numeric stats + placeholder stats for enums
+        let mut feature_stats = numeric_feature_stats;
+        for (values, _) in &enum_col_major {
+            // For enum features, slider range is 0 to num_values-1
+            let num_values = values.len();
+            let max_val = num_values as f32;
+            feature_stats.push(FeatureStats {
+                slider_min: 0.0,
+                slider_max: (num_values.saturating_sub(1)) as f32,
+                histogram: Histogram {
+                    min: 0.0,
+                    max: max_val,
+                    p1: 0.0,
+                    p99: max_val,
+                    counts: vec![0; num_values.max(1)],
+                },
+            });
+            // Enum features: not quantized, stored directly as u16
+            quant_min.push(0.0);
+            quant_range.push(0.0);
+        }
+        let dequant_a: Vec<f32> = quant_range
+            .iter()
+            .map(|&r| if r > 0.0 { r / QUANT_SCALE } else { 0.0 })
+            .collect();
+
+        // Transpose to row-major AND apply spatial permutation in one pass.
+        // Combines numeric and enum features into a single feature_data array, quantized to u16.
+        tracing::info!("Transposing to row-major layout (spatially sorted, quantized to u16)");
+        let mut feature_data = vec![NAN_U16; row_count * num_features];
+        feature_data
+            .par_chunks_mut(num_features)
+            .enumerate()
+            .for_each(|(new_row, row_slice)| {
+                let old_index = perm[new_row] as usize;
+                // Numeric features: quantize to u16
+                for (feat_idx, col_vec) in numeric_col_major.iter().enumerate() {
+                    let value = col_vec[old_index];
+                    row_slice[feat_idx] = if value.is_finite() {
+                        let range = quant_range[feat_idx];
+                        if range > 0.0 {
+                            let normalized = (value - quant_min[feat_idx]) / range;
+                            (normalized * QUANT_SCALE).round().clamp(0.0, QUANT_SCALE) as u16
+                        } else {
+                            0
+                        }
+                    } else {
+                        NAN_U16
+                    };
+                }
+                // Enum features: store as u16 directly
+                for (enum_idx, (_, encoded)) in enum_col_major.iter().enumerate() {
+                    let value = encoded[old_index];
+                    row_slice[num_numeric + enum_idx] = if value.is_finite() {
+                        value as u16
+                    } else {
+                        NAN_U16
+                    };
+                }
+            });
+
+        tracing::info!("Data loading complete");
+
+        Ok(PropertyData {
+            lat,
+            lon,
+            feature_names,
+            num_features,
+            num_numeric,
+            feature_data,
+            dequant_a,
+            quant_min,
+            quant_range,
+            feature_stats,
+            poi_metrics,
+            last_known_price_raw,
+            address_buffer,
+            address_offsets,
+            address_lengths,
+            postcode_interner,
+            postcode_keys,
+            postcode_row_index,
+            address_token_index,
+            address_prefix_index,
+            address_search_interner,
+            address_search_token_keys,
+            address_search_token_offsets,
+            address_search_token_lengths,
+            enum_values,
+            enum_counts,
+            approx_build_date_bits,
+            renovation_history,
+            historical_prices,
+            property_sub_type,
+            price_qualifier,
+        })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn country_column_detection_prefers_reliable_country_code() {
+        let df = df!(
+            "Postcode" => &["SW1A 1AA"],
+            "ctry25cd" => &["E92000001"],
+            "lat" => &[Some(51.501_f64)],
+            "lon" => &[Some(-0.141_f64)],
+        )
+        .expect("test dataframe should build");
+
+        assert_eq!(
+            find_country_column(df.schema()).as_deref(),
+            Some("ctry25cd")
+        );
+    }
+
+    #[test]
+    fn missing_coordinate_summary_counts_rows_and_distinct_samples() {
+        let summary = summarize_missing_coordinate_postcodes([
+            Some("SW1A 1AA"),
+            Some("SW1A 1AA"),
+            Some("E14 2DG"),
+            Some(""),
+            None,
+        ]);
+
+        assert_eq!(
+            summary,
+            MissingCoordinateSummary {
+                row_count: 5,
+                unique_postcode_count: 2,
+                sample_postcodes: vec!["E14 2DG".to_string(), "SW1A 1AA".to_string()],
+            }
+        );
+    }
+
+    #[test]
+    fn missing_england_coordinates_error_includes_counts_and_samples() {
+        let summary = MissingCoordinateSummary {
+            row_count: 3,
+            unique_postcode_count: 2,
+            sample_postcodes: vec!["E14 2DG".to_string(), "SW1A 1AA".to_string()],
+        };
+
+        let message = missing_england_coordinates_error(&summary, "ctry25cd");
+
+        assert!(message.contains("3 rows across 2 postcodes"));
+        assert!(message.contains("country column 'ctry25cd'"));
+        assert!(message.contains("E14 2DG, SW1A 1AA"));
+    }
+
+    #[test]
+    fn coordinate_validation_errors_for_england_rows() {
+        let lf = df!(
+            "Postcode" => &["SW1A 1AA", "E14 2DG"],
+            "ctry25cd" => &["E92000001", "E92000001"],
+            "lat" => &[Some(51.501_f64), None],
+            "lon" => &[Some(-0.141_f64), Some(-0.001_f64)],
+        )
+        .expect("test dataframe should build")
+        .lazy();
+        let schema = lf.clone().collect_schema().expect("schema should collect");
+
+        let err = validate_no_england_rows_missing_coordinates(&lf, &schema)
+            .expect_err("England row with missing coordinate should error");
+        let message = err.to_string();
+
+        assert!(message.contains("1 rows across 1 postcodes"));
+        assert!(message.contains("E14 2DG"));
+    }
+
+    #[test]
+    fn coordinate_validation_allows_non_england_rows() {
+        let lf = df!(
+            "Postcode" => &["CF10 1AA", "SW1A 1AA"],
+            "ctry25cd" => &["W92000004", "E92000001"],
+            "lat" => &[None, Some(51.501_f64)],
+            "lon" => &[Some(-3.179_f64), Some(-0.141_f64)],
+        )
+        .expect("test dataframe should build")
+        .lazy();
+        let schema = lf.clone().collect_schema().expect("schema should collect");
+
+        validate_no_england_rows_missing_coordinates(&lf, &schema)
+            .expect("non-England row with missing coordinate should be skipped");
+    }
+
+    #[test]
+    fn coordinate_validation_requires_country_column() {
+        let lf = df!(
+            "Postcode" => &["SW1A 1AA"],
+            "lat" => &[None::<f64>],
+            "lon" => &[Some(-0.141_f64)],
+        )
+        .expect("test dataframe should build")
+        .lazy();
+        let schema = lf.clone().collect_schema().expect("schema should collect");
+
+        let err = validate_no_england_rows_missing_coordinates(&lf, &schema)
+            .expect_err("missing country provenance should error");
+
+        assert!(err.to_string().contains("no reliable country column"));
+    }
+
+    #[test]
+    fn enum_value_counting() {
+        let values = vec![0.0_f32, 1.0, 1.0, 2.0, f32::NAN, 3.0, 1.0];
+        let enum_count = 4;
+
+        let mut counts = vec![0u64; enum_count];
+        for &v in &values {
+            if v.is_finite() {
+                let idx = v as usize;
+                if idx < enum_count {
+                    counts[idx] += 1;
+                }
+            }
+        }
+
+        assert_eq!(counts[0], 1);
+        assert_eq!(counts[1], 3);
+        assert_eq!(counts[2], 1);
+        assert_eq!(counts[3], 1);
+    }
+}
diff --git a/server-rs/src/data/property/mod.rs b/server-rs/src/data/property/mod.rs
new file mode 100644
index 0000000..17b39c7
--- /dev/null
+++ b/server-rs/src/data/property/mod.rs
@@ -0,0 +1,238 @@
+//! Property data: the row-major quantized feature matrix plus the side tables
+//! (addresses, postcodes, renovation/price history, POI metrics) built from the
+//! properties + postcode parquet files.
+//!
+//! Split by concern:
+//! - [`loading`]: parquet ingestion, validation, spatial sort and matrix build
+//! - [`stats`]: histograms, percentiles and slider-bound computation
+//! - [`quant`]: u16 quantization encode/decode
+//! - [`poi_metrics`]: postcode-level POI metric side table
+//! - [`address_search`]: address tokenization, indexing and ranked search
+//! - [`h3`]: H3 cell precomputation
+
+mod address_search;
+mod h3;
+mod loading;
+mod poi_metrics;
+mod quant;
+mod stats;
+
+pub use h3::precompute_h3;
+pub use poi_metrics::PostcodePoiMetrics;
+pub use quant::QuantRef;
+pub use stats::{FeatureStats, Histogram};
+
+use rustc_hash::FxHashMap;
+use serde::Serialize;
+
+use crate::consts::NAN_U16;
+
+#[derive(Serialize, Clone)]
+pub struct RenovationEvent {
+    pub year: i32,
+    pub event: String,
+}
+
+#[derive(Serialize, Clone)]
+pub struct HistoricalPrice {
+    pub year: i32,
+    pub month: u8,
+    pub price: i64,
+}
+
+pub struct PropertyData {
+    pub lat: Vec<f32>,
+    pub lon: Vec<f32>,
+    pub feature_names: Vec<String>,
+    pub num_features: usize,
+    /// Number of numeric features (enum features start at this index).
+    pub num_numeric: usize,
+    /// Row-major flat array: feature_data[row * num_features + feat_idx].
+    /// Quantized to u16. NaN sentinel = u16::MAX (65535).
+    /// Numeric features: encoded via (val - min) / range * 65534.
+    /// Enum features: stored directly as u16 cast of the f32 index.
+    pub feature_data: Vec<u16>,
+    /// Per-feature: range / QUANT_SCALE for fast decode.
+    dequant_a: Vec<f32>,
+    /// Per-feature: minimum value (offset for dequantization).
+    quant_min: Vec<f32>,
+    /// Per-feature: max - min (for encoding filter bounds).
+    quant_range: Vec<f32>,
+    pub feature_stats: Vec<FeatureStats>,
+    pub poi_metrics: PostcodePoiMetrics,
+    /// Unquantized last sale price used by the price-history chart.
+    last_known_price_raw: Vec<f32>,
+    /// Contiguous buffer holding all address strings end-to-end.
+    address_buffer: String,
+    /// Byte offset into `address_buffer` where each row's address starts.
+    address_offsets: Vec<u32>,
+    /// Length in bytes of each row's address.
+    address_lengths: Vec<u16>,
+    /// Interned postcodes: reader is thread-safe, keys index into it.
+    postcode_interner: lasso::RodeoReader,
+    postcode_keys: Vec<lasso::Spur>,
+    /// Rows for each postcode, keyed by the interned postcode key.
+    postcode_row_index: FxHashMap<lasso::Spur, Vec<u32>>,
+    /// Inverted index from address tokens to property rows.
+    address_token_index: FxHashMap<String, Vec<u32>>,
+    /// Prefix lookup from typed address-token prefix to indexed full address tokens.
+    address_prefix_index: FxHashMap<String, Vec<String>>,
+    /// Interned normalized address-search tokens used for per-row scoring.
+    address_search_interner: lasso::RodeoReader,
+    /// Flat per-row normalized address-search token keys.
+    address_search_token_keys: Vec<lasso::Spur>,
+    /// Offset into `address_search_token_keys` for each row.
+    address_search_token_offsets: Vec<u32>,
+    /// Number of normalized address-search token keys for each row.
+    address_search_token_lengths: Vec<u16>,
+    /// For enum features: maps feature index to list of possible string values.
+    /// Index in values list corresponds to the u16 value stored in feature_data.
+    pub enum_values: rustc_hash::FxHashMap<usize, Vec<String>>,
+    /// For enum features: maps feature index to per-value global counts (same order as enum_values).
+    pub enum_counts: rustc_hash::FxHashMap<usize, Vec<u64>>,
+    /// Per-row flag: true = construction date is approximate (from EPC band),
+    /// false = exact (from new-build transaction date).
+    /// Bit-packed: byte `row / 8`, bit `row % 8`. 8x smaller than Vec<bool>.
+    approx_build_date_bits: Vec<u8>,
+    /// Per-row renovation events. Keyed by (permuted) row index.
+    /// Only rows with events are present in the map.
+    renovation_history: FxHashMap<u32, Vec<RenovationEvent>>,
+    /// Per-row historical sale transactions (Land Registry price-paid).
+    /// Keyed by (permuted) row index. Only rows with prices are present.
+    historical_prices: FxHashMap<u32, Vec<HistoricalPrice>>,
+    property_sub_type: FxHashMap<u32, String>,
+    price_qualifier: FxHashMap<u32, String>,
+}
+
+impl PropertyData {
+    /// Get the address string for a given row.
+    pub fn address(&self, row: usize) -> &str {
+        let offset = self.address_offsets[row] as usize;
+        let length = self.address_lengths[row] as usize;
+        &self.address_buffer[offset..offset + length]
+    }
+
+    /// Get the postcode string for a given row.
+    pub fn postcode(&self, row: usize) -> &str {
+        self.postcode_interner.resolve(&self.postcode_keys[row])
+    }
+
+    /// Get postcode components for field-level borrowing (avoids conflicting borrows with feature_data).
+    pub fn postcode_parts(&self) -> (&lasso::RodeoReader, &[lasso::Spur]) {
+        (&self.postcode_interner, &self.postcode_keys)
+    }
+
+    /// Property rows for a given postcode string, or empty if unknown.
+    pub fn rows_for_postcode(&self, postcode: &str) -> &[u32] {
+        self.postcode_interner
+            .get(postcode)
+            .and_then(|key| self.postcode_row_index.get(&key))
+            .map(Vec::as_slice)
+            .unwrap_or(&[])
+    }
+
+    /// Get the is_approx_build_date flag for a given row (bit-packed).
+    pub fn is_approx_build_date(&self, row: usize) -> bool {
+        let byte = self.approx_build_date_bits[row / 8];
+        byte & (1 << (row % 8)) != 0
+    }
+
+    /// Get renovation events for a given row (empty slice if none).
+    pub fn renovation_history(&self, row: usize) -> &[RenovationEvent] {
+        self.renovation_history
+            .get(&(row as u32))
+            .map(|v| v.as_slice())
+            .unwrap_or(&[])
+    }
+
+    /// Get historical sale transactions for a given row (empty slice if none).
+    pub fn historical_prices(&self, row: usize) -> &[HistoricalPrice] {
+        self.historical_prices
+            .get(&(row as u32))
+            .map(|v| v.as_slice())
+            .unwrap_or(&[])
+    }
+
+    /// Get property sub-type for a given row.
+    pub fn property_sub_type(&self, row: usize) -> Option<&str> {
+        self.property_sub_type
+            .get(&(row as u32))
+            .map(String::as_str)
+    }
+
+    /// Get price qualifier for a given row.
+    pub fn price_qualifier(&self, row: usize) -> Option<&str> {
+        self.price_qualifier.get(&(row as u32)).map(String::as_str)
+    }
+
+    /// Get the unquantized last sale price for charting.
+    #[inline]
+    pub fn last_known_price_raw(&self, row: usize) -> f32 {
+        self.last_known_price_raw[row]
+    }
+
+    /// Decode a single feature value from quantized u16 storage.
+    #[inline]
+    pub fn get_feature(&self, row: usize, feat_idx: usize) -> f32 {
+        let raw = self.feature_data[row * self.num_features + feat_idx];
+        if raw == NAN_U16 {
+            return f32::NAN;
+        }
+        if feat_idx >= self.num_numeric {
+            raw as f32
+        } else {
+            raw as f32 * self.dequant_a[feat_idx] + self.quant_min[feat_idx]
+        }
+    }
+
+    /// Get a QuantRef for passing to aggregation/filter functions.
+    pub fn quant_ref(&self) -> QuantRef<'_> {
+        QuantRef {
+            dequant_a: &self.dequant_a,
+            quant_min: &self.quant_min,
+            quant_range: &self.quant_range,
+            num_numeric: self.num_numeric,
+        }
+    }
+}
+
+#[cfg(test)]
+impl PropertyData {
+    /// Minimal empty instance for integration tests that need an `AppState`
+    /// but never touch property data (e.g. checkout/webhook/invite flows).
+    pub(crate) fn empty_for_tests() -> Self {
+        PropertyData {
+            lat: Vec::new(),
+            lon: Vec::new(),
+            feature_names: Vec::new(),
+            num_features: 0,
+            num_numeric: 0,
+            feature_data: Vec::new(),
+            dequant_a: Vec::new(),
+            quant_min: Vec::new(),
+            quant_range: Vec::new(),
+            feature_stats: Vec::new(),
+            poi_metrics: PostcodePoiMetrics::empty(0),
+            last_known_price_raw: Vec::new(),
+            address_buffer: String::new(),
+            address_offsets: Vec::new(),
+            address_lengths: Vec::new(),
+            postcode_interner: lasso::Rodeo::default().into_reader(),
+            postcode_keys: Vec::new(),
+            postcode_row_index: FxHashMap::default(),
+            address_token_index: FxHashMap::default(),
+            address_prefix_index: FxHashMap::default(),
+            address_search_interner: lasso::Rodeo::default().into_reader(),
+            address_search_token_keys: Vec::new(),
+            address_search_token_offsets: Vec::new(),
+            address_search_token_lengths: Vec::new(),
+            enum_values: rustc_hash::FxHashMap::default(),
+            enum_counts: rustc_hash::FxHashMap::default(),
+            approx_build_date_bits: Vec::new(),
+            renovation_history: FxHashMap::default(),
+            historical_prices: FxHashMap::default(),
+            property_sub_type: FxHashMap::default(),
+            price_qualifier: FxHashMap::default(),
+        }
+    }
+}
diff --git a/server-rs/src/data/property/poi_metrics.rs b/server-rs/src/data/property/poi_metrics.rs
new file mode 100644
index 0000000..491eb77
--- /dev/null
+++ b/server-rs/src/data/property/poi_metrics.rs
@@ -0,0 +1,200 @@
+//! Postcode-level POI metric side table: dynamic POI features are stored once
+//! per postcode (not per property row) to keep the hot row-major feature matrix
+//! narrow, with a per-property row mapping for lookups.
+
+use anyhow::Context;
+use polars::prelude::*;
+use rayon::prelude::*;
+use rustc_hash::FxHashMap;
+
+use crate::consts::{NAN_U16, QUANT_SCALE};
+use crate::features::{self, Bounds};
+
+use super::quant::QuantRef;
+use super::stats::{column_to_f32_vec, compute_feature_stats, FeatureStats};
+
+pub(super) const NO_POI_METRIC_ROW: u32 = u32::MAX;
+
+pub struct PostcodePoiMetrics {
+    pub feature_names: Vec<String>,
+    pub name_to_index: FxHashMap<String, usize>,
+    /// Metric-major storage: columns[metric_idx][postcode_metric_idx].
+    pub columns: Vec<Vec<u16>>,
+    pub feature_stats: Vec<FeatureStats>,
+    /// Per-property row lookup into the postcode metric table.
+    row_to_metric_idx: Vec<u32>,
+    dequant_a: Vec<f32>,
+    quant_min: Vec<f32>,
+    quant_range: Vec<f32>,
+}
+
+impl PostcodePoiMetrics {
+    pub(super) fn empty(row_count: usize) -> Self {
+        Self {
+            feature_names: Vec::new(),
+            name_to_index: FxHashMap::default(),
+            columns: Vec::new(),
+            feature_stats: Vec::new(),
+            row_to_metric_idx: vec![NO_POI_METRIC_ROW; row_count],
+            dequant_a: Vec::new(),
+            quant_min: Vec::new(),
+            quant_range: Vec::new(),
+        }
+    }
+
+    pub(super) fn from_postcode_df(
+        df: &DataFrame,
+        feature_names: Vec<String>,
+    ) -> anyhow::Result<Self> {
+        if feature_names.is_empty() {
+            return Ok(Self::empty(0));
+        }
+
+        tracing::info!(
+            metrics = feature_names.len(),
+            postcodes = df.height(),
+            "Building postcode POI metric side table"
+        );
+
+        let col_major: Vec<Vec<f32>> = feature_names
+            .par_iter()
+            .map(|name| {
+                let column = df
+                    .column(name.as_str())
+                    .with_context(|| format!("Missing POI metric column '{name}'"))?;
+                column_to_f32_vec(column)
+            })
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        let feature_stats: Vec<FeatureStats> = col_major
+            .par_iter()
+            .enumerate()
+            .map(|(metric_idx, vals)| {
+                let name = feature_names[metric_idx].as_str();
+                let bounds = features::bounds_for(name)
+                    .with_context(|| format!("No bounds config for POI metric '{name}'"))?;
+                Ok(compute_feature_stats(
+                    vals,
+                    &bounds,
+                    features::has_integer_bins(name),
+                ))
+            })
+            .collect::<anyhow::Result<Vec<_>>>()?;
+
+        let mut quant_min = Vec::with_capacity(feature_names.len());
+        let mut quant_range = Vec::with_capacity(feature_names.len());
+        for (metric_idx, stats) in feature_stats.iter().enumerate() {
+            let (min, max) = match features::bounds_for(feature_names[metric_idx].as_str()) {
+                Some(Bounds::Fixed { min, max }) => (min, max),
+                _ => (stats.histogram.min, stats.histogram.max),
+            };
+            quant_min.push(min);
+            quant_range.push(if max > min { max - min } else { 0.0 });
+        }
+        let dequant_a: Vec<f32> = quant_range
+            .iter()
+            .map(|&range| {
+                if range > 0.0 {
+                    range / QUANT_SCALE
+                } else {
+                    0.0
+                }
+            })
+            .collect();
+
+        let columns: Vec<Vec<u16>> = col_major
+            .par_iter()
+            .enumerate()
+            .map(|(metric_idx, vals)| {
+                let range = quant_range[metric_idx];
+                let min = quant_min[metric_idx];
+                vals.iter()
+                    .map(|&value| {
+                        if !value.is_finite() {
+                            NAN_U16
+                        } else if range > 0.0 {
+                            let normalized = (value - min) / range;
+                            (normalized * QUANT_SCALE).round().clamp(0.0, QUANT_SCALE) as u16
+                        } else {
+                            0
+                        }
+                    })
+                    .collect()
+            })
+            .collect();
+
+        let name_to_index = feature_names
+            .iter()
+            .enumerate()
+            .map(|(idx, name)| (name.clone(), idx))
+            .collect();
+
+        Ok(Self {
+            feature_names,
+            name_to_index,
+            columns,
+            feature_stats,
+            row_to_metric_idx: Vec::new(),
+            dequant_a,
+            quant_min,
+            quant_range,
+        })
+    }
+
+    pub(super) fn set_row_mapping(&mut self, row_to_metric_idx: Vec<u32>) {
+        self.row_to_metric_idx = row_to_metric_idx;
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.feature_names.is_empty()
+    }
+
+    pub fn num_features(&self) -> usize {
+        self.feature_names.len()
+    }
+
+    pub fn quant_ref(&self) -> QuantRef<'_> {
+        QuantRef {
+            dequant_a: &self.dequant_a,
+            quant_min: &self.quant_min,
+            quant_range: &self.quant_range,
+            num_numeric: self.feature_names.len(),
+        }
+    }
+
+    #[inline]
+    pub fn metric_row_for_property(&self, row: usize) -> Option<usize> {
+        self.row_to_metric_idx
+            .get(row)
+            .copied()
+            .filter(|&idx| idx != NO_POI_METRIC_ROW)
+            .map(|idx| idx as usize)
+    }
+
+    #[inline]
+    pub fn raw_for_metric_row(&self, metric_row: usize, metric_idx: usize) -> u16 {
+        self.columns[metric_idx][metric_row]
+    }
+
+    #[inline]
+    pub fn raw_for_property_row(&self, row: usize, metric_idx: usize) -> u16 {
+        let Some(metric_row) = self.metric_row_for_property(row) else {
+            return NAN_U16;
+        };
+        self.raw_for_metric_row(metric_row, metric_idx)
+    }
+
+    #[inline]
+    pub fn decode_raw(&self, metric_idx: usize, raw: u16) -> f32 {
+        if raw == NAN_U16 {
+            f32::NAN
+        } else {
+            raw as f32 * self.dequant_a[metric_idx] + self.quant_min[metric_idx]
+        }
+    }
+
+    #[inline]
+    pub fn get_for_property_row(&self, row: usize, metric_idx: usize) -> f32 {
+        self.decode_raw(metric_idx, self.raw_for_property_row(row, metric_idx))
+    }
+}
diff --git a/server-rs/src/data/property/quant.rs b/server-rs/src/data/property/quant.rs
new file mode 100644
index 0000000..ef0c6e2
--- /dev/null
+++ b/server-rs/src/data/property/quant.rs
@@ -0,0 +1,46 @@
+//! u16 quantization: decoding stored feature values and encoding filter bounds.
+
+use crate::consts::{NAN_U16, QUANT_SCALE};
+
+/// Lightweight reference to quantization parameters for decoding u16 feature data.
+pub struct QuantRef<'a> {
+    pub dequant_a: &'a [f32],
+    pub quant_min: &'a [f32],
+    pub quant_range: &'a [f32],
+    pub num_numeric: usize,
+}
+
+impl QuantRef<'_> {
+    /// Decode a raw u16 value back to f32.
+    #[inline]
+    pub fn decode(&self, feat_idx: usize, raw: u16) -> f32 {
+        if raw == NAN_U16 {
+            return f32::NAN;
+        }
+        if feat_idx >= self.num_numeric {
+            raw as f32
+        } else {
+            raw as f32 * self.dequant_a[feat_idx] + self.quant_min[feat_idx]
+        }
+    }
+
+    /// Encode a filter minimum bound to u16 (floors to include boundary values).
+    #[inline]
+    pub fn encode_min(&self, feat_idx: usize, value: f32) -> u16 {
+        if !value.is_finite() || self.quant_range[feat_idx] == 0.0 {
+            return 0;
+        }
+        let norm = (value - self.quant_min[feat_idx]) / self.quant_range[feat_idx];
+        (norm * QUANT_SCALE).floor().clamp(0.0, QUANT_SCALE) as u16
+    }
+
+    /// Encode a filter maximum bound to u16 (ceils to include boundary values).
+    #[inline]
+    pub fn encode_max(&self, feat_idx: usize, value: f32) -> u16 {
+        if !value.is_finite() || self.quant_range[feat_idx] == 0.0 {
+            return QUANT_SCALE as u16;
+        }
+        let norm = (value - self.quant_min[feat_idx]) / self.quant_range[feat_idx];
+        (norm * QUANT_SCALE).ceil().clamp(0.0, QUANT_SCALE) as u16
+    }
+}
diff --git a/server-rs/src/data/property/stats.rs b/server-rs/src/data/property/stats.rs
new file mode 100644
index 0000000..bdce42b
--- /dev/null
+++ b/server-rs/src/data/property/stats.rs
@@ -0,0 +1,544 @@
+//! Feature statistics: outlier-bracketed histograms, percentile estimation and
+//! slider-bound computation.
+
+use anyhow::Context;
+use polars::prelude::*;
+use serde::Serialize;
+
+use crate::consts::HISTOGRAM_BINS;
+use crate::features::Bounds;
+
+/// Histogram with outlier buckets at the edges.
+/// - Bin 0: [min, p1) — low outliers
+/// - Bins 1 to n-2: [p1, p99) — main distribution, evenly divided
+/// - Bin n-1: [p99, max] — high outliers
+#[derive(Serialize, Clone)]
+pub struct Histogram {
+    pub min: f32,
+    pub max: f32,
+    /// 1st percentile (left edge of main distribution)
+    pub p1: f32,
+    /// 99th percentile (right edge of main distribution)
+    pub p99: f32,
+    pub counts: Vec<u64>,
+}
+
+impl Histogram {
+    /// Return the bin index for a given value using the outlier-bracket layout.
+    #[cfg(test)]
+    pub fn bin_for_value(&self, value: f32) -> usize {
+        let num_bins = self.counts.len();
+        if value < self.p1 {
+            0
+        } else if value >= self.p99 {
+            num_bins - 1
+        } else {
+            let middle_bins = num_bins.saturating_sub(2);
+            if middle_bins > 0 && self.p99 > self.p1 {
+                let width = (self.p99 - self.p1) / middle_bins as f32;
+                let middle_bin = ((value - self.p1) / width) as usize;
+                (1 + middle_bin).min(num_bins - 2)
+            } else {
+                num_bins / 2
+            }
+        }
+    }
+
+    /// Width of a single middle bin (bins 1..n-2).
+    #[cfg(test)]
+    pub fn middle_bin_width(&self) -> f32 {
+        let middle_bins = self.counts.len().saturating_sub(2);
+        if middle_bins > 0 && self.p99 > self.p1 {
+            (self.p99 - self.p1) / middle_bins as f32
+        } else {
+            0.0
+        }
+    }
+}
+
+pub struct FeatureStats {
+    pub slider_min: f32,
+    pub slider_max: f32,
+    pub histogram: Histogram,
+}
+
+/// Compute a percentile from a uniformly-binned histogram.
+/// `prelim_counts` are uniform bins over [min, max].
+fn percentile_from_uniform_histogram(
+    count: usize,
+    min: f32,
+    max: f32,
+    prelim_counts: &[u64],
+    percentile: f32,
+) -> f32 {
+    if count == 0 || prelim_counts.is_empty() {
+        return min;
+    }
+    let target = (count as f64 * percentile as f64 / 100.0).floor() as u64;
+    let bin_width = (max - min) / prelim_counts.len() as f32;
+    let mut cumulative = 0u64;
+    for (i, &bin_count) in prelim_counts.iter().enumerate() {
+        let prev_cumulative = cumulative;
+        cumulative += bin_count;
+        if cumulative > target {
+            // Interpolate within this bin
+            let bin_start = min + i as f32 * bin_width;
+            let fraction = if bin_count > 0 {
+                (target - prev_cumulative) as f32 / bin_count as f32
+            } else {
+                0.0
+            };
+            return bin_start + fraction * bin_width;
+        }
+    }
+    max
+}
+
+/// Build a histogram and compute slider bounds based on the feature's Bounds config.
+pub fn compute_feature_stats(vals: &[f32], bounds: &Bounds, integer_bins: bool) -> FeatureStats {
+    // Single pass: min, max, count (skipping NaN and infinity)
+    let mut min = f32::INFINITY;
+    let mut max = f32::NEG_INFINITY;
+    let mut count = 0usize;
+    for &value in vals {
+        if value.is_finite() {
+            if value < min {
+                min = value;
+            }
+            if value > max {
+                max = value;
+            }
+            count += 1;
+        }
+    }
+
+    if count == 0 {
+        let (slider_min, slider_max) = match bounds {
+            Bounds::Fixed {
+                min: fmin,
+                max: fmax,
+            } => (*fmin, *fmax),
+            Bounds::Percentile { .. } => (0.0, 0.0),
+        };
+        return FeatureStats {
+            slider_min,
+            slider_max,
+            histogram: Histogram {
+                min: 0.0,
+                max: 0.0,
+                p1: 0.0,
+                p99: 0.0,
+                counts: vec![0; HISTOGRAM_BINS],
+            },
+        };
+    }
+
+    // Build preliminary histogram with uniform bins to compute percentiles
+    // Use full HISTOGRAM_BINS for percentile precision
+    let range = if max == min { 1.0 } else { max - min };
+    let prelim_max = min + range * (1.0 + 1e-6);
+    let prelim_bin_width = (prelim_max - min) / HISTOGRAM_BINS as f32;
+
+    let mut prelim_counts = vec![0u64; HISTOGRAM_BINS];
+    for &value in vals {
+        if value.is_finite() {
+            let bin = ((value - min) / prelim_bin_width) as usize;
+            prelim_counts[bin.min(HISTOGRAM_BINS - 1)] += 1;
+        }
+    }
+
+    // Compute p1 and p99 from preliminary histogram
+    let mut p1 = percentile_from_uniform_histogram(count, min, max, &prelim_counts, 1.0);
+    let mut p99 = percentile_from_uniform_histogram(count, min, max, &prelim_counts, 99.0);
+
+    // Iterative refinement for outlier-dominated distributions.
+    // When extreme outliers (e.g. 317M sqm from web scraping) dominate the range,
+    // the uniform histogram puts all real data in one bin, making percentile
+    // estimation useless. Zoom into the estimated data region and recompute.
+    let mut refined_counts = prelim_counts;
+    let mut refined_count = count;
+    let mut refined_min = min;
+    let mut refined_max = max;
+    for _ in 0..3 {
+        let iqr = p99 - p1;
+        if iqr <= 0.0 || (refined_max - refined_min) <= 5.0 * iqr {
+            break;
+        }
+        let new_min = (p1 - iqr).max(min);
+        let new_max = p99 + iqr;
+        if new_max <= new_min {
+            break;
+        }
+        let bin_width = (new_max - new_min) / HISTOGRAM_BINS as f32;
+        let mut counts = vec![0u64; HISTOGRAM_BINS];
+        let mut cnt = 0usize;
+        for &value in vals {
+            if value.is_finite() && value >= new_min && value <= new_max {
+                let bin = ((value - new_min) / bin_width) as usize;
+                counts[bin.min(HISTOGRAM_BINS - 1)] += 1;
+                cnt += 1;
+            }
+        }
+        if cnt == 0 {
+            break;
+        }
+        p1 = percentile_from_uniform_histogram(cnt, new_min, new_max, &counts, 1.0);
+        p99 = percentile_from_uniform_histogram(cnt, new_min, new_max, &counts, 99.0);
+        refined_counts = counts;
+        refined_count = cnt;
+        refined_min = new_min;
+        refined_max = new_max;
+    }
+
+    // For integer-binned features, snap p1/p99 to integer boundaries
+    // so each middle bin is exactly 1 unit wide.
+    if integer_bins {
+        p1 = p1.floor();
+        p99 = p99.ceil();
+    }
+
+    // Determine number of histogram bins
+    let num_bins = if integer_bins && p99 > p1 {
+        // One middle bin per integer + 2 outlier bins
+        (p99 - p1) as usize + 2
+    } else {
+        // Count unique values within the p1–p99 range to cap histogram bins.
+        // Using the full-range cardinality would over-allocate bins when outliers
+        // inflate it (e.g. bedrooms: 1–137 unique values but only ~10 within p1–p99).
+        let cardinality = {
+            let mut unique_set = rustc_hash::FxHashSet::default();
+            for &val in vals {
+                if val.is_finite() && val >= p1 && val <= p99 {
+                    unique_set.insert(val.to_bits());
+                }
+            }
+            unique_set.len()
+        };
+        HISTOGRAM_BINS.min(cardinality).max(3)
+    };
+
+    // Build final histogram with outlier bins at edges:
+    // - Bin 0: [min, p1) — low outliers
+    // - Bins 1 to n-2: [p1, p99) — main distribution, evenly divided
+    // - Bin n-1: [p99, max] — high outliers
+    let mut counts = vec![0u64; num_bins];
+    let middle_bins = num_bins.saturating_sub(2);
+    let middle_width = if middle_bins > 0 && p99 > p1 {
+        (p99 - p1) / middle_bins as f32
+    } else {
+        0.0
+    };
+
+    for &value in vals {
+        if value.is_finite() {
+            let bin = if value < p1 {
+                0 // Low outlier bin
+            } else if value >= p99 {
+                num_bins - 1 // High outlier bin
+            } else if middle_width > 0.0 {
+                // Middle bins (1 to n-2)
+                let middle_bin = ((value - p1) / middle_width) as usize;
+                (1 + middle_bin).min(num_bins - 2)
+            } else {
+                num_bins / 2 // Fallback if p1 == p99
+            };
+            counts[bin] += 1;
+        }
+    }
+
+    let histogram = Histogram {
+        min: refined_min,
+        max: refined_max,
+        p1,
+        p99,
+        counts,
+    };
+
+    // Compute slider bounds (use refined histogram for accurate percentiles)
+    let (slider_min, slider_max) = match bounds {
+        Bounds::Fixed {
+            min: fmin,
+            max: fmax,
+        } => (*fmin, *fmax),
+        Bounds::Percentile { low, high } => {
+            let p_low = percentile_from_uniform_histogram(
+                refined_count,
+                refined_min,
+                refined_max,
+                &refined_counts,
+                *low as f32,
+            );
+            let p_high = percentile_from_uniform_histogram(
+                refined_count,
+                refined_min,
+                refined_max,
+                &refined_counts,
+                *high as f32,
+            );
+            (p_low, p_high)
+        }
+    };
+
+    FeatureStats {
+        slider_min,
+        slider_max,
+        histogram,
+    }
+}
+
+pub(super) fn column_to_f32_vec(column: &Column) -> anyhow::Result<Vec<f32>> {
+    let float_series = column
+        .cast(&DataType::Float32)
+        .context("Failed to cast column to Float32")?;
+    let chunked = float_series
+        .f32()
+        .context("Failed to get f32 chunked array")?;
+    Ok(chunked
+        .into_iter()
+        .map(|value| value.unwrap_or(f32::NAN))
+        .collect())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::consts::QUANT_SCALE;
+    use crate::features::Bounds;
+
+    fn make_fixed_bounds(min: f32, max: f32) -> Bounds {
+        Bounds::Fixed { min, max }
+    }
+
+    fn make_percentile_bounds(low: f64, high: f64) -> Bounds {
+        Bounds::Percentile { low, high }
+    }
+
+    #[test]
+    fn histogram_empty_data() {
+        let data: Vec<f32> = vec![];
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.slider_min, 0.0);
+        assert_eq!(stats.slider_max, 100.0);
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 0);
+    }
+
+    #[test]
+    fn histogram_single_value() {
+        let data = vec![50.0_f32];
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.min, 50.0);
+        assert_eq!(stats.histogram.max, 50.0);
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 1);
+    }
+
+    #[test]
+    fn histogram_uniform_distribution() {
+        let data: Vec<f32> = (0..100).map(|i| i as f32).collect();
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.min, 0.0);
+        assert_eq!(stats.histogram.max, 99.0);
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 100);
+    }
+
+    #[test]
+    fn histogram_with_nan_values() {
+        let data = vec![10.0_f32, f32::NAN, 20.0, f32::NAN, 30.0];
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 3);
+        assert_eq!(stats.histogram.min, 10.0);
+        assert_eq!(stats.histogram.max, 30.0);
+    }
+
+    #[test]
+    fn histogram_all_nan() {
+        let data = vec![f32::NAN, f32::NAN, f32::NAN];
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 0);
+    }
+
+    #[test]
+    fn histogram_all_same_value() {
+        let data = vec![42.0_f32; 1000];
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.min, 42.0);
+        assert_eq!(stats.histogram.max, 42.0);
+        assert_eq!(stats.histogram.p1, 42.0);
+        assert_eq!(stats.histogram.p99, 42.0);
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 1000);
+    }
+
+    #[test]
+    fn histogram_percentile_bounds() {
+        let mut data: Vec<f32> = vec![0.0]; // Low outlier
+        data.extend((1..99).map(|i| 50.0 + i as f32 * 0.01));
+        data.push(1000.0); // High outlier
+
+        let bounds = make_percentile_bounds(2.0, 98.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert!(stats.slider_min > 0.0);
+        assert!(stats.slider_max < 1000.0);
+    }
+
+    #[test]
+    fn fixed_price_bounds_keep_slider_cap() {
+        let data = vec![400_000.0_f32, 2_500_000.0, 3_750_000.0];
+        let bounds = make_fixed_bounds(0.0, 2_500_000.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.slider_min, 0.0);
+        assert_eq!(stats.slider_max, 2_500_000.0);
+    }
+
+    #[test]
+    fn histogram_bin_for_value() {
+        let hist = Histogram {
+            min: 0.0,
+            max: 100.0,
+            p1: 10.0,
+            p99: 90.0,
+            counts: vec![0; 10],
+        };
+
+        assert_eq!(hist.bin_for_value(5.0), 0); // Low outlier bin
+        assert_eq!(hist.bin_for_value(95.0), 9); // High outlier bin
+
+        let mid_value = 50.0;
+        let bin = hist.bin_for_value(mid_value);
+        assert!((1..=8).contains(&bin));
+    }
+
+    #[test]
+    fn histogram_middle_bin_width() {
+        let hist = Histogram {
+            min: 0.0,
+            max: 100.0,
+            p1: 10.0,
+            p99: 90.0,
+            counts: vec![0; 10],
+        };
+
+        let expected_width = (90.0 - 10.0) / 8.0;
+        assert!((hist.middle_bin_width() - expected_width).abs() < 0.001);
+    }
+
+    #[test]
+    fn histogram_cardinality_caps_bins() {
+        let data = vec![1.0_f32, 1.0, 2.0, 2.0, 3.0, 3.0];
+        let bounds = make_fixed_bounds(0.0, 100.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.counts.len(), 3);
+    }
+
+    #[test]
+    fn min_max_skips_nan() {
+        let values = vec![10.0_f32, f32::NAN, 20.0, f32::NAN, 5.0];
+
+        let mut min = f32::INFINITY;
+        let mut max = f32::NEG_INFINITY;
+        for &v in &values {
+            if v.is_finite() {
+                if v < min {
+                    min = v;
+                }
+                if v > max {
+                    max = v;
+                }
+            }
+        }
+
+        assert_eq!(min, 5.0);
+        assert_eq!(max, 20.0);
+    }
+
+    #[test]
+    fn count_skips_nan() {
+        let values = [1.0_f32, f32::NAN, 2.0, f32::NAN, 3.0];
+        let count = values.iter().filter(|v| v.is_finite()).count();
+        assert_eq!(count, 3);
+    }
+
+    #[test]
+    fn infinity_values_excluded() {
+        let data = vec![f32::INFINITY, f32::NEG_INFINITY, 50.0];
+        let bounds = Bounds::Fixed {
+            min: 0.0,
+            max: 100.0,
+        };
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.min, 50.0);
+        assert_eq!(stats.histogram.max, 50.0);
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 1);
+    }
+
+    #[test]
+    fn only_finite_values() {
+        let data = vec![10.0_f32, 20.0, 30.0];
+        let bounds = Bounds::Fixed {
+            min: 0.0,
+            max: 100.0,
+        };
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        assert_eq!(stats.histogram.min, 10.0);
+        assert_eq!(stats.histogram.max, 30.0);
+        assert_eq!(stats.histogram.counts.iter().sum::<u64>(), 3);
+    }
+
+    #[test]
+    fn extreme_outlier_does_not_destroy_quantization() {
+        // Simulate floor area: 10k normal values (50-200 sqm) + one 317M outlier
+        let mut data: Vec<f32> = (0..10_000).map(|i| 50.0 + (i % 150) as f32).collect();
+        data.push(317_000_000.0); // Extreme outlier from web scraping
+
+        let bounds = make_percentile_bounds(0.0, 98.0);
+        let stats = compute_feature_stats(&data, &bounds, false);
+
+        // After refinement, histogram range should be much tighter than 317M
+        assert!(
+            stats.histogram.max < 1_000_000.0,
+            "histogram.max should be refined, got {}",
+            stats.histogram.max,
+        );
+        // p1 should be near 50, not millions
+        assert!(
+            stats.histogram.p1 < 100.0,
+            "p1 should be near real data, got {}",
+            stats.histogram.p1,
+        );
+        // Slider min should reflect actual data range
+        assert!(
+            stats.slider_min < 100.0,
+            "slider_min should be near real data, got {}",
+            stats.slider_min,
+        );
+
+        // Quantization using histogram.min/max should give usable range
+        let qmin = stats.histogram.min;
+        let qrange = stats.histogram.max - stats.histogram.min;
+        assert!(qrange > 0.0 && qrange < 1_000_000.0);
+
+        // A typical floor area (100 sqm) should be distinguishable from min
+        let normalized = (100.0 - qmin) / qrange;
+        let encoded = (normalized * QUANT_SCALE).round() as u16;
+        assert!(
+            encoded > 100,
+            "100 sqm should encode to a meaningful u16 value, got {}",
+            encoded,
+        );
+    }
+}
diff --git a/server-rs/src/data/travel_time.rs b/server-rs/src/data/travel_time.rs
index c36ddff..2c617e0 100644
--- a/server-rs/src/data/travel_time.rs
+++ b/server-rs/src/data/travel_time.rs
@@ -272,6 +272,21 @@ pub fn slugify(name: &str) -> String {
     result
 }
 
+#[cfg(test)]
+impl TravelTimeStore {
+    /// Minimal empty instance for integration tests that need an `AppState`
+    /// but never touch travel time data.
+    pub(crate) fn empty_for_tests() -> Self {
+        Self {
+            base_dir: PathBuf::new(),
+            available_modes: Vec::new(),
+            destinations: FxHashMap::default(),
+            slug_to_file: FxHashMap::default(),
+            cache: Mutex::new(LruCache::new(1)),
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/server-rs/src/main.rs b/server-rs/src/main.rs
index 351874d..b92d4bb 100644
--- a/server-rs/src/main.rs
+++ b/server-rs/src/main.rs
@@ -1042,7 +1042,44 @@ async fn main() -> anyhow::Result<()> {
         listener,
         app.into_make_service_with_connect_info::<std::net::SocketAddr>(),
     )
+    .with_graceful_shutdown(shutdown_signal())
     .await
     .context("Server error")?;
+    info!("Server shut down cleanly");
     Ok(())
 }
+
+/// Resolves on SIGTERM or SIGINT so in-flight requests (exports, checkouts)
+/// can drain before the process exits. The realtime SSE proxy connections
+/// never complete, so a watchdog force-exits before Docker's default 10s
+/// stop grace period elapses and it sends SIGKILL.
+async fn shutdown_signal() {
+    let ctrl_c = async {
+        tokio::signal::ctrl_c()
+            .await
+            .expect("Failed to install SIGINT handler");
+    };
+
+    #[cfg(unix)]
+    let terminate = async {
+        tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())
+            .expect("Failed to install SIGTERM handler")
+            .recv()
+            .await;
+    };
+
+    #[cfg(not(unix))]
+    let terminate = std::future::pending::<()>();
+
+    tokio::select! {
+        () = ctrl_c => {},
+        () = terminate => {},
+    }
+    info!("Shutdown signal received; draining in-flight requests");
+
+    tokio::spawn(async {
+        tokio::time::sleep(Duration::from_secs(8)).await;
+        tracing::warn!("Graceful shutdown drain timed out after 8s; forcing exit");
+        std::process::exit(0);
+    });
+}
diff --git a/server-rs/src/routes/ai_filters.rs b/server-rs/src/routes/ai_filters.rs
deleted file mode 100644
index bddba35..0000000
--- a/server-rs/src/routes/ai_filters.rs
+++ /dev/null
@@ -1,1585 +0,0 @@
-use std::sync::Arc;
-
-use axum::extract::State;
-use axum::http::StatusCode;
-use axum::response::Json;
-use axum::Extension;
-use metrics::counter;
-use serde::{Deserialize, Serialize};
-use serde_json::{json, Map, Value};
-use tracing::{info, warn};
-
-use crate::auth::OptionalUser;
-use crate::consts::{AI_FILTERS_MAX_TOKENS, AI_FILTERS_TEMPERATURE, AI_FILTERS_WEEKLY_TOKEN_LIMIT};
-use crate::data::slugify;
-use crate::data::travel_time::TravelData;
-use crate::parsing::{parse_filters_with_poi, row_passes_filters, row_passes_poi_filters};
-use crate::pocketbase::{get_superuser_token, log_ai_query};
-use crate::routes::{FeatureInfo, FeaturesResponse};
-use crate::state::{AppState, SharedState};
-use crate::utils::gemini_chat;
-
-#[derive(Deserialize)]
-pub struct AiFiltersContext {
-    filters: Value,
-    #[serde(default)]
-    travel_time: Vec<AiTravelTimeContext>,
-}
-
-#[derive(Deserialize)]
-pub struct AiTravelTimeContext {
-    mode: String,
-    label: String,
-    min: Option<f32>,
-    max: Option<f32>,
-}
-
-#[derive(Deserialize)]
-pub struct AiFiltersRequest {
-    query: String,
-    /// Current filters for conversational refinement (e.g. "make it cheaper")
-    context: Option<AiFiltersContext>,
-}
-
-#[derive(Serialize)]
-pub struct TravelTimeFilter {
-    mode: String,
-    slug: String,
-    label: String,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    min: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    max: Option<f32>,
-}
-
-#[derive(Serialize)]
-pub struct AiFiltersResponse {
-    filters: Value,
-    #[serde(skip_serializing_if = "Vec::is_empty")]
-    travel_time_filters: Vec<TravelTimeFilter>,
-    /// What the LLM couldn't map to existing filters (empty if everything matched)
-    #[serde(skip_serializing_if = "String::is_empty")]
-    notes: String,
-    /// Number of properties matching the proposed property and travel time filters.
-    match_count: usize,
-    /// Bounding box of the matching properties so the client can move the
-    /// camera to where matches actually are. Absent when nothing matches.
-    #[serde(skip_serializing_if = "Option::is_none")]
-    match_bounds: Option<MatchBounds>,
-}
-
-#[derive(Serialize)]
-pub struct MatchBounds {
-    south: f32,
-    west: f32,
-    north: f32,
-    east: f32,
-}
-
-/// Bounding box over matched coordinates, trimmed to the 5th–95th percentile
-/// per axis (when there are enough points) so a handful of remote outliers
-/// doesn't zoom the camera out to all of England.
-fn percentile_trimmed_bounds(mut lats: Vec<f32>, mut lons: Vec<f32>) -> Option<MatchBounds> {
-    if lats.is_empty() || lats.len() != lons.len() {
-        return None;
-    }
-    lats.sort_unstable_by(f32::total_cmp);
-    lons.sort_unstable_by(f32::total_cmp);
-    let last = lats.len() - 1;
-    let (lo, hi) = if lats.len() >= 20 {
-        let trim = lats.len() / 20;
-        (trim, last - trim)
-    } else {
-        (0, last)
-    };
-    Some(MatchBounds {
-        south: lats[lo],
-        north: lats[hi],
-        west: lons[lo],
-        east: lons[hi],
-    })
-}
-
-/// Strip markdown code fences (```json ... ``` or ``` ... ```) from LLM output.
-/// Models occasionally wrap JSON in markdown fencing even when told not to.
-fn strip_markdown_fences(text: &str) -> &str {
-    let trimmed = text.trim();
-
-    // Try ```json\n...\n``` or ```\n...\n```
-    if let Some(rest) = trimmed.strip_prefix("```") {
-        // Skip optional language tag (e.g. "json")
-        let rest = if let Some(newline_pos) = rest.find('\n') {
-            &rest[newline_pos + 1..]
-        } else {
-            return trimmed;
-        };
-        if let Some(content) = rest.strip_suffix("```") {
-            return content.trim();
-        }
-    }
-
-    trimmed
-}
-
-fn school_feature_name_from_key(name: &str) -> Option<&'static str> {
-    let rest = name.strip_prefix("Schools:")?;
-    let mut parts = rest.split(':');
-    let phase = parts.next()?;
-    let rating = parts.next()?;
-
-    match (phase, rating) {
-        ("primary", "good") => Some("Good+ primary school catchments"),
-        ("secondary", "good") => Some("Good+ secondary school catchments"),
-        ("primary", "outstanding") => Some("Outstanding primary school catchments"),
-        ("secondary", "outstanding") => Some("Outstanding secondary school catchments"),
-        _ => None,
-    }
-}
-
-fn decode_synthetic_feature_key(name: &str, prefix: &str) -> Option<String> {
-    let rest = name.strip_prefix(prefix)?;
-    let (encoded, _id) = rest.rsplit_once(':')?;
-    urlencoding::decode(encoded)
-        .ok()
-        .map(|decoded| decoded.into_owned())
-}
-
-/// Convert frontend synthetic filter keys back to backend feature names.
-///
-/// The React filter UI stores configurable cards under keys such as
-/// `Political vote share:%25%20Labour:0`. The LLM and backend validators need
-/// the real feature name (`% Labour`) instead.
-fn backend_filter_name(name: &str) -> Option<String> {
-    if let Some(feature_name) = school_feature_name_from_key(name) {
-        return Some(feature_name.to_string());
-    }
-
-    for prefix in [
-        "Specific crimes:",
-        "Political vote share:",
-        "Ethnicities:",
-        "Amenity distance:",
-        "Transport distance:",
-        "Amenities within 2km:",
-        "Amenities within 5km:",
-    ] {
-        if let Some(feature_name) = decode_synthetic_feature_key(name, prefix) {
-            return Some(feature_name);
-        }
-    }
-
-    None
-}
-
-fn canonical_filter_name(name: &str) -> String {
-    backend_filter_name(name).unwrap_or_else(|| name.to_string())
-}
-
-fn normalize_context_filters(filters: &Value) -> Value {
-    let Some(obj) = filters.as_object() else {
-        return filters.clone();
-    };
-
-    let mut normalized = Map::with_capacity(obj.len());
-    for (name, value) in obj {
-        normalized.insert(canonical_filter_name(name), value.clone());
-    }
-    Value::Object(normalized)
-}
-
-/// Build the Gemini tool declaration for destination search.
-fn build_tool_declarations(state: &AppState) -> Value {
-    let modes: Vec<&str> = state
-        .travel_time_store
-        .available_modes
-        .iter()
-        .map(|mode| mode.as_str())
-        .collect();
-
-    json!([{
-        "functionDeclarations": [{
-            "name": "search_destinations",
-            "description": "Search for available travel time destinations (cities, stations, towns) that have precomputed travel time data. Call this when the user mentions wanting to be near, close to, or within a certain travel time of a specific place.",
-            "parameters": {
-                "type": "object",
-                "properties": {
-                    "query": {
-                        "type": "string",
-                        "description": "Place name to search for (e.g. 'Manchester', 'Kings Cross', 'Heathrow')"
-                    },
-                    "mode": {
-                        "type": "string",
-                        "enum": modes,
-                        "description": "Transport mode to search destinations for"
-                    }
-                },
-                "required": ["query", "mode"]
-            }
-        }]
-    }])
-}
-
-/// Execute a destination search against PlaceData + TravelTimeStore.
-/// Returns matching destinations as a JSON value with `results` and optional `message`.
-///
-/// Uses word-based matching: all words in the query must appear somewhere in the
-/// place name (order-independent). Also matches against slugs for short queries.
-fn execute_destination_search(state: &AppState, query: &str, mode: &str) -> Value {
-    let query_lower = query.to_lowercase();
-    let query_words: Vec<&str> = query_lower.split_whitespace().collect();
-    let query_slug = slugify(query);
-    let tt_store = &state.travel_time_store;
-    let pd = &state.place_data;
-
-    let slug_set = match tt_store.destinations.get(mode) {
-        Some(slugs) => slugs,
-        None => {
-            return json!({ "results": [], "message": format!("No travel data available for mode '{}'", mode) })
-        }
-    };
-
-    // Find places matching the query that have travel time data.
-    // A place matches if ALL query words appear in its name, OR its slug matches the query slug.
-    let mut matches: Vec<(usize, String, u8, u32)> = pd
-        .name_lower
-        .iter()
-        .enumerate()
-        .filter_map(|(idx, name_lower)| {
-            if !pd.travel_destination[idx] {
-                return None;
-            }
-            let words_match = query_words.iter().all(|word| name_lower.contains(word));
-            let slug = slugify(&pd.name[idx]);
-            let slug_match = slug.contains(&query_slug) || query_slug.contains(&slug);
-            if !words_match && !slug_match {
-                return None;
-            }
-            if slug_set.contains(&slug) {
-                Some((idx, slug, pd.type_rank[idx], pd.population[idx]))
-            } else {
-                None
-            }
-        })
-        .collect();
-
-    // Sort: type rank asc, population desc
-    matches.sort_unstable_by(|a, b| a.2.cmp(&b.2).then(b.3.cmp(&a.3)));
-    matches.truncate(10);
-
-    if matches.is_empty() {
-        // Check if the query matched a city that lacks its own travel data.
-        // If so, return nearby stations within that city as suggestions.
-        let matched_city_name: Option<&str> =
-            pd.name_lower
-                .iter()
-                .enumerate()
-                .find_map(|(idx, name_lower)| {
-                    if !pd.travel_destination[idx] {
-                        return None;
-                    }
-                    let words_match = query_words.iter().all(|word| name_lower.contains(word));
-                    let slug = slugify(&pd.name[idx]);
-                    let slug_match = slug.contains(&query_slug) || query_slug.contains(&slug);
-                    if (words_match || slug_match) && pd.type_rank[idx] == 0 {
-                        Some(pd.name[idx].as_str())
-                    } else {
-                        None
-                    }
-                });
-
-        if let Some(city_name) = matched_city_name {
-            let city_lower = city_name.to_lowercase();
-            let mut city_matches: Vec<(usize, String, u8, u32)> = pd
-                .city
-                .iter()
-                .enumerate()
-                .filter_map(|(idx, city_opt)| {
-                    if !pd.travel_destination[idx] {
-                        return None;
-                    }
-                    let city = city_opt.as_deref()?;
-                    if city.to_lowercase() != city_lower {
-                        return None;
-                    }
-                    let slug = slugify(&pd.name[idx]);
-                    if slug_set.contains(&slug) {
-                        Some((idx, slug, pd.type_rank[idx], pd.population[idx]))
-                    } else {
-                        None
-                    }
-                })
-                .collect();
-
-            city_matches.sort_unstable_by(|a, b| a.2.cmp(&b.2).then(b.3.cmp(&a.3)));
-            city_matches.truncate(10);
-
-            if !city_matches.is_empty() {
-                let results: Vec<Value> = city_matches
-                    .into_iter()
-                    .map(|(idx, slug, ..)| {
-                        json!({
-                            "name": pd.name[idx],
-                            "slug": slug,
-                            "place_type": pd.place_type.get(idx).to_string(),
-                        })
-                    })
-                    .collect();
-
-                info!(
-                    query = query,
-                    city = city_name,
-                    results = results.len(),
-                    "Destination search fell back to city stations"
-                );
-
-                return json!({
-                    "results": results,
-                    "message": format!(
-                        "No travel data for '{}' directly. Pick one of these nearby stations:",
-                        city_name
-                    )
-                });
-            }
-        }
-
-        info!(
-            query = query,
-            mode = mode,
-            "Destination search returned no results"
-        );
-        return json!({
-            "results": [],
-            "message": format!("No travel time data available for '{}' by {}. This destination cannot be used as a travel time filter.", query, mode)
-        });
-    }
-
-    let results: Vec<Value> = matches
-        .into_iter()
-        .map(|(idx, slug, ..)| {
-            json!({
-                "name": pd.name[idx],
-                "slug": slug,
-                "place_type": pd.place_type.get(idx).to_string(),
-            })
-        })
-        .collect();
-
-    json!({ "results": results })
-}
-
-/// Build the complete system prompt for AI filters.
-///
-/// Contains: role instructions, feature catalogue, travel time info,
-/// few-shot examples, output rules.
-/// Precomputed at startup and cached in AppState.
-pub fn build_system_prompt(
-    features: &FeaturesResponse,
-    mode_destinations: &[(String, usize)],
-) -> String {
-    let mut parts = Vec::new();
-
-    parts.push(
-        "You are a UK property search assistant. \
-         The user describes their ideal property or area in natural language. \
-         Translate their description into filter settings using ONLY the features listed below.\n\
-         \n\
-         Rules:\n\
-         - ONLY set filters the user explicitly mentioned or clearly implied.\n\
-         - Leave out any filter the user did not mention. Empty arrays are fine.\n\
-         - Each numeric filter sets ONE bound only: \"min\" (at least this value) \
-           or \"max\" (at most this value). Never set two filters on the same feature.\n\
-         - Use EXACT feature names from the list — spelling, capitalisation, and punctuation must match.\n\
-         - \"cheap\" / \"affordable\" = lower price range. \"expensive\" = higher price range.\n\
-         - \"low crime\" / \"safe\" = low values on the Serious crime (avg/yr) and Minor crime (avg/yr) \
-           features (area-normalised incident density near the postcode). Prefer these aggregates for broad \
-           area safety; use specific crime features only when the user names a crime type.\n\
-         - \"quiet\" = low Noise (dB). \"green\" / \"near parks\" = high Number of amenities (Park) within 2km \
-           or low Distance to nearest park (km), depending on wording.\n\
-         - \"good schools\" = Good+ school features. \"outstanding schools\" = Outstanding school features.\n\
-         - Amenities and transport stops are normal filters in the feature catalogue. \
-           For \"near a bus stop\", \"near a station\", \"near shops\", etc., use the exact \
-           Distance to nearest amenity (...) or Number of amenities (...) feature when available.\n\
-         - Politics/elections are normal filters in the Neighbours group. Use exact vote share \
-           features such as % Labour, % Conservative, % Liberal Democrat, % Reform UK, % Green, \
-           % Other parties, or Voter turnout (%) when the user asks for political character.\n\
-         - When the user says a number like \"under 400k\", interpret it as 400000.\n\
-         - When the user says \"3 bed\" or \"3 bedroom\", use Number of bedrooms & living rooms \
-           (note: this counts bedrooms + living rooms combined, so 3 bed ~ min 4).\n\
-         - If the user mentions something that has no matching filter, put it in \"notes\" \
-           as a short phrase (e.g. \"No filter for: garden, sea view\"). \
-           If everything was matched, set \"notes\" to an empty string.\n\
-         \n\
-         CONVERSATIONAL REFINEMENT:\n\
-         The user's message may include their currently active filters as context. \
-         When context is provided:\n\
-         - \"make it cheaper\" / \"lower the price\" = adjust the existing price filter down\n\
-         - \"also add ...\" / \"and good schools\" = keep existing filters and add new ones\n\
-         - \"remove the ...\" / \"drop the ...\" = return filters WITHOUT the mentioned one\n\
-         - If the request is a completely new search (not a refinement), ignore the context \
-           and build filters from scratch.\n\
-         - Always output the COMPLETE set of filters (existing + modified), not just the changes."
-            .to_string(),
-    );
-
-    // Travel time section with available modes
-    let modes_list = mode_destinations
-        .iter()
-        .map(|(mode, count)| format!("- {} ({} destinations available)", mode, count))
-        .collect::<Vec<_>>()
-        .join("\n");
-
-    parts.push(format!(
-        "\n--- TRAVEL TIME FILTERS ---\n\
-         You can add travel time filters when the user mentions commute times, \
-         proximity to places, or wanting to be near/within X minutes of somewhere.\n\
-         \n\
-         Available travel-time modes (only use modes that have destinations):\n\
-         {}\n\
-         - \"car\" / \"drive\" / \"driving\" = car mode\n\
-         - \"cycle\" / \"bike\" / \"cycling\" = bicycle mode\n\
-         - \"walk\" / \"walking\" / \"on foot\" = walking mode\n\
-         - \"train\" / \"tube\" / \"bus\" / \"public transport\" / \"commute\" = transit mode\n\
-         - \"without buses\" / \"no bus\" / \"rail only\" = transit-no-bus mode\n\
-         - \"no change\" / \"no transfer\" / \"direct\" / \"single bus/train\" = transit-no-change mode\n\
-         - \"no change and no bus\" / \"direct rail/tube\" = transit-no-change-no-bus mode\n\
-         - If a mode appears in the available mode list but is not named above, you may still \
-           use the exact mode string from the list.\n\
-         \n\
-         When the user mentions a specific place, you MUST call the search_destinations \
-         tool to find the exact slug. Use the name and slug from the search results.\n\
-         If search_destinations returns an empty array, the destination is not available — \
-         mention it in \"notes\" (e.g. \"No travel data for: Gatwick Airport\") and do NOT \
-         include a travel_time_filter for it.\n\
-         \n\
-         Travel time values are in MINUTES (0-90 range; data is capped at 90 min).\n\
-         - \"within 30 minutes\" = set \"max\": 30\n\
-         - \"at least 10 minutes\" = set \"min\": 10\n\
-         - \"30-45 minute commute\" = set \"min\": 30 and \"max\": 45 on the same travel_time_filter\n\
-         - If only a max is given, omit min (and vice versa). Do not use bound/value for travel time.\n\
-         \n\
-         INFERRING TRANSPORT MODE (when the user does not specify one explicitly):\n\
-         - \"commute\" to a major city centre or station = transit\n\
-         - \"near\" / \"close to\" a city centre or station = transit\n\
-         - \"near\" / \"close to\" a smaller town, village, or rural area = car\n\
-         - \"drive\" / \"driving distance\" / \"driving time\" = always car\n\
-         - If multiple modes are plausible, prefer transit for urban destinations \
-           (London, Manchester, Birmingham, Leeds, etc.) and car for everything else.",
-        modes_list,
-    ));
-
-    // Feature guidance
-    parts.push(
-        "\n--- DATA SOURCE ---\n\
-         The data is historical property sales from the Land Registry.\n\
-         \n\
-         Use these features for price queries:\n\
-         - For purchase price: use \"Estimated current price\" or \"Last known price\"\n\
-         - For price per sqm: use \"Est. price per sqm\"\n\
-         - For rent estimates: use \"Estimated monthly rent\""
-            .to_string(),
-    );
-
-    // Feature catalogue
-    parts.push("\n--- AVAILABLE FEATURES ---\n".to_string());
-    for group in &features.groups {
-        parts.push(format!("## {}", group.name));
-        for feature in &group.features {
-            match feature {
-                FeatureInfo::Numeric {
-                    name,
-                    min,
-                    max,
-                    description,
-                    prefix,
-                    suffix,
-                    ..
-                } => {
-                    parts.push(format!(
-                        "- \"{}\" (numeric, {}{:.0}{} to {}{:.0}{}): {}",
-                        name, prefix, min, suffix, prefix, max, suffix, description
-                    ));
-                }
-                FeatureInfo::Enum {
-                    name,
-                    values,
-                    description,
-                    ..
-                } => {
-                    parts.push(format!(
-                        "- \"{}\" (enum, values: [{}]): {}",
-                        name,
-                        values
-                            .iter()
-                            .map(|val| format!("\"{}\"", val))
-                            .collect::<Vec<_>>()
-                            .join(", "),
-                        description
-                    ));
-                }
-            }
-        }
-    }
-
-    // Few-shot examples
-    parts.push("\n--- EXAMPLES ---\n".to_string());
-
-    parts.push(
-        "User: \"cheap freehold house under 400k\"\n\
-         Output: {\"numeric_filters\": [{\"name\": \"Last known price\", \"bound\": \"max\", \"value\": 400000}], \
-         \"enum_filters\": [{\"name\": \"Leasehold/Freehold\", \"values\": [\"Freehold\"]}, \
-         {\"name\": \"Property type\", \"values\": [\"Detached\", \"Semi-Detached\", \"Terraced\"]}], \
-         \"travel_time_filters\": [], \
-         \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"safe quiet area with good schools and parks\"\n\
-         Output: {\"numeric_filters\": [\
-         {\"name\": \"Serious crime (avg/yr)\", \"bound\": \"max\", \"value\": 5}, \
-         {\"name\": \"Minor crime (avg/yr)\", \"bound\": \"max\", \"value\": 20}, \
-         {\"name\": \"Noise (dB)\", \"bound\": \"max\", \"value\": 55}, \
-         {\"name\": \"Good+ primary school catchments\", \"bound\": \"min\", \"value\": 2}, \
-         {\"name\": \"Good+ secondary school catchments\", \"bound\": \"min\", \"value\": 1}, \
-         {\"name\": \"Number of amenities (Park) within 2km\", \"bound\": \"min\", \"value\": 3}], \
-         \"enum_filters\": [], \"travel_time_filters\": [], \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"quiet area with outstanding schools\"\n\
-         Output: {\"numeric_filters\": [\
-         {\"name\": \"Noise (dB)\", \"bound\": \"max\", \"value\": 55}, \
-         {\"name\": \"Outstanding primary school catchments\", \"bound\": \"min\", \"value\": 1}, \
-         {\"name\": \"Outstanding secondary school catchments\", \"bound\": \"min\", \"value\": 1}], \
-         \"enum_filters\": [], \"travel_time_filters\": [], \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"3 bed flat under 300k with fast broadband near the beach\"\n\
-         Output: {\"numeric_filters\": [\
-         {\"name\": \"Last known price\", \"bound\": \"max\", \"value\": 300000}, \
-         {\"name\": \"Number of bedrooms & living rooms\", \"bound\": \"min\", \"value\": 4}], \
-         \"enum_filters\": [{\"name\": \"Property type\", \"values\": [\"Flats/Maisonettes\"]}, \
-         {\"name\": \"Max available download speed (Mbps)\", \"values\": [\"100\", \"300\", \"1000\"]}], \
-         \"travel_time_filters\": [], \
-         \"notes\": \"No filter for: beach proximity\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"within 30 minutes commute of Kings Cross, under 500k\"\n\
-         (After calling search_destinations for \"Kings Cross\" with mode \"transit\" \
-         and getting [{\"name\": \"Kings Cross\", \"slug\": \"kings-cross\", \"place_type\": \"station\"}])\n\
-         Output: {\"numeric_filters\": [\
-         {\"name\": \"Last known price\", \"bound\": \"max\", \"value\": 500000}], \
-         \"enum_filters\": [], \
-         \"travel_time_filters\": [{\"mode\": \"transit\", \"slug\": \"kings-cross\", \
-         \"label\": \"Kings Cross\", \"max\": 30}], \
-         \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"family home with garden, 45 min drive from Manchester, good schools\"\n\
-         (After calling search_destinations for \"Manchester\" with mode \"car\" \
-         and getting [{\"name\": \"Manchester\", \"slug\": \"manchester\", \"place_type\": \"city\"}])\n\
-         Output: {\"numeric_filters\": [\
-         {\"name\": \"Total floor area (sqm)\", \"bound\": \"min\", \"value\": 100}, \
-         {\"name\": \"Number of bedrooms & living rooms\", \"bound\": \"min\", \"value\": 5}, \
-         {\"name\": \"Good+ primary school catchments\", \"bound\": \"min\", \"value\": 2}, \
-         {\"name\": \"Good+ secondary school catchments\", \"bound\": \"min\", \"value\": 1}], \
-         \"enum_filters\": [{\"name\": \"Property type\", \
-         \"values\": [\"Detached\", \"Semi-Detached\"]}], \
-         \"travel_time_filters\": [{\"mode\": \"car\", \"slug\": \"manchester\", \
-         \"label\": \"Manchester\", \"max\": 45}], \
-         \"notes\": \"No filter for: garden\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"Labour-voting area with low burglary and a station nearby\"\n\
-         Output: {\"numeric_filters\": [\
-         {\"name\": \"% Labour\", \"bound\": \"min\", \"value\": 40}, \
-         {\"name\": \"Burglary (avg/yr)\", \"bound\": \"max\", \"value\": 10}, \
-         {\"name\": \"Distance to nearest amenity (Rail station) (km)\", \"bound\": \"max\", \"value\": 1}], \
-         \"enum_filters\": [], \"travel_time_filters\": [], \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    // Examples showing rent and price features
-    parts.push(
-        "\nUser: \"2 bed flat with rent under £1500/month\"\n\
-         Output: {\
-         \"numeric_filters\": [{\"name\": \"Estimated monthly rent\", \"bound\": \"max\", \"value\": 1500}], \
-         \"enum_filters\": [{\"name\": \"Property type\", \"values\": [\"Flats/Maisonettes\"]}], \
-         \"travel_time_filters\": [], \
-         \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    parts.push(
-        "\nUser: \"3 bed house under 500k with good schools\"\n\
-         Output: {\
-         \"numeric_filters\": [{\"name\": \"Estimated current price\", \"bound\": \"max\", \"value\": 500000}, \
-         {\"name\": \"Good+ primary school catchments\", \"bound\": \"min\", \"value\": 2}], \
-         \"enum_filters\": [{\"name\": \"Property type\", \
-         \"values\": [\"Detached\", \"Semi-Detached\", \"Terraced\"]}], \
-         \"travel_time_filters\": [], \
-         \"notes\": \"\"}"
-            .to_string(),
-    );
-
-    // Output format reminder
-    parts.push(
-        "\n--- OUTPUT FORMAT ---\n\
-         {\"numeric_filters\": [...], \"enum_filters\": [...], \
-         \"travel_time_filters\": [{\"mode\": \"...\", \"slug\": \"...\", \"label\": \"...\", \
-         \"min\": N, \"max\": N}, ...], \"notes\": \"...\"}\n\
-         - travel_time_filters: min and max are both optional, but include at least one. \
-         Use ONLY slugs returned by search_destinations. If a place isn't found, mention it in notes.\n\
-         Respond with ONLY the JSON object. No explanation."
-            .to_string(),
-    );
-
-    parts.join("\n")
-}
-
-/// Monotonically increasing week number derived from Unix epoch.
-/// Resets every 7 days (604800 seconds). Used for weekly rate limiting.
-fn current_week_number() -> u64 {
-    let secs = std::time::SystemTime::now()
-        .duration_since(std::time::UNIX_EPOCH)
-        .expect("system time before epoch")
-        .as_secs();
-    secs / 604_800
-}
-
-/// Fetch the user's current AI token usage from PocketBase.
-/// Returns `(tokens_used, week_number)`.
-async fn fetch_ai_usage(
-    state: &AppState,
-    user_id: &str,
-) -> Result<(u64, u64), (StatusCode, String)> {
-    let token = get_superuser_token(state).await.map_err(|err| {
-        warn!("Failed to auth superuser for AI usage check: {err}");
-        (StatusCode::BAD_GATEWAY, "Internal error".into())
-    })?;
-
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/users/records/{user_id}");
-    let resp = state
-        .http_client
-        .get(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .send()
-        .await
-        .map_err(|err| {
-            warn!("Failed to fetch user record for AI usage: {err}");
-            (StatusCode::BAD_GATEWAY, "Internal error".into())
-        })?;
-
-    if !resp.status().is_success() {
-        let status = resp.status();
-        warn!("PocketBase user fetch failed ({status})");
-        return Err((StatusCode::BAD_GATEWAY, "Internal error".into()));
-    }
-
-    let body: Value = resp.json().await.map_err(|err| {
-        warn!("Failed to parse user record: {err}");
-        (StatusCode::BAD_GATEWAY, "Internal error".into())
-    })?;
-
-    let tokens_used = body
-        .get("ai_tokens_used")
-        .and_then(|val| val.as_u64())
-        .unwrap_or(0);
-    let week = body
-        .get("ai_tokens_week")
-        .and_then(|val| val.as_u64())
-        .unwrap_or(0);
-
-    Ok((tokens_used, week))
-}
-
-/// Update the user's AI token usage in PocketBase.
-/// Best-effort — logs warnings on failure but does not propagate errors.
-async fn update_ai_usage(state: &AppState, user_id: &str, tokens_used: u64, week: u64) {
-    let token = match get_superuser_token(state).await {
-        Ok(tk) => tk,
-        Err(err) => {
-            warn!("Failed to auth superuser for AI usage update: {err}");
-            return;
-        }
-    };
-
-    let pb_url = state.pocketbase_url.trim_end_matches('/');
-    let url = format!("{pb_url}/api/collections/users/records/{user_id}");
-    let res = state
-        .http_client
-        .patch(&url)
-        .header("Authorization", format!("Bearer {token}"))
-        .json(&json!({
-            "ai_tokens_used": tokens_used,
-            "ai_tokens_week": week,
-        }))
-        .send()
-        .await;
-
-    match res {
-        Ok(resp) if resp.status().is_success() => {}
-        Ok(resp) => {
-            let status = resp.status();
-            warn!("Failed to update AI usage ({status})");
-        }
-        Err(err) => warn!("Failed to update AI usage: {err}"),
-    }
-}
-
-async fn record_ai_request_usage(
-    state: &AppState,
-    user_id: &str,
-    existing_tokens_used: u64,
-    week: u64,
-    request_tokens_used: u64,
-    status: &'static str,
-) {
-    if request_tokens_used > 0 {
-        let new_total = existing_tokens_used.saturating_add(request_tokens_used);
-        update_ai_usage(state, user_id, new_total, week).await;
-        counter!("ai_tokens_total").increment(request_tokens_used);
-    }
-    counter!("ai_requests_total", "status" => status).increment(1);
-}
-
-/// Convert validated filter JSON back to the `;;`-separated filter string format
-/// that `parse_filters` expects.
-///
-/// Numeric: `{"name": [min, max]}` → `name:min:max`
-/// Enum: `{"name": ["val1", "val2"]}` → `name:val1|val2`
-fn filters_to_filter_string(filters: &Value) -> String {
-    let obj = match filters.as_object() {
-        Some(obj) => obj,
-        None => return String::new(),
-    };
-
-    let mut parts = Vec::new();
-    for (name, value) in obj {
-        if let Some(arr) = value.as_array() {
-            if arr.len() == 2 && arr[0].is_number() && arr[1].is_number() {
-                let min = arr[0].as_f64().unwrap_or(0.0);
-                let max = arr[1].as_f64().unwrap_or(0.0);
-                parts.push(format!("{name}:{min}:{max}"));
-            } else if !arr.is_empty() && arr[0].is_string() {
-                let values: Vec<&str> = arr.iter().filter_map(|v| v.as_str()).collect();
-                if !values.is_empty() {
-                    parts.push(format!("{name}:{}", values.join("|")));
-                }
-            }
-        }
-    }
-
-    parts.join(";;")
-}
-
-/// Count how many rows in the property dataset pass the given property filters
-/// AND travel time filters. Travel time data is loaded from the TravelTimeStore
-/// and checked per-postcode (same logic as hexagons.rs).
-fn count_matching_rows(
-    state: &AppState,
-    filters: &Value,
-    travel_time_filters: &[TravelTimeFilter],
-) -> (usize, Option<MatchBounds>) {
-    let filter_str = filters_to_filter_string(filters);
-
-    let quant = state.data.quant_ref();
-    let poi_quant = state.data.poi_metrics.quant_ref();
-    let (parsed_filters, parsed_enum_filters, parsed_poi_filters) = if filter_str.is_empty() {
-        (Vec::new(), Vec::new(), Vec::new())
-    } else {
-        match parse_filters_with_poi(
-            Some(&filter_str),
-            &state.feature_name_to_index,
-            &state.data.enum_values,
-            &quant,
-            &state.data.poi_metrics.name_to_index,
-            &poi_quant,
-        ) {
-            Ok(f) => f,
-            Err(err) => {
-                warn!("Failed to parse filters for match count: {err}");
-                return (0, None);
-            }
-        }
-    };
-
-    // Load travel time data for each filter entry
-    let travel_data: Vec<(TravelData, Option<f32>, Option<f32>)> = travel_time_filters
-        .iter()
-        .filter_map(|ttf| {
-            let data = state.travel_time_store.get(&ttf.mode, &ttf.slug).ok()?;
-            Some((data, ttf.min, ttf.max))
-        })
-        .collect();
-    let has_travel = !travel_data.is_empty();
-
-    let feature_data = &state.data.feature_data;
-    let num_features = state.data.num_features;
-    let num_rows = state.data.lat.len();
-    let (pc_interner, pc_keys) = state.data.postcode_parts();
-    let has_poi_filters = !parsed_poi_filters.is_empty();
-
-    let mut count = 0usize;
-    let mut matched_lats: Vec<f32> = Vec::new();
-    let mut matched_lons: Vec<f32> = Vec::new();
-    for (row, pc_key) in pc_keys.iter().enumerate().take(num_rows) {
-        if !row_passes_filters(
-            row,
-            &parsed_filters,
-            &parsed_enum_filters,
-            feature_data,
-            num_features,
-        ) {
-            continue;
-        }
-        if has_poi_filters
-            && !row_passes_poi_filters(row, &parsed_poi_filters, &state.data.poi_metrics)
-        {
-            continue;
-        }
-
-        if has_travel {
-            let postcode = pc_interner.resolve(pc_key);
-            let mut passes_travel = true;
-            for (data, fmin, fmax) in &travel_data {
-                let pass = if let Some(mins) = data.get(postcode).map(|r| r.minutes as f32) {
-                    fmin.is_none_or(|min| mins >= min) && fmax.is_none_or(|max| mins <= max)
-                } else {
-                    false // no travel data → postcode not reachable
-                };
-                if !pass {
-                    passes_travel = false;
-                    break;
-                }
-            }
-            if !passes_travel {
-                continue;
-            }
-        }
-
-        count += 1;
-        matched_lats.push(state.data.lat[row]);
-        matched_lons.push(state.data.lon[row]);
-    }
-
-    (count, percentile_trimmed_bounds(matched_lats, matched_lons))
-}
-
-/// Budget limits for the Gemini conversation loop. Separate counters prevent
-/// tool calls (destination searches) from starving JSON retries or zero-match
-/// refinements.
-const MAX_TOOL_CALLS: usize = 4;
-const MAX_RETRIES: usize = 3;
-const MAX_REFINEMENTS: u32 = 3;
-const MAX_TOTAL_ROUNDS: usize = 10;
-
-const MAX_AI_QUERY_CHARS: usize = 5000;
-
-pub async fn post_ai_filters(
-    State(shared): State<Arc<SharedState>>,
-    Extension(user): Extension<OptionalUser>,
-    Json(req): Json<AiFiltersRequest>,
-) -> Result<Json<AiFiltersResponse>, (StatusCode, String)> {
-    let state = shared.load_state();
-    // Auth check
-    let user = user
-        .0
-        .ok_or((StatusCode::UNAUTHORIZED, "Login required".into()))?;
-
-    if req.query.chars().count() > MAX_AI_QUERY_CHARS {
-        counter!("ai_requests_total", "status" => "query_too_long").increment(1);
-        return Err((
-            StatusCode::PAYLOAD_TOO_LARGE,
-            format!("Query too long (max {MAX_AI_QUERY_CHARS} chars)"),
-        ));
-    }
-
-    // Check weekly token usage
-    let current_week = current_week_number();
-    let (stored_tokens, stored_week) = fetch_ai_usage(&state, &user.id).await?;
-    let tokens_used = if stored_week == current_week {
-        stored_tokens
-    } else {
-        0
-    };
-
-    if tokens_used >= AI_FILTERS_WEEKLY_TOKEN_LIMIT {
-        counter!("ai_requests_total", "status" => "rate_limited").increment(1);
-        return Err((
-            StatusCode::TOO_MANY_REQUESTS,
-            "Weekly AI usage limit reached. Resets next week.".into(),
-        ));
-    }
-
-    info!(query = %req.query, user_id = %user.id, "POST /api/ai-filters");
-
-    let tools = build_tool_declarations(&state);
-
-    // Build user message with optional context for conversational refinement
-    let user_text = if let Some(ref ctx) = req.context {
-        let mut msg = String::new();
-        msg.push_str("Currently active filters:\n");
-        let normalized_filters = normalize_context_filters(&ctx.filters);
-        msg.push_str(&serde_json::to_string(&normalized_filters).unwrap_or_default());
-        if !ctx.travel_time.is_empty() {
-            msg.push_str("\nCurrently active travel time filters:\n");
-            for tt in &ctx.travel_time {
-                let bounds = match (tt.min, tt.max) {
-                    (Some(min), Some(max)) => format!("{}-{} min", min, max),
-                    (Some(min), None) => format!("min {} min", min),
-                    (None, Some(max)) => format!("max {} min", max),
-                    (None, None) => "no range".to_string(),
-                };
-                msg.push_str(&format!("- {} to {} ({})\n", tt.mode, tt.label, bounds));
-            }
-        }
-        msg.push_str(&format!("\nUser request: {}", req.query));
-        msg
-    } else {
-        req.query.clone()
-    };
-
-    let mut contents = vec![json!({
-        "role": "user",
-        "parts": [{ "text": user_text }]
-    })];
-
-    let mut total_tokens_accumulated: u64 = 0;
-    let mut tool_call_count = 0usize;
-    let mut retry_count = 0usize;
-    let mut refinement_attempts = 0u32;
-
-    // Function calling loop: model may call search_destinations, we execute and feed back
-    for round in 0..MAX_TOTAL_ROUNDS {
-        let body = json!({
-            "systemInstruction": {
-                "parts": [{ "text": state.ai_filters_system_prompt }]
-            },
-            "contents": contents,
-            "tools": tools,
-            "generationConfig": {
-                "temperature": AI_FILTERS_TEMPERATURE,
-                "maxOutputTokens": AI_FILTERS_MAX_TOKENS,
-                "thinkingConfig": { "thinkingLevel": "LOW" },
-            }
-        });
-
-        let json_resp = match gemini_chat(
-            &state.http_client,
-            &state.gemini_api_key,
-            &state.gemini_model,
-            &body,
-        )
-        .await
-        {
-            Ok(resp) => resp,
-            Err(err) => {
-                record_ai_request_usage(
-                    &state,
-                    &user.id,
-                    tokens_used,
-                    current_week,
-                    total_tokens_accumulated,
-                    "llm_error",
-                )
-                .await;
-                return Err(err);
-            }
-        };
-
-        // Accumulate token usage
-        total_tokens_accumulated += json_resp
-            .get("usageMetadata")
-            .and_then(|md| md.get("totalTokenCount"))
-            .and_then(|tc| tc.as_u64())
-            .unwrap_or(0);
-
-        let candidate = match json_resp
-            .get("candidates")
-            .and_then(|cs| cs.get(0))
-            .and_then(|c| c.get("content"))
-        {
-            Some(candidate) => candidate,
-            None => {
-                warn!("Malformed Gemini response: missing candidates[0].content");
-                record_ai_request_usage(
-                    &state,
-                    &user.id,
-                    tokens_used,
-                    current_week,
-                    total_tokens_accumulated,
-                    "malformed_response",
-                )
-                .await;
-                return Err((StatusCode::BAD_GATEWAY, "Malformed Gemini response".into()));
-            }
-        };
-
-        let parts = match candidate.get("parts").and_then(|p| p.as_array()) {
-            Some(parts) => parts,
-            None => {
-                warn!("Malformed Gemini response: missing parts array");
-                record_ai_request_usage(
-                    &state,
-                    &user.id,
-                    tokens_used,
-                    current_week,
-                    total_tokens_accumulated,
-                    "malformed_response",
-                )
-                .await;
-                return Err((StatusCode::BAD_GATEWAY, "Malformed Gemini response".into()));
-            }
-        };
-
-        // Check if the model made a function call.
-        // Find the full part (includes thoughtSignature required by Gemini 3 models).
-        if let Some(fc_part) = parts.iter().find(|part| part.get("functionCall").is_some()) {
-            let fc = fc_part.get("functionCall").unwrap();
-            let fn_name = fc.get("name").and_then(|n| n.as_str()).unwrap_or("");
-            let fn_args = fc.get("args").cloned().unwrap_or(json!({}));
-
-            tool_call_count += 1;
-            info!(
-                function = fn_name,
-                round = round,
-                tool_call = tool_call_count,
-                "AI called tool"
-            );
-
-            if tool_call_count > MAX_TOOL_CALLS {
-                warn!("Tool call budget exhausted, forcing text output");
-                contents.push(candidate.clone());
-                contents.push(json!({
-                    "role": "user",
-                    "parts": [{ "text": "Tool call limit reached. Output your best JSON now using the destinations you already found. Do not call any more tools." }]
-                }));
-                continue;
-            }
-
-            let fn_result = if fn_name == "search_destinations" {
-                let query = fn_args.get("query").and_then(|q| q.as_str()).unwrap_or("");
-                let mode = fn_args
-                    .get("mode")
-                    .and_then(|m| m.as_str())
-                    .unwrap_or("transit");
-                execute_destination_search(&state, query, mode)
-            } else {
-                json!({"error": "unknown function"})
-            };
-
-            // Append the model's full response (preserves thoughtSignature) + our function result
-            contents.push(candidate.clone());
-            contents.push(json!({
-                "role": "user",
-                "parts": [{
-                    "functionResponse": {
-                        "name": fn_name,
-                        "response": fn_result
-                    }
-                }]
-            }));
-
-            // Continue the loop — model will process the results
-            continue;
-        }
-
-        // Model returned text — extract and parse as JSON
-        let text = parts
-            .iter()
-            .find_map(|part| part.get("text").and_then(|t| t.as_str()))
-            .unwrap_or("");
-        let text = strip_markdown_fences(text);
-        let text = text.trim();
-
-        if text.is_empty() {
-            retry_count += 1;
-            warn!(
-                "Gemini returned empty text content (round {}, retry {})",
-                round, retry_count
-            );
-            if retry_count > MAX_RETRIES {
-                record_ai_request_usage(
-                    &state,
-                    &user.id,
-                    tokens_used,
-                    current_week,
-                    total_tokens_accumulated,
-                    "empty_response",
-                )
-                .await;
-                return Err((
-                    StatusCode::BAD_GATEWAY,
-                    "AI returned empty responses".into(),
-                ));
-            }
-            contents.push(candidate.clone());
-            contents.push(json!({
-                "role": "user",
-                "parts": [{ "text": "Your response was empty. Please output the JSON object." }]
-            }));
-            continue;
-        }
-
-        let raw: Value = match serde_json::from_str(text) {
-            Ok(val) => val,
-            Err(err) => {
-                retry_count += 1;
-                warn!(error = %err, round = round, retry = retry_count, "Failed to parse Gemini JSON output");
-                if retry_count > MAX_RETRIES {
-                    record_ai_request_usage(
-                        &state,
-                        &user.id,
-                        tokens_used,
-                        current_week,
-                        total_tokens_accumulated,
-                        "invalid_json",
-                    )
-                    .await;
-                    return Err((StatusCode::BAD_GATEWAY, "AI returned invalid JSON".into()));
-                }
-                contents.push(candidate.clone());
-                contents.push(json!({
-                    "role": "user",
-                    "parts": [{ "text": "That was not valid JSON. Please output ONLY the JSON object with numeric_filters, enum_filters, travel_time_filters, and notes." }]
-                }));
-                continue;
-            }
-        };
-
-        let filters = validate_and_convert(&raw, &state.features_response);
-        let travel_time_filters = validate_travel_time_filters(&raw, &state);
-        let notes = raw
-            .get("notes")
-            .and_then(|val| val.as_str())
-            .unwrap_or("")
-            .to_string();
-
-        // Count matching properties and refine if too restrictive
-        let (match_count, match_bounds) =
-            count_matching_rows(&state, &filters, &travel_time_filters);
-        info!(
-            match_count = match_count,
-            round = round,
-            "AI filter match count"
-        );
-
-        if match_count == 0 {
-            refinement_attempts += 1;
-            let total_rows = state.data.lat.len();
-            info!(
-                attempt = refinement_attempts,
-                "0 matches out of {total_rows} — asking AI to relax filters"
-            );
-
-            if refinement_attempts > MAX_REFINEMENTS {
-                warn!("Refinement budget exhausted, returning filters with 0 matches");
-                record_ai_request_usage(
-                    &state,
-                    &user.id,
-                    tokens_used,
-                    current_week,
-                    total_tokens_accumulated,
-                    "zero_matches",
-                )
-                .await;
-
-                let notes = if notes.is_empty() {
-                    "No properties match these filters. Try relaxing some constraints.".to_string()
-                } else {
-                    format!(
-                        "{}. No properties match. Try relaxing some constraints.",
-                        notes
-                    )
-                };
-
-                return Ok(Json(AiFiltersResponse {
-                    filters,
-                    travel_time_filters,
-                    notes,
-                    match_count: 0,
-                    match_bounds: None,
-                }));
-            }
-
-            let feedback = match refinement_attempts {
-                1 => format!(
-                    "Your proposed filters matched 0 properties out of {total_rows} total. \
-                     The combination is too restrictive. Please widen some numeric ranges \
-                     or add more enum values while keeping the user's intent. \
-                     Output the adjusted JSON."
-                ),
-                2 => format!(
-                    "Still 0 matches out of {total_rows}. Please widen ranges further. \
-                     Output the adjusted JSON."
-                ),
-                _ => format!(
-                    "Still 0 matches out of {total_rows}. Please remove additional filters \
-                     until some properties match, keeping the user's core priority. \
-                     Output the adjusted JSON."
-                ),
-            };
-            contents.push(candidate.clone());
-            contents.push(json!({
-                "role": "user",
-                "parts": [{ "text": feedback }]
-            }));
-            continue;
-        }
-
-        record_ai_request_usage(
-            &state,
-            &user.id,
-            tokens_used,
-            current_week,
-            total_tokens_accumulated,
-            "success",
-        )
-        .await;
-
-        // Log the query to PocketBase (fire-and-forget)
-        let filters_json = serde_json::to_string(&filters).unwrap_or_default();
-        let log_state = state.clone();
-        let log_user_id = user.id.clone();
-        let log_query = req.query.clone();
-        let log_notes = notes.clone();
-        let log_rounds = (round + 1) as u64;
-        tokio::spawn(async move {
-            log_ai_query(
-                &log_state,
-                &log_user_id,
-                &log_query,
-                &filters_json,
-                &log_notes,
-                total_tokens_accumulated,
-                log_rounds,
-            )
-            .await;
-        });
-
-        return Ok(Json(AiFiltersResponse {
-            filters,
-            travel_time_filters,
-            notes,
-            match_count,
-            match_bounds,
-        }));
-    }
-
-    // Exhausted total round budget without getting a valid response
-    warn!(
-        "AI exhausted {} total rounds without final response (tools={}, retries={}, refinements={})",
-        MAX_TOTAL_ROUNDS, tool_call_count, retry_count, refinement_attempts
-    );
-    record_ai_request_usage(
-        &state,
-        &user.id,
-        tokens_used,
-        current_week,
-        total_tokens_accumulated,
-        "incomplete",
-    )
-    .await;
-    Err((
-        StatusCode::BAD_GATEWAY,
-        "AI could not complete the request".into(),
-    ))
-}
-
-/// Maximum travel-time minutes the data can contain. Matches the Java pipeline's
-/// MAX_TRIP_DURATION_MINUTES and the frontend's MAX_TRAVEL_MINUTES.
-const TRAVEL_TIME_MAX_MINUTES: f64 = 90.0;
-
-fn travel_time_minute_field(item: &Value, key: &str) -> Option<f32> {
-    item.get(key)
-        .and_then(|val| val.as_f64())
-        .filter(|val| val.is_finite())
-        .map(|val| val.clamp(0.0, TRAVEL_TIME_MAX_MINUTES) as f32)
-}
-
-fn parse_travel_time_bounds(item: &Value) -> (Option<f32>, Option<f32>) {
-    let explicit_min = travel_time_minute_field(item, "min");
-    let explicit_max = travel_time_minute_field(item, "max");
-
-    let (mut min, mut max) = if explicit_min.is_some() || explicit_max.is_some() {
-        (explicit_min, explicit_max)
-    } else {
-        let value = travel_time_minute_field(item, "value");
-        match (item.get("bound").and_then(|val| val.as_str()), value) {
-            (Some("min"), Some(val)) => (Some(val), None),
-            (Some("max"), Some(val)) => (None, Some(val)),
-            _ => (None, None),
-        }
-    };
-
-    if let (Some(min_val), Some(max_val)) = (min, max) {
-        if min_val > max_val {
-            min = Some(max_val);
-            max = Some(min_val);
-        }
-    }
-
-    (min, max)
-}
-
-/// Validate travel time filters from LLM output against available destinations.
-fn validate_travel_time_filters(raw: &Value, state: &AppState) -> Vec<TravelTimeFilter> {
-    let arr = match raw
-        .get("travel_time_filters")
-        .and_then(|val| val.as_array())
-    {
-        Some(arr) => arr,
-        None => return Vec::new(),
-    };
-
-    let tt_store = &state.travel_time_store;
-    let mut results = Vec::new();
-
-    for item in arr {
-        let mode = match item.get("mode").and_then(|val| val.as_str()) {
-            Some(mode) => mode,
-            None => continue,
-        };
-        let slug = match item.get("slug").and_then(|val| val.as_str()) {
-            Some(slug) => slug,
-            None => continue,
-        };
-        let label = item
-            .get("label")
-            .and_then(|val| val.as_str())
-            .unwrap_or(slug);
-
-        // Verify this destination actually exists
-        if !tt_store.has_destination(mode, slug) {
-            warn!(
-                mode = mode,
-                slug = slug,
-                "AI suggested non-existent destination"
-            );
-            continue;
-        }
-
-        let (min, max) = parse_travel_time_bounds(item);
-
-        // Only include if at least one bound is set
-        if min.is_some() || max.is_some() {
-            results.push(TravelTimeFilter {
-                mode: mode.to_string(),
-                slug: slug.to_string(),
-                label: label.to_string(),
-                min,
-                max,
-            });
-        }
-    }
-
-    results
-}
-
-/// Validate LLM output against feature metadata and convert to FeatureFilters format.
-///
-/// Input format (array-based, each numeric filter sets one bound):
-/// ```json
-/// {
-///   "numeric_filters": [{"name": "Last known price", "bound": "max", "value": 300000}],
-///   "enum_filters": [{"name": "Leasehold/Freehold", "values": ["Freehold"]}]
-/// }
-/// ```
-///
-/// Output format (FeatureFilters):
-/// ```json
-/// { "Last known price": [0, 300000], "Leasehold/Freehold": ["Freehold"] }
-/// ```
-fn validate_and_convert(raw: &Value, features: &FeaturesResponse) -> Value {
-    let mut result = serde_json::Map::new();
-
-    // Build lookup maps from feature metadata.
-    // Store both slider bounds (min/max from percentiles) and true data bounds
-    // (histogram.min/max) so one-sided AI filters use the full data range.
-    let mut numeric_features: rustc_hash::FxHashMap<&str, (f32, f32, f32, f32)> =
-        rustc_hash::FxHashMap::default();
-    let mut enum_features: rustc_hash::FxHashMap<&str, &[String]> =
-        rustc_hash::FxHashMap::default();
-
-    for group in &features.groups {
-        for feature in &group.features {
-            match feature {
-                FeatureInfo::Numeric {
-                    name,
-                    min,
-                    max,
-                    histogram,
-                    ..
-                } => {
-                    numeric_features.insert(name, (*min, *max, histogram.min, histogram.max));
-                }
-                FeatureInfo::Enum { name, values, .. } => {
-                    enum_features.insert(name, values);
-                }
-            }
-        }
-    }
-
-    // Process numeric filters — each sets one bound (min or max).
-    // The unset side uses the true data min/max (from histogram), not
-    // the slider bounds (percentile-based), so a "max" filter for crime
-    // produces [0, value] rather than [2nd-percentile, value].
-    if let Some(arr) = raw.get("numeric_filters").and_then(|val| val.as_array()) {
-        for item in arr {
-            let raw_name = match item.get("name").and_then(|val| val.as_str()) {
-                Some(name) => name,
-                None => continue,
-            };
-            let name = canonical_filter_name(raw_name);
-            let (slider_min, slider_max, data_min, data_max) =
-                match numeric_features.get(name.as_str()) {
-                    Some(range) => *range,
-                    None => continue,
-                };
-            let bound = match item.get("bound").and_then(|val| val.as_str()) {
-                Some(b) => b,
-                None => continue,
-            };
-            // Clamp value to true data range (not slider range)
-            let value = match item.get("value").and_then(|val| val.as_f64()) {
-                Some(v) => v.max(data_min as f64).min(data_max as f64) as f32,
-                None => continue,
-            };
-            let (filter_min, filter_max) = match bound {
-                "min" => (value, data_max),
-                "max" => (data_min, value),
-                _ => continue,
-            };
-            // Only include if range is narrower than full slider range
-            if filter_min > slider_min || filter_max < slider_max {
-                result.insert(name, json!([filter_min, filter_max]));
-            }
-        }
-    }
-
-    // Process enum filters
-    if let Some(arr) = raw.get("enum_filters").and_then(|val| val.as_array()) {
-        for item in arr {
-            let raw_name = match item.get("name").and_then(|val| val.as_str()) {
-                Some(name) => name,
-                None => continue,
-            };
-            let name = canonical_filter_name(raw_name);
-            let valid_values = match enum_features.get(name.as_str()) {
-                Some(values) => *values,
-                None => continue,
-            };
-            if let Some(selected) = item.get("values").and_then(|val| val.as_array()) {
-                let valid: Vec<&str> = selected
-                    .iter()
-                    .filter_map(|item| item.as_str())
-                    .filter(|str_val| valid_values.iter().any(|known| known == str_val))
-                    .collect();
-                if !valid.is_empty() && valid.len() < valid_values.len() {
-                    result.insert(name, json!(valid));
-                }
-            }
-        }
-    }
-
-    Value::Object(result)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn strip_fences_json_tag() {
-        let input = "```json\n{\"a\": 1}\n```";
-        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
-    }
-
-    #[test]
-    fn strip_fences_no_tag() {
-        let input = "```\n{\"a\": 1}\n```";
-        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
-    }
-
-    #[test]
-    fn strip_fences_passthrough() {
-        let input = "{\"a\": 1}";
-        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
-    }
-
-    #[test]
-    fn strip_fences_whitespace() {
-        let input = "  ```json\n  {\"a\": 1}  \n```  ";
-        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
-    }
-
-    #[test]
-    fn synthetic_filter_keys_are_normalized_to_backend_names() {
-        assert_eq!(
-            canonical_filter_name("Schools:primary:good:0"),
-            "Good+ primary school catchments"
-        );
-        // Legacy keys still carry a distance segment; it is ignored.
-        assert_eq!(
-            canonical_filter_name("Schools:primary:good:2:0"),
-            "Good+ primary school catchments"
-        );
-        assert_eq!(
-            canonical_filter_name("Specific crimes:Burglary%20%28avg%2Fyr%29:1"),
-            "Burglary (avg/yr)"
-        );
-        assert_eq!(
-            canonical_filter_name("Political vote share:%25%20Labour:0"),
-            "% Labour"
-        );
-        assert_eq!(
-            canonical_filter_name(
-                "Transport distance:Distance%20to%20nearest%20amenity%20%28Bus%20stop%29%20%28km%29:0"
-            ),
-            "Distance to nearest amenity (Bus stop) (km)"
-        );
-    }
-
-    #[test]
-    fn context_filters_are_normalized_before_prompting() {
-        let filters = json!({
-            "Political vote share:%25%20Green:0": [40, 100],
-            "Estimated current price": [0, 500000],
-        });
-
-        let normalized = normalize_context_filters(&filters);
-        assert_eq!(normalized["% Green"], json!([40, 100]));
-        assert_eq!(normalized["Estimated current price"], json!([0, 500000]));
-    }
-
-    #[test]
-    fn travel_time_bounds_accept_min_max_schema() {
-        let item = json!({ "min": 30, "max": 45 });
-        assert_eq!(parse_travel_time_bounds(&item), (Some(30.0), Some(45.0)));
-    }
-
-    #[test]
-    fn travel_time_bounds_accept_legacy_bound_value_schema() {
-        let item = json!({ "bound": "max", "value": 30 });
-        assert_eq!(parse_travel_time_bounds(&item), (None, Some(30.0)));
-    }
-
-    #[test]
-    fn travel_time_bounds_clamp_and_order_range() {
-        // Data ceiling is 90 (matches Java MAX_TRIP_DURATION_MINUTES).
-        // Inputs outside [0, 90] clamp; min/max ordering is preserved as-given here.
-        let item = json!({ "min": 150, "max": -10 });
-        assert_eq!(parse_travel_time_bounds(&item), (Some(0.0), Some(90.0)));
-    }
-}
diff --git a/server-rs/src/routes/ai_filters/handler.rs b/server-rs/src/routes/ai_filters/handler.rs
new file mode 100644
index 0000000..090bdab
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/handler.rs
@@ -0,0 +1,448 @@
+//! The `POST /api/ai-filters` route handler: rate limiting, the Gemini
+//! function-calling conversation loop, and zero-match refinement.
+
+use std::sync::Arc;
+
+use axum::extract::State;
+use axum::http::StatusCode;
+use axum::response::Json;
+use axum::Extension;
+use metrics::counter;
+use serde_json::{json, Value};
+use tracing::{info, warn};
+
+use crate::auth::OptionalUser;
+use crate::consts::{AI_FILTERS_MAX_TOKENS, AI_FILTERS_TEMPERATURE, AI_FILTERS_WEEKLY_TOKEN_LIMIT};
+use crate::pocketbase::log_ai_query;
+use crate::state::SharedState;
+use crate::utils::gemini_chat;
+
+use super::matching::count_matching_rows;
+use super::parsing::{
+    normalize_context_filters, strip_markdown_fences, validate_and_convert,
+    validate_travel_time_filters,
+};
+use super::tools::{build_tool_declarations, execute_destination_search};
+use super::usage::{current_week_number, fetch_ai_usage, record_ai_request_usage};
+use super::{AiFiltersRequest, AiFiltersResponse};
+
+/// Budget limits for the Gemini conversation loop. Separate counters prevent
+/// tool calls (destination searches) from starving JSON retries or zero-match
+/// refinements.
+const MAX_TOOL_CALLS: usize = 4;
+const MAX_RETRIES: usize = 3;
+const MAX_REFINEMENTS: u32 = 3;
+const MAX_TOTAL_ROUNDS: usize = 10;
+
+const MAX_AI_QUERY_CHARS: usize = 5000;
+
+pub async fn post_ai_filters(
+    State(shared): State<Arc<SharedState>>,
+    Extension(user): Extension<OptionalUser>,
+    Json(req): Json<AiFiltersRequest>,
+) -> Result<Json<AiFiltersResponse>, (StatusCode, String)> {
+    let state = shared.load_state();
+    // Auth check
+    let user = user
+        .0
+        .ok_or((StatusCode::UNAUTHORIZED, "Login required".into()))?;
+
+    if req.query.chars().count() > MAX_AI_QUERY_CHARS {
+        counter!("ai_requests_total", "status" => "query_too_long").increment(1);
+        return Err((
+            StatusCode::PAYLOAD_TOO_LARGE,
+            format!("Query too long (max {MAX_AI_QUERY_CHARS} chars)"),
+        ));
+    }
+
+    // Check weekly token usage
+    let current_week = current_week_number();
+    let (stored_tokens, stored_week) = fetch_ai_usage(&state, &user.id).await?;
+    let tokens_used = if stored_week == current_week {
+        stored_tokens
+    } else {
+        0
+    };
+
+    if tokens_used >= AI_FILTERS_WEEKLY_TOKEN_LIMIT {
+        counter!("ai_requests_total", "status" => "rate_limited").increment(1);
+        return Err((
+            StatusCode::TOO_MANY_REQUESTS,
+            "Weekly AI usage limit reached. Resets next week.".into(),
+        ));
+    }
+
+    info!(query = %req.query, user_id = %user.id, "POST /api/ai-filters");
+
+    let tools = build_tool_declarations(&state);
+
+    // Build user message with optional context for conversational refinement
+    let user_text = if let Some(ref ctx) = req.context {
+        let mut msg = String::new();
+        msg.push_str("Currently active filters:\n");
+        let normalized_filters = normalize_context_filters(&ctx.filters);
+        msg.push_str(&serde_json::to_string(&normalized_filters).unwrap_or_default());
+        if !ctx.travel_time.is_empty() {
+            msg.push_str("\nCurrently active travel time filters:\n");
+            for tt in &ctx.travel_time {
+                let bounds = match (tt.min, tt.max) {
+                    (Some(min), Some(max)) => format!("{}-{} min", min, max),
+                    (Some(min), None) => format!("min {} min", min),
+                    (None, Some(max)) => format!("max {} min", max),
+                    (None, None) => "no range".to_string(),
+                };
+                msg.push_str(&format!("- {} to {} ({})\n", tt.mode, tt.label, bounds));
+            }
+        }
+        msg.push_str(&format!("\nUser request: {}", req.query));
+        msg
+    } else {
+        req.query.clone()
+    };
+
+    let mut contents = vec![json!({
+        "role": "user",
+        "parts": [{ "text": user_text }]
+    })];
+
+    let mut total_tokens_accumulated: u64 = 0;
+    let mut tool_call_count = 0usize;
+    let mut retry_count = 0usize;
+    let mut refinement_attempts = 0u32;
+
+    // Function calling loop: model may call search_destinations, we execute and feed back
+    for round in 0..MAX_TOTAL_ROUNDS {
+        let body = json!({
+            "systemInstruction": {
+                "parts": [{ "text": state.ai_filters_system_prompt }]
+            },
+            "contents": contents,
+            "tools": tools,
+            "generationConfig": {
+                "temperature": AI_FILTERS_TEMPERATURE,
+                "maxOutputTokens": AI_FILTERS_MAX_TOKENS,
+                "thinkingConfig": { "thinkingLevel": "LOW" },
+            }
+        });
+
+        let json_resp = match gemini_chat(
+            &state.http_client,
+            &state.gemini_api_key,
+            &state.gemini_model,
+            &body,
+        )
+        .await
+        {
+            Ok(resp) => resp,
+            Err(err) => {
+                record_ai_request_usage(
+                    &state,
+                    &user.id,
+                    tokens_used,
+                    current_week,
+                    total_tokens_accumulated,
+                    "llm_error",
+                )
+                .await;
+                return Err(err);
+            }
+        };
+
+        // Accumulate token usage
+        total_tokens_accumulated += json_resp
+            .get("usageMetadata")
+            .and_then(|md| md.get("totalTokenCount"))
+            .and_then(|tc| tc.as_u64())
+            .unwrap_or(0);
+
+        let candidate = match json_resp
+            .get("candidates")
+            .and_then(|cs| cs.get(0))
+            .and_then(|c| c.get("content"))
+        {
+            Some(candidate) => candidate,
+            None => {
+                warn!("Malformed Gemini response: missing candidates[0].content");
+                record_ai_request_usage(
+                    &state,
+                    &user.id,
+                    tokens_used,
+                    current_week,
+                    total_tokens_accumulated,
+                    "malformed_response",
+                )
+                .await;
+                return Err((StatusCode::BAD_GATEWAY, "Malformed Gemini response".into()));
+            }
+        };
+
+        let parts = match candidate.get("parts").and_then(|p| p.as_array()) {
+            Some(parts) => parts,
+            None => {
+                warn!("Malformed Gemini response: missing parts array");
+                record_ai_request_usage(
+                    &state,
+                    &user.id,
+                    tokens_used,
+                    current_week,
+                    total_tokens_accumulated,
+                    "malformed_response",
+                )
+                .await;
+                return Err((StatusCode::BAD_GATEWAY, "Malformed Gemini response".into()));
+            }
+        };
+
+        // Check if the model made a function call.
+        // Find the full part (includes thoughtSignature required by Gemini 3 models).
+        if let Some(fc_part) = parts.iter().find(|part| part.get("functionCall").is_some()) {
+            let fc = fc_part.get("functionCall").unwrap();
+            let fn_name = fc.get("name").and_then(|n| n.as_str()).unwrap_or("");
+            let fn_args = fc.get("args").cloned().unwrap_or(json!({}));
+
+            tool_call_count += 1;
+            info!(
+                function = fn_name,
+                round = round,
+                tool_call = tool_call_count,
+                "AI called tool"
+            );
+
+            if tool_call_count > MAX_TOOL_CALLS {
+                warn!("Tool call budget exhausted, forcing text output");
+                contents.push(candidate.clone());
+                contents.push(json!({
+                    "role": "user",
+                    "parts": [{ "text": "Tool call limit reached. Output your best JSON now using the destinations you already found. Do not call any more tools." }]
+                }));
+                continue;
+            }
+
+            let fn_result = if fn_name == "search_destinations" {
+                let query = fn_args.get("query").and_then(|q| q.as_str()).unwrap_or("");
+                let mode = fn_args
+                    .get("mode")
+                    .and_then(|m| m.as_str())
+                    .unwrap_or("transit");
+                execute_destination_search(&state, query, mode)
+            } else {
+                json!({"error": "unknown function"})
+            };
+
+            // Append the model's full response (preserves thoughtSignature) + our function result
+            contents.push(candidate.clone());
+            contents.push(json!({
+                "role": "user",
+                "parts": [{
+                    "functionResponse": {
+                        "name": fn_name,
+                        "response": fn_result
+                    }
+                }]
+            }));
+
+            // Continue the loop — model will process the results
+            continue;
+        }
+
+        // Model returned text — extract and parse as JSON
+        let text = parts
+            .iter()
+            .find_map(|part| part.get("text").and_then(|t| t.as_str()))
+            .unwrap_or("");
+        let text = strip_markdown_fences(text);
+        let text = text.trim();
+
+        if text.is_empty() {
+            retry_count += 1;
+            warn!(
+                "Gemini returned empty text content (round {}, retry {})",
+                round, retry_count
+            );
+            if retry_count > MAX_RETRIES {
+                record_ai_request_usage(
+                    &state,
+                    &user.id,
+                    tokens_used,
+                    current_week,
+                    total_tokens_accumulated,
+                    "empty_response",
+                )
+                .await;
+                return Err((
+                    StatusCode::BAD_GATEWAY,
+                    "AI returned empty responses".into(),
+                ));
+            }
+            contents.push(candidate.clone());
+            contents.push(json!({
+                "role": "user",
+                "parts": [{ "text": "Your response was empty. Please output the JSON object." }]
+            }));
+            continue;
+        }
+
+        let raw: Value = match serde_json::from_str(text) {
+            Ok(val) => val,
+            Err(err) => {
+                retry_count += 1;
+                warn!(error = %err, round = round, retry = retry_count, "Failed to parse Gemini JSON output");
+                if retry_count > MAX_RETRIES {
+                    record_ai_request_usage(
+                        &state,
+                        &user.id,
+                        tokens_used,
+                        current_week,
+                        total_tokens_accumulated,
+                        "invalid_json",
+                    )
+                    .await;
+                    return Err((StatusCode::BAD_GATEWAY, "AI returned invalid JSON".into()));
+                }
+                contents.push(candidate.clone());
+                contents.push(json!({
+                    "role": "user",
+                    "parts": [{ "text": "That was not valid JSON. Please output ONLY the JSON object with numeric_filters, enum_filters, travel_time_filters, and notes." }]
+                }));
+                continue;
+            }
+        };
+
+        let filters = validate_and_convert(&raw, &state.features_response);
+        let travel_time_filters = validate_travel_time_filters(&raw, &state);
+        let notes = raw
+            .get("notes")
+            .and_then(|val| val.as_str())
+            .unwrap_or("")
+            .to_string();
+
+        // Count matching properties and refine if too restrictive
+        let (match_count, match_bounds) =
+            count_matching_rows(&state, &filters, &travel_time_filters);
+        info!(
+            match_count = match_count,
+            round = round,
+            "AI filter match count"
+        );
+
+        if match_count == 0 {
+            refinement_attempts += 1;
+            let total_rows = state.data.lat.len();
+            info!(
+                attempt = refinement_attempts,
+                "0 matches out of {total_rows} — asking AI to relax filters"
+            );
+
+            if refinement_attempts > MAX_REFINEMENTS {
+                warn!("Refinement budget exhausted, returning filters with 0 matches");
+                record_ai_request_usage(
+                    &state,
+                    &user.id,
+                    tokens_used,
+                    current_week,
+                    total_tokens_accumulated,
+                    "zero_matches",
+                )
+                .await;
+
+                let notes = if notes.is_empty() {
+                    "No properties match these filters. Try relaxing some constraints.".to_string()
+                } else {
+                    format!(
+                        "{}. No properties match. Try relaxing some constraints.",
+                        notes
+                    )
+                };
+
+                return Ok(Json(AiFiltersResponse {
+                    filters,
+                    travel_time_filters,
+                    notes,
+                    match_count: 0,
+                    match_bounds: None,
+                }));
+            }
+
+            let feedback = match refinement_attempts {
+                1 => format!(
+                    "Your proposed filters matched 0 properties out of {total_rows} total. \
+                     The combination is too restrictive. Please widen some numeric ranges \
+                     or add more enum values while keeping the user's intent. \
+                     Output the adjusted JSON."
+                ),
+                2 => format!(
+                    "Still 0 matches out of {total_rows}. Please widen ranges further. \
+                     Output the adjusted JSON."
+                ),
+                _ => format!(
+                    "Still 0 matches out of {total_rows}. Please remove additional filters \
+                     until some properties match, keeping the user's core priority. \
+                     Output the adjusted JSON."
+                ),
+            };
+            contents.push(candidate.clone());
+            contents.push(json!({
+                "role": "user",
+                "parts": [{ "text": feedback }]
+            }));
+            continue;
+        }
+
+        record_ai_request_usage(
+            &state,
+            &user.id,
+            tokens_used,
+            current_week,
+            total_tokens_accumulated,
+            "success",
+        )
+        .await;
+
+        // Log the query to PocketBase (fire-and-forget)
+        let filters_json = serde_json::to_string(&filters).unwrap_or_default();
+        let log_state = state.clone();
+        let log_user_id = user.id.clone();
+        let log_query = req.query.clone();
+        let log_notes = notes.clone();
+        let log_rounds = (round + 1) as u64;
+        tokio::spawn(async move {
+            log_ai_query(
+                &log_state,
+                &log_user_id,
+                &log_query,
+                &filters_json,
+                &log_notes,
+                total_tokens_accumulated,
+                log_rounds,
+            )
+            .await;
+        });
+
+        return Ok(Json(AiFiltersResponse {
+            filters,
+            travel_time_filters,
+            notes,
+            match_count,
+            match_bounds,
+        }));
+    }
+
+    // Exhausted total round budget without getting a valid response
+    warn!(
+        "AI exhausted {} total rounds without final response (tools={}, retries={}, refinements={})",
+        MAX_TOTAL_ROUNDS, tool_call_count, retry_count, refinement_attempts
+    );
+    record_ai_request_usage(
+        &state,
+        &user.id,
+        tokens_used,
+        current_week,
+        total_tokens_accumulated,
+        "incomplete",
+    )
+    .await;
+    Err((
+        StatusCode::BAD_GATEWAY,
+        "AI could not complete the request".into(),
+    ))
+}
diff --git a/server-rs/src/routes/ai_filters/matching.rs b/server-rs/src/routes/ai_filters/matching.rs
new file mode 100644
index 0000000..74686d3
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/matching.rs
@@ -0,0 +1,158 @@
+//! Counting properties that match the AI-proposed property and travel time
+//! filters, and computing a camera-friendly bounding box of the matches.
+
+use serde_json::Value;
+use tracing::warn;
+
+use crate::data::travel_time::TravelData;
+use crate::parsing::{parse_filters_with_poi, row_passes_filters, row_passes_poi_filters};
+use crate::state::AppState;
+
+use super::{MatchBounds, TravelTimeFilter};
+
+/// Bounding box over matched coordinates, trimmed to the 5th–95th percentile
+/// per axis (when there are enough points) so a handful of remote outliers
+/// doesn't zoom the camera out to all of England.
+fn percentile_trimmed_bounds(mut lats: Vec<f32>, mut lons: Vec<f32>) -> Option<MatchBounds> {
+    if lats.is_empty() || lats.len() != lons.len() {
+        return None;
+    }
+    lats.sort_unstable_by(f32::total_cmp);
+    lons.sort_unstable_by(f32::total_cmp);
+    let last = lats.len() - 1;
+    let (lo, hi) = if lats.len() >= 20 {
+        let trim = lats.len() / 20;
+        (trim, last - trim)
+    } else {
+        (0, last)
+    };
+    Some(MatchBounds {
+        south: lats[lo],
+        north: lats[hi],
+        west: lons[lo],
+        east: lons[hi],
+    })
+}
+
+/// Convert validated filter JSON back to the `;;`-separated filter string format
+/// that `parse_filters` expects.
+///
+/// Numeric: `{"name": [min, max]}` → `name:min:max`
+/// Enum: `{"name": ["val1", "val2"]}` → `name:val1|val2`
+fn filters_to_filter_string(filters: &Value) -> String {
+    let obj = match filters.as_object() {
+        Some(obj) => obj,
+        None => return String::new(),
+    };
+
+    let mut parts = Vec::new();
+    for (name, value) in obj {
+        if let Some(arr) = value.as_array() {
+            if arr.len() == 2 && arr[0].is_number() && arr[1].is_number() {
+                let min = arr[0].as_f64().unwrap_or(0.0);
+                let max = arr[1].as_f64().unwrap_or(0.0);
+                parts.push(format!("{name}:{min}:{max}"));
+            } else if !arr.is_empty() && arr[0].is_string() {
+                let values: Vec<&str> = arr.iter().filter_map(|v| v.as_str()).collect();
+                if !values.is_empty() {
+                    parts.push(format!("{name}:{}", values.join("|")));
+                }
+            }
+        }
+    }
+
+    parts.join(";;")
+}
+
+/// Count how many rows in the property dataset pass the given property filters
+/// AND travel time filters. Travel time data is loaded from the TravelTimeStore
+/// and checked per-postcode (same logic as hexagons.rs).
+pub(super) fn count_matching_rows(
+    state: &AppState,
+    filters: &Value,
+    travel_time_filters: &[TravelTimeFilter],
+) -> (usize, Option<MatchBounds>) {
+    let filter_str = filters_to_filter_string(filters);
+
+    let quant = state.data.quant_ref();
+    let poi_quant = state.data.poi_metrics.quant_ref();
+    let (parsed_filters, parsed_enum_filters, parsed_poi_filters) = if filter_str.is_empty() {
+        (Vec::new(), Vec::new(), Vec::new())
+    } else {
+        match parse_filters_with_poi(
+            Some(&filter_str),
+            &state.feature_name_to_index,
+            &state.data.enum_values,
+            &quant,
+            &state.data.poi_metrics.name_to_index,
+            &poi_quant,
+        ) {
+            Ok(f) => f,
+            Err(err) => {
+                warn!("Failed to parse filters for match count: {err}");
+                return (0, None);
+            }
+        }
+    };
+
+    // Load travel time data for each filter entry
+    let travel_data: Vec<(TravelData, Option<f32>, Option<f32>)> = travel_time_filters
+        .iter()
+        .filter_map(|ttf| {
+            let data = state.travel_time_store.get(&ttf.mode, &ttf.slug).ok()?;
+            Some((data, ttf.min, ttf.max))
+        })
+        .collect();
+    let has_travel = !travel_data.is_empty();
+
+    let feature_data = &state.data.feature_data;
+    let num_features = state.data.num_features;
+    let num_rows = state.data.lat.len();
+    let (pc_interner, pc_keys) = state.data.postcode_parts();
+    let has_poi_filters = !parsed_poi_filters.is_empty();
+
+    let mut count = 0usize;
+    let mut matched_lats: Vec<f32> = Vec::new();
+    let mut matched_lons: Vec<f32> = Vec::new();
+    for (row, pc_key) in pc_keys.iter().enumerate().take(num_rows) {
+        if !row_passes_filters(
+            row,
+            &parsed_filters,
+            &parsed_enum_filters,
+            feature_data,
+            num_features,
+        ) {
+            continue;
+        }
+        if has_poi_filters
+            && !row_passes_poi_filters(row, &parsed_poi_filters, &state.data.poi_metrics)
+        {
+            continue;
+        }
+
+        if has_travel {
+            let postcode = pc_interner.resolve(pc_key);
+            let mut passes_travel = true;
+            for (data, fmin, fmax) in &travel_data {
+                let pass = if let Some(mins) = data.get(postcode).map(|r| r.minutes as f32) {
+                    fmin.is_none_or(|min| mins >= min) && fmax.is_none_or(|max| mins <= max)
+                } else {
+                    false // no travel data → postcode not reachable
+                };
+                if !pass {
+                    passes_travel = false;
+                    break;
+                }
+            }
+            if !passes_travel {
+                continue;
+            }
+        }
+
+        count += 1;
+        matched_lats.push(state.data.lat[row]);
+        matched_lons.push(state.data.lon[row]);
+    }
+
+    (count, percentile_trimmed_bounds(matched_lats, matched_lons))
+}
diff --git a/server-rs/src/routes/ai_filters/mod.rs b/server-rs/src/routes/ai_filters/mod.rs
new file mode 100644
index 0000000..8cfce63
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/mod.rs
@@ -0,0 +1,81 @@
+//! AI filters: translate a natural-language property query into validated
+//! filter settings via Gemini.
+//!
+//! Split by concern:
+//! - [`handler`]: the `POST /api/ai-filters` route handler and Gemini
+//!   conversation loop
+//! - [`prompt`]: system prompt building (precomputed at startup)
+//! - [`tools`]: the `search_destinations` tool declaration and execution
+//! - [`parsing`]: LLM response parsing and validation against feature metadata
+//! - [`matching`]: counting properties that match the proposed filters
+//! - [`usage`]: weekly token usage tracking / rate limiting
+
+mod handler;
+mod matching;
+mod parsing;
+mod prompt;
+mod tools;
+mod usage;
+
+pub use handler::post_ai_filters;
+pub use prompt::build_system_prompt;
+
+use serde::{Deserialize, Serialize};
+use serde_json::Value;
+
+#[derive(Deserialize)]
+pub struct AiFiltersContext {
+    filters: Value,
+    #[serde(default)]
+    travel_time: Vec<AiTravelTimeContext>,
+}
+
+#[derive(Deserialize)]
+pub struct AiTravelTimeContext {
+    mode: String,
+    label: String,
+    min: Option<f32>,
+    max: Option<f32>,
+}
+
+#[derive(Deserialize)]
+pub struct AiFiltersRequest {
+    query: String,
+    /// Current filters for conversational refinement (e.g. "make it cheaper")
+    context: Option<AiFiltersContext>,
+}
+
+#[derive(Serialize)]
+pub struct TravelTimeFilter {
+    mode: String,
+    slug: String,
+    label: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    min: Option<f32>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    max: Option<f32>,
+}
+
+#[derive(Serialize)]
+pub struct AiFiltersResponse {
+    filters: Value,
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    travel_time_filters: Vec<TravelTimeFilter>,
+    /// What the LLM couldn't map to existing filters (empty if everything matched)
+    #[serde(skip_serializing_if = "String::is_empty")]
+    notes: String,
+    /// Number of properties matching the proposed property and travel time filters.
+    match_count: usize,
+    /// Bounding box of the matching properties so the client can move the
+    /// camera to where matches actually are. Absent when nothing matches.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    match_bounds: Option<MatchBounds>,
+}
+
+#[derive(Serialize)]
+pub struct MatchBounds {
+    south: f32,
+    west: f32,
+    north: f32,
+    east: f32,
+}
diff --git a/server-rs/src/routes/ai_filters/parsing.rs b/server-rs/src/routes/ai_filters/parsing.rs
new file mode 100644
index 0000000..2ab5044
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/parsing.rs
@@ -0,0 +1,385 @@
+//! LLM response parsing: stripping markdown fences, normalizing frontend
+//! synthetic filter keys, and validating proposed filters against feature
+//! metadata and available travel destinations.
+
+use serde_json::{json, Map, Value};
+use tracing::warn;
+
+use crate::routes::{FeatureInfo, FeaturesResponse};
+use crate::state::AppState;
+
+use super::TravelTimeFilter;
+
+/// Strip markdown code fences (```json ... ``` or ``` ... ```) from LLM output.
+/// Models occasionally wrap JSON in markdown fencing even when told not to.
+pub(super) fn strip_markdown_fences(text: &str) -> &str {
+    let trimmed = text.trim();
+
+    // Try ```json\n...\n``` or ```\n...\n```
+    if let Some(rest) = trimmed.strip_prefix("```") {
+        // Skip optional language tag (e.g. "json")
+        let rest = if let Some(newline_pos) = rest.find('\n') {
+            &rest[newline_pos + 1..]
+        } else {
+            return trimmed;
+        };
+        if let Some(content) = rest.strip_suffix("```") {
+            return content.trim();
+        }
+    }
+
+    trimmed
+}
+
+fn school_feature_name_from_key(name: &str) -> Option<&'static str> {
+    let rest = name.strip_prefix("Schools:")?;
+    let mut parts = rest.split(':');
+    let phase = parts.next()?;
+    let rating = parts.next()?;
+
+    match (phase, rating) {
+        ("primary", "good") => Some("Good+ primary school catchments"),
+        ("secondary", "good") => Some("Good+ secondary school catchments"),
+        ("primary", "outstanding") => Some("Outstanding primary school catchments"),
+        ("secondary", "outstanding") => Some("Outstanding secondary school catchments"),
+        _ => None,
+    }
+}
+
+fn decode_synthetic_feature_key(name: &str, prefix: &str) -> Option<String> {
+    let rest = name.strip_prefix(prefix)?;
+    let (encoded, _id) = rest.rsplit_once(':')?;
+    urlencoding::decode(encoded)
+        .ok()
+        .map(|decoded| decoded.into_owned())
+}
+
+/// Convert frontend synthetic filter keys back to backend feature names.
+///
+/// The React filter UI stores configurable cards under keys such as
+/// `Political vote share:%25%20Labour:0`. The LLM and backend validators need
+/// the real feature name (`% Labour`) instead.
+fn backend_filter_name(name: &str) -> Option<String> {
+    if let Some(feature_name) = school_feature_name_from_key(name) {
+        return Some(feature_name.to_string());
+    }
+
+    for prefix in [
+        "Specific crimes:",
+        "Political vote share:",
+        "Ethnicities:",
+        "Amenity distance:",
+        "Transport distance:",
+        "Amenities within 2km:",
+        "Amenities within 5km:",
+    ] {
+        if let Some(feature_name) = decode_synthetic_feature_key(name, prefix) {
+            return Some(feature_name);
+        }
+    }
+
+    None
+}
+
+fn canonical_filter_name(name: &str) -> String {
+    backend_filter_name(name).unwrap_or_else(|| name.to_string())
+}
+
+pub(super) fn normalize_context_filters(filters: &Value) -> Value {
+    let Some(obj) = filters.as_object() else {
+        return filters.clone();
+    };
+
+    let mut normalized = Map::with_capacity(obj.len());
+    for (name, value) in obj {
+        normalized.insert(canonical_filter_name(name), value.clone());
+    }
+    Value::Object(normalized)
+}
+
+/// Maximum travel-time minutes the data can contain. Matches the Java pipeline's
+/// MAX_TRIP_DURATION_MINUTES and the frontend's MAX_TRAVEL_MINUTES.
+const TRAVEL_TIME_MAX_MINUTES: f64 = 90.0;
+
+fn travel_time_minute_field(item: &Value, key: &str) -> Option<f32> {
+    item.get(key)
+        .and_then(|val| val.as_f64())
+        .filter(|val| val.is_finite())
+        .map(|val| val.clamp(0.0, TRAVEL_TIME_MAX_MINUTES) as f32)
+}
+
+fn parse_travel_time_bounds(item: &Value) -> (Option<f32>, Option<f32>) {
+    let explicit_min = travel_time_minute_field(item, "min");
+    let explicit_max = travel_time_minute_field(item, "max");
+
+    let (mut min, mut max) = if explicit_min.is_some() || explicit_max.is_some() {
+        (explicit_min, explicit_max)
+    } else {
+        let value = travel_time_minute_field(item, "value");
+        match (item.get("bound").and_then(|val| val.as_str()), value) {
+            (Some("min"), Some(val)) => (Some(val), None),
+            (Some("max"), Some(val)) => (None, Some(val)),
+            _ => (None, None),
+        }
+    };
+
+    if let (Some(min_val), Some(max_val)) = (min, max) {
+        if min_val > max_val {
+            min = Some(max_val);
+            max = Some(min_val);
+        }
+    }
+
+    (min, max)
+}
+
+/// Validate travel time filters from LLM output against available destinations.
+pub(super) fn validate_travel_time_filters(raw: &Value, state: &AppState) -> Vec<TravelTimeFilter> {
+    let arr = match raw
+        .get("travel_time_filters")
+        .and_then(|val| val.as_array())
+    {
+        Some(arr) => arr,
+        None => return Vec::new(),
+    };
+
+    let tt_store = &state.travel_time_store;
+    let mut results = Vec::new();
+
+    for item in arr {
+        let mode = match item.get("mode").and_then(|val| val.as_str()) {
+            Some(mode) => mode,
+            None => continue,
+        };
+        let slug = match item.get("slug").and_then(|val| val.as_str()) {
+            Some(slug) => slug,
+            None => continue,
+        };
+        let label = item
+            .get("label")
+            .and_then(|val| val.as_str())
+            .unwrap_or(slug);
+
+        // Verify this destination actually exists
+        if !tt_store.has_destination(mode, slug) {
+            warn!(
+                mode = mode,
+                slug = slug,
+                "AI suggested non-existent destination"
+            );
+            continue;
+        }
+
+        let (min, max) = parse_travel_time_bounds(item);
+
+        // Only include if at least one bound is set
+        if min.is_some() || max.is_some() {
+            results.push(TravelTimeFilter {
+                mode: mode.to_string(),
+                slug: slug.to_string(),
+                label: label.to_string(),
+                min,
+                max,
+            });
+        }
+    }
+
+    results
+}
+
+/// Validate LLM output against feature metadata and convert to FeatureFilters format.
+///
+/// Input format (array-based, each numeric filter sets one bound):
+/// ```json
+/// {
+///   "numeric_filters": [{"name": "Last known price", "bound": "max", "value": 300000}],
+///   "enum_filters": [{"name": "Leasehold/Freehold", "values": ["Freehold"]}]
+/// }
+/// ```
+///
+/// Output format (FeatureFilters):
+/// ```json
+/// { "Last known price": [0, 300000], "Leasehold/Freehold": ["Freehold"] }
+/// ```
+pub(super) fn validate_and_convert(raw: &Value, features: &FeaturesResponse) -> Value {
+    let mut result = serde_json::Map::new();
+
+    // Build lookup maps from feature metadata.
+    // Store both slider bounds (min/max from percentiles) and true data bounds
+    // (histogram.min/max) so one-sided AI filters use the full data range.
+    let mut numeric_features: rustc_hash::FxHashMap<&str, (f32, f32, f32, f32)> =
+        rustc_hash::FxHashMap::default();
+    let mut enum_features: rustc_hash::FxHashMap<&str, &[String]> =
+        rustc_hash::FxHashMap::default();
+
+    for group in &features.groups {
+        for feature in &group.features {
+            match feature {
+                FeatureInfo::Numeric {
+                    name,
+                    min,
+                    max,
+                    histogram,
+                    ..
+                } => {
+                    numeric_features.insert(name, (*min, *max, histogram.min, histogram.max));
+                }
+                FeatureInfo::Enum { name, values, .. } => {
+                    enum_features.insert(name, values);
+                }
+            }
+        }
+    }
+
+    // Process numeric filters — each sets one bound (min or max).
+    // The unset side uses the true data min/max (from histogram), not
+    // the slider bounds (percentile-based), so a "max" filter for crime
+    // produces [0, value] rather than [2nd-percentile, value].
+    if let Some(arr) = raw.get("numeric_filters").and_then(|val| val.as_array()) {
+        for item in arr {
+            let raw_name = match item.get("name").and_then(|val| val.as_str()) {
+                Some(name) => name,
+                None => continue,
+            };
+            let name = canonical_filter_name(raw_name);
+            let (slider_min, slider_max, data_min, data_max) =
+                match numeric_features.get(name.as_str()) {
+                    Some(range) => *range,
+                    None => continue,
+                };
+            let bound = match item.get("bound").and_then(|val| val.as_str()) {
+                Some(b) => b,
+                None => continue,
+            };
+            // Clamp value to true data range (not slider range)
+            let value = match item.get("value").and_then(|val| val.as_f64()) {
+                Some(v) => v.max(data_min as f64).min(data_max as f64) as f32,
+                None => continue,
+            };
+            let (filter_min, filter_max) = match bound {
+                "min" => (value, data_max),
+                "max" => (data_min, value),
+                _ => continue,
+            };
+            // Only include if range is narrower than full slider range
+            if filter_min > slider_min || filter_max < slider_max {
+                result.insert(name, json!([filter_min, filter_max]));
+            }
+        }
+    }
+
+    // Process enum filters
+    if let Some(arr) = raw.get("enum_filters").and_then(|val| val.as_array()) {
+        for item in arr {
+            let raw_name = match item.get("name").and_then(|val| val.as_str()) {
+                Some(name) => name,
+                None => continue,
+            };
+            let name = canonical_filter_name(raw_name);
+            let valid_values = match enum_features.get(name.as_str()) {
+                Some(values) => *values,
+                None => continue,
+            };
+            if let Some(selected) = item.get("values").and_then(|val| val.as_array()) {
+                let valid: Vec<&str> = selected
+                    .iter()
+                    .filter_map(|item| item.as_str())
+                    .filter(|str_val| valid_values.iter().any(|known| known == str_val))
+                    .collect();
+                if !valid.is_empty() && valid.len() < valid_values.len() {
+                    result.insert(name, json!(valid));
+                }
+            }
+        }
+    }
+
+    Value::Object(result)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn strip_fences_json_tag() {
+        let input = "```json\n{\"a\": 1}\n```";
+        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
+    }
+
+    #[test]
+    fn strip_fences_no_tag() {
+        let input = "```\n{\"a\": 1}\n```";
+        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
+    }
+
+    #[test]
+    fn strip_fences_passthrough() {
+        let input = "{\"a\": 1}";
+        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
+    }
+
+    #[test]
+    fn strip_fences_whitespace() {
+        let input = "  ```json\n  {\"a\": 1}  \n```  ";
+        assert_eq!(strip_markdown_fences(input), "{\"a\": 1}");
+    }
+
+    #[test]
+    fn synthetic_filter_keys_are_normalized_to_backend_names() {
+        assert_eq!(
+            canonical_filter_name("Schools:primary:good:0"),
+            "Good+ primary school catchments"
+        );
+        // Legacy keys still carry a distance segment; it is ignored.
+        assert_eq!(
+            canonical_filter_name("Schools:primary:good:2:0"),
+            "Good+ primary school catchments"
+        );
+        assert_eq!(
+            canonical_filter_name("Specific crimes:Burglary%20%28avg%2Fyr%29:1"),
+            "Burglary (avg/yr)"
+        );
+        assert_eq!(
+            canonical_filter_name("Political vote share:%25%20Labour:0"),
+            "% Labour"
+        );
+        assert_eq!(
+            canonical_filter_name(
+                "Transport distance:Distance%20to%20nearest%20amenity%20%28Bus%20stop%29%20%28km%29:0"
+            ),
+            "Distance to nearest amenity (Bus stop) (km)"
+        );
+    }
+
+    #[test]
+    fn context_filters_are_normalized_before_prompting() {
+        let filters = json!({
+            "Political vote share:%25%20Green:0": [40, 100],
+            "Estimated current price": [0, 500000],
+        });
+
+        let normalized = normalize_context_filters(&filters);
+        assert_eq!(normalized["% Green"], json!([40, 100]));
+        assert_eq!(normalized["Estimated current price"], json!([0, 500000]));
+    }
+
+    #[test]
+    fn travel_time_bounds_accept_min_max_schema() {
+        let item = json!({ "min": 30, "max": 45 });
+        assert_eq!(parse_travel_time_bounds(&item), (Some(30.0), Some(45.0)));
+    }
+
+    #[test]
+    fn travel_time_bounds_accept_legacy_bound_value_schema() {
+        let item = json!({ "bound": "max", "value": 30 });
+        assert_eq!(parse_travel_time_bounds(&item), (None, Some(30.0)));
+    }
+
+    #[test]
+    fn travel_time_bounds_clamp_and_order_range() {
+        // Data ceiling is 90 (matches Java MAX_TRIP_DURATION_MINUTES).
+        // Inputs outside [0, 90] clamp; min/max ordering is preserved as-given here.
+        let item = json!({ "min": 150, "max": -10 });
+        assert_eq!(parse_travel_time_bounds(&item), (Some(0.0), Some(90.0)));
+    }
+}
diff --git a/server-rs/src/routes/ai_filters/prompt.rs b/server-rs/src/routes/ai_filters/prompt.rs
new file mode 100644
index 0000000..b12fb3c
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/prompt.rs
@@ -0,0 +1,282 @@
+//! System prompt building for the AI filters assistant.
+
+use crate::routes::{FeatureInfo, FeaturesResponse};
+
+/// Build the complete system prompt for AI filters.
+///
+/// Contains: role instructions, feature catalogue, travel time info,
+/// few-shot examples, output rules.
+/// Precomputed at startup and cached in AppState.
+pub fn build_system_prompt(
+    features: &FeaturesResponse,
+    mode_destinations: &[(String, usize)],
+) -> String {
+    let mut parts = Vec::new();
+
+    parts.push(
+        "You are a UK property search assistant. \
+         The user describes their ideal property or area in natural language. \
+         Translate their description into filter settings using ONLY the features listed below.\n\
+         \n\
+         Rules:\n\
+         - ONLY set filters the user explicitly mentioned or clearly implied.\n\
+         - Leave out any filter the user did not mention. Empty arrays are fine.\n\
+         - Each numeric filter sets ONE bound only: \"min\" (at least this value) \
+           or \"max\" (at most this value). Never set two filters on the same feature.\n\
+         - Use EXACT feature names from the list — spelling, capitalisation, and punctuation must match.\n\
+         - \"cheap\" / \"affordable\" = lower price range. \"expensive\" = higher price range.\n\
+         - \"low crime\" / \"safe\" = low values on the Serious crime (avg/yr) and Minor crime (avg/yr) \
+           features (area-normalised incident density near the postcode). Prefer these aggregates for broad \
+           area safety; use specific crime features only when the user names a crime type.\n\
+         - \"quiet\" = low Noise (dB). \"green\" / \"near parks\" = high Number of amenities (Park) within 2km \
+           or low Distance to nearest park (km), depending on wording.\n\
+         - \"good schools\" = Good+ school features. \"outstanding schools\" = Outstanding school features.\n\
+         - Amenities and transport stops are normal filters in the feature catalogue. \
+           For \"near a bus stop\", \"near a station\", \"near shops\", etc., use the exact \
+           Distance to nearest amenity (...) or Number of amenities (...) feature when available.\n\
+         - Politics/elections are normal filters in the Neighbours group. Use exact vote share \
+           features such as % Labour, % Conservative, % Liberal Democrat, % Reform UK, % Green, \
+           % Other parties, or Voter turnout (%) when the user asks for political character.\n\
+         - When the user says a number like \"under 400k\", interpret it as 400000.\n\
+         - When the user says \"3 bed\" or \"3 bedroom\", use Number of bedrooms & living rooms \
+           (note: this counts bedrooms + living rooms combined, so 3 bed ~ min 4).\n\
+         - If the user mentions something that has no matching filter, put it in \"notes\" \
+           as a short phrase (e.g. \"No filter for: garden, sea view\"). \
+           If everything was matched, set \"notes\" to an empty string.\n\
+         \n\
+         CONVERSATIONAL REFINEMENT:\n\
+         The user's message may include their currently active filters as context. \
+         When context is provided:\n\
+         - \"make it cheaper\" / \"lower the price\" = adjust the existing price filter down\n\
+         - \"also add ...\" / \"and good schools\" = keep existing filters and add new ones\n\
+         - \"remove the ...\" / \"drop the ...\" = return filters WITHOUT the mentioned one\n\
+         - If the request is a completely new search (not a refinement), ignore the context \
+           and build filters from scratch.\n\
+         - Always output the COMPLETE set of filters (existing + modified), not just the changes."
+            .to_string(),
+    );
+
+    // Travel time section with available modes
+    let modes_list = mode_destinations
+        .iter()
+        .map(|(mode, count)| format!("- {} ({} destinations available)", mode, count))
+        .collect::<Vec<_>>()
+        .join("\n");
+
+    parts.push(format!(
+        "\n--- TRAVEL TIME FILTERS ---\n\
+         You can add travel time filters when the user mentions commute times, \
+         proximity to places, or wanting to be near/within X minutes of somewhere.\n\
+         \n\
+         Available travel-time modes (only use modes that have destinations):\n\
+         {}\n\
+         - \"car\" / \"drive\" / \"driving\" = car mode\n\
+         - \"cycle\" / \"bike\" / \"cycling\" = bicycle mode\n\
+         - \"walk\" / \"walking\" / \"on foot\" = walking mode\n\
+         - \"train\" / \"tube\" / \"bus\" / \"public transport\" / \"commute\" = transit mode\n\
+         - \"without buses\" / \"no bus\" / \"rail only\" = transit-no-bus mode\n\
+         - \"no change\" / \"no transfer\" / \"direct\" / \"single bus/train\" = transit-no-change mode\n\
+         - \"no change and no bus\" / \"direct rail/tube\" = transit-no-change-no-bus mode\n\
+         - If a mode appears in the available mode list but is not named above, you may still \
+           use the exact mode string from the list.\n\
+         \n\
+         When the user mentions a specific place, you MUST call the search_destinations \
+         tool to find the exact slug. Use the name and slug from the search results.\n\
+         If search_destinations returns an empty array, the destination is not available — \
+         mention it in \"notes\" (e.g. \"No travel data for: Gatwick Airport\") and do NOT \
+         include a travel_time_filter for it.\n\
+         \n\
+         Travel time values are in MINUTES (0-90 range; data is capped at 90 min).\n\
+         - \"within 30 minutes\" = set \"max\": 30\n\
+         - \"at least 10 minutes\" = set \"min\": 10\n\
+         - \"30-45 minute commute\" = set \"min\": 30 and \"max\": 45 on the same travel_time_filter\n\
+         - If only a max is given, omit min (and vice versa). Do not use bound/value for travel time.\n\
+         \n\
+         INFERRING TRANSPORT MODE (when the user does not specify one explicitly):\n\
+         - \"commute\" to a major city centre or station = transit\n\
+         - \"near\" / \"close to\" a city centre or station = transit\n\
+         - \"near\" / \"close to\" a smaller town, village, or rural area = car\n\
+         - \"drive\" / \"driving distance\" / \"driving time\" = always car\n\
+         - If multiple modes are plausible, prefer transit for urban destinations \
+           (London, Manchester, Birmingham, Leeds, etc.) and car for everything else.",
+        modes_list,
+    ));
+
+    // Feature guidance
+    parts.push(
+        "\n--- DATA SOURCE ---\n\
+         The data is historical property sales from the Land Registry.\n\
+         \n\
+         Use these features for price queries:\n\
+         - For purchase price: use \"Estimated current price\" or \"Last known price\"\n\
+         - For price per sqm: use \"Est. price per sqm\"\n\
+         - For rent estimates: use \"Estimated monthly rent\""
+            .to_string(),
+    );
+
+    // Feature catalogue
+    parts.push("\n--- AVAILABLE FEATURES ---\n".to_string());
+    for group in &features.groups {
+        parts.push(format!("## {}", group.name));
+        for feature in &group.features {
+            match feature {
+                FeatureInfo::Numeric {
+                    name,
+                    min,
+                    max,
+                    description,
+                    prefix,
+                    suffix,
+                    ..
+                } => {
+                    parts.push(format!(
+                        "- \"{}\" (numeric, {}{:.0}{} to {}{:.0}{}): {}",
+                        name, prefix, min, suffix, prefix, max, suffix, description
+                    ));
+                }
+                FeatureInfo::Enum {
+                    name,
+                    values,
+                    description,
+                    ..
+                } => {
+                    parts.push(format!(
+                        "- \"{}\" (enum, values: [{}]): {}",
+                        name,
+                        values
+                            .iter()
+                            .map(|val| format!("\"{}\"", val))
+                            .collect::<Vec<_>>()
+                            .join(", "),
+                        description
+                    ));
+                }
+            }
+        }
+    }
+
+    // Few-shot examples
+    parts.push("\n--- EXAMPLES ---\n".to_string());
+
+    parts.push(
+        "User: \"cheap freehold house under 400k\"\n\
+         Output: {\"numeric_filters\": [{\"name\": \"Last known price\", \"bound\": \"max\", \"value\": 400000}], \
+         \"enum_filters\": [{\"name\": \"Leasehold/Freehold\", \"values\": [\"Freehold\"]}, \
+         {\"name\": \"Property type\", \"values\": [\"Detached\", \"Semi-Detached\", \"Terraced\"]}], \
+         \"travel_time_filters\": [], \
+         \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"safe quiet area with good schools and parks\"\n\
+         Output: {\"numeric_filters\": [\
+         {\"name\": \"Serious crime (avg/yr)\", \"bound\": \"max\", \"value\": 5}, \
+         {\"name\": \"Minor crime (avg/yr)\", \"bound\": \"max\", \"value\": 20}, \
+         {\"name\": \"Noise (dB)\", \"bound\": \"max\", \"value\": 55}, \
+         {\"name\": \"Good+ primary school catchments\", \"bound\": \"min\", \"value\": 2}, \
+         {\"name\": \"Good+ secondary school catchments\", \"bound\": \"min\", \"value\": 1}, \
+         {\"name\": \"Number of amenities (Park) within 2km\", \"bound\": \"min\", \"value\": 3}], \
+         \"enum_filters\": [], \"travel_time_filters\": [], \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"quiet area with outstanding schools\"\n\
+         Output: {\"numeric_filters\": [\
+         {\"name\": \"Noise (dB)\", \"bound\": \"max\", \"value\": 55}, \
+         {\"name\": \"Outstanding primary school catchments\", \"bound\": \"min\", \"value\": 1}, \
+         {\"name\": \"Outstanding secondary school catchments\", \"bound\": \"min\", \"value\": 1}], \
+         \"enum_filters\": [], \"travel_time_filters\": [], \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"3 bed flat under 300k with fast broadband near the beach\"\n\
+         Output: {\"numeric_filters\": [\
+         {\"name\": \"Last known price\", \"bound\": \"max\", \"value\": 300000}, \
+         {\"name\": \"Number of bedrooms & living rooms\", \"bound\": \"min\", \"value\": 4}], \
+         \"enum_filters\": [{\"name\": \"Property type\", \"values\": [\"Flats/Maisonettes\"]}, \
+         {\"name\": \"Max available download speed (Mbps)\", \"values\": [\"100\", \"300\", \"1000\"]}], \
+         \"travel_time_filters\": [], \
+         \"notes\": \"No filter for: beach proximity\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"within 30 minutes commute of Kings Cross, under 500k\"\n\
+         (After calling search_destinations for \"Kings Cross\" with mode \"transit\" \
+         and getting [{\"name\": \"Kings Cross\", \"slug\": \"kings-cross\", \"place_type\": \"station\"}])\n\
+         Output: {\"numeric_filters\": [\
+         {\"name\": \"Last known price\", \"bound\": \"max\", \"value\": 500000}], \
+         \"enum_filters\": [], \
+         \"travel_time_filters\": [{\"mode\": \"transit\", \"slug\": \"kings-cross\", \
+         \"label\": \"Kings Cross\", \"max\": 30}], \
+         \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"family home with garden, 45 min drive from Manchester, good schools\"\n\
+         (After calling search_destinations for \"Manchester\" with mode \"car\" \
+         and getting [{\"name\": \"Manchester\", \"slug\": \"manchester\", \"place_type\": \"city\"}])\n\
+         Output: {\"numeric_filters\": [\
+         {\"name\": \"Total floor area (sqm)\", \"bound\": \"min\", \"value\": 100}, \
+         {\"name\": \"Number of bedrooms & living rooms\", \"bound\": \"min\", \"value\": 5}, \
+         {\"name\": \"Good+ primary school catchments\", \"bound\": \"min\", \"value\": 2}, \
+         {\"name\": \"Good+ secondary school catchments\", \"bound\": \"min\", \"value\": 1}], \
+         \"enum_filters\": [{\"name\": \"Property type\", \
+         \"values\": [\"Detached\", \"Semi-Detached\"]}], \
+         \"travel_time_filters\": [{\"mode\": \"car\", \"slug\": \"manchester\", \
+         \"label\": \"Manchester\", \"max\": 45}], \
+         \"notes\": \"No filter for: garden\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"Labour-voting area with low burglary and a station nearby\"\n\
+         Output: {\"numeric_filters\": [\
+         {\"name\": \"% Labour\", \"bound\": \"min\", \"value\": 40}, \
+         {\"name\": \"Burglary (avg/yr)\", \"bound\": \"max\", \"value\": 10}, \
+         {\"name\": \"Distance to nearest amenity (Rail station) (km)\", \"bound\": \"max\", \"value\": 1}], \
+         \"enum_filters\": [], \"travel_time_filters\": [], \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    // Examples showing rent and price features
+    parts.push(
+        "\nUser: \"2 bed flat with rent under £1500/month\"\n\
+         Output: {\
+         \"numeric_filters\": [{\"name\": \"Estimated monthly rent\", \"bound\": \"max\", \"value\": 1500}], \
+         \"enum_filters\": [{\"name\": \"Property type\", \"values\": [\"Flats/Maisonettes\"]}], \
+         \"travel_time_filters\": [], \
+         \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    parts.push(
+        "\nUser: \"3 bed house under 500k with good schools\"\n\
+         Output: {\
+         \"numeric_filters\": [{\"name\": \"Estimated current price\", \"bound\": \"max\", \"value\": 500000}, \
+         {\"name\": \"Good+ primary school catchments\", \"bound\": \"min\", \"value\": 2}], \
+         \"enum_filters\": [{\"name\": \"Property type\", \
+         \"values\": [\"Detached\", \"Semi-Detached\", \"Terraced\"]}], \
+         \"travel_time_filters\": [], \
+         \"notes\": \"\"}"
+            .to_string(),
+    );
+
+    // Output format reminder
+    parts.push(
+        "\n--- OUTPUT FORMAT ---\n\
+         {\"numeric_filters\": [...], \"enum_filters\": [...], \
+         \"travel_time_filters\": [{\"mode\": \"...\", \"slug\": \"...\", \"label\": \"...\", \
+         \"min\": N, \"max\": N}, ...], \"notes\": \"...\"}\n\
+         - travel_time_filters: min and max are both optional, but include at least one. \
+         Use ONLY slugs returned by search_destinations. If a place isn't found, mention it in notes.\n\
+         Respond with ONLY the JSON object. No explanation."
+            .to_string(),
+    );
+
+    parts.join("\n")
+}
diff --git a/server-rs/src/routes/ai_filters/tools.rs b/server-rs/src/routes/ai_filters/tools.rs
new file mode 100644
index 0000000..b5b44d9
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/tools.rs
@@ -0,0 +1,188 @@
+//! The `search_destinations` Gemini tool: declaration and execution against
+//! PlaceData + TravelTimeStore.
+
+use serde_json::{json, Value};
+use tracing::info;
+
+use crate::data::slugify;
+use crate::state::AppState;
+
+/// Build the Gemini tool declaration for destination search.
+pub(super) fn build_tool_declarations(state: &AppState) -> Value {
+    let modes: Vec<&str> = state
+        .travel_time_store
+        .available_modes
+        .iter()
+        .map(|mode| mode.as_str())
+        .collect();
+
+    json!([{
+        "functionDeclarations": [{
+            "name": "search_destinations",
+            "description": "Search for available travel time destinations (cities, stations, towns) that have precomputed travel time data. Call this when the user mentions wanting to be near, close to, or within a certain travel time of a specific place.",
+            "parameters": {
+                "type": "object",
+                "properties": {
+                    "query": {
+                        "type": "string",
+                        "description": "Place name to search for (e.g. 'Manchester', 'Kings Cross', 'Heathrow')"
+                    },
+                    "mode": {
+                        "type": "string",
+                        "enum": modes,
+                        "description": "Transport mode to search destinations for"
+                    }
+                },
+                "required": ["query", "mode"]
+            }
+        }]
+    }])
+}
+
+/// Execute a destination search against PlaceData + TravelTimeStore.
+/// Returns matching destinations as a JSON value with `results` and optional `message`.
+///
+/// Uses word-based matching: all words in the query must appear somewhere in the
+/// place name (order-independent). Also matches against slugs for short queries.
+pub(super) fn execute_destination_search(state: &AppState, query: &str, mode: &str) -> Value {
+    let query_lower = query.to_lowercase();
+    let query_words: Vec<&str> = query_lower.split_whitespace().collect();
+    let query_slug = slugify(query);
+    let tt_store = &state.travel_time_store;
+    let pd = &state.place_data;
+
+    let slug_set = match tt_store.destinations.get(mode) {
+        Some(slugs) => slugs,
+        None => {
+            return json!({ "results": [], "message": format!("No travel data available for mode '{}'", mode) })
+        }
+    };
+
+    // Find places matching the query that have travel time data.
+    // A place matches if ALL query words appear in its name, OR its slug matches the query slug.
+    let mut matches: Vec<(usize, String, u8, u32)> = pd
+        .name_lower
+        .iter()
+        .enumerate()
+        .filter_map(|(idx, name_lower)| {
+            if !pd.travel_destination[idx] {
+                return None;
+            }
+            let words_match = query_words.iter().all(|word| name_lower.contains(word));
+            let slug = slugify(&pd.name[idx]);
+            let slug_match = slug.contains(&query_slug) || query_slug.contains(&slug);
+            if !words_match && !slug_match {
+                return None;
+            }
+            if slug_set.contains(&slug) {
+                Some((idx, slug, pd.type_rank[idx], pd.population[idx]))
+            } else {
+                None
+            }
+        })
+        .collect();
+
+    // Sort: type rank asc, population desc
+    matches.sort_unstable_by(|a, b| a.2.cmp(&b.2).then(b.3.cmp(&a.3)));
+    matches.truncate(10);
+
+    if matches.is_empty() {
+        // Check if the query matched a city that lacks its own travel data.
+        // If so, return nearby stations within that city as suggestions.
+        let matched_city_name: Option<&str> =
+            pd.name_lower
+                .iter()
+                .enumerate()
+                .find_map(|(idx, name_lower)| {
+                    if !pd.travel_destination[idx] {
+                        return None;
+                    }
+                    let words_match = query_words.iter().all(|word| name_lower.contains(word));
+                    let slug = slugify(&pd.name[idx]);
+                    let slug_match = slug.contains(&query_slug) || query_slug.contains(&slug);
+                    if (words_match || slug_match) && pd.type_rank[idx] == 0 {
+                        Some(pd.name[idx].as_str())
+                    } else {
+                        None
+                    }
+                });
+
+        if let Some(city_name) = matched_city_name {
+            let city_lower = city_name.to_lowercase();
+            let mut city_matches: Vec<(usize, String, u8, u32)> = pd
+                .city
+                .iter()
+                .enumerate()
+                .filter_map(|(idx, city_opt)| {
+                    if !pd.travel_destination[idx] {
+                        return None;
+                    }
+                    let city = city_opt.as_deref()?;
+                    if city.to_lowercase() != city_lower {
+                        return None;
+                    }
+                    let slug = slugify(&pd.name[idx]);
+                    if slug_set.contains(&slug) {
+                        Some((idx, slug, pd.type_rank[idx], pd.population[idx]))
+                    } else {
+                        None
+                    }
+                })
+                .collect();
+
+            city_matches.sort_unstable_by(|a, b| a.2.cmp(&b.2).then(b.3.cmp(&a.3)));
+            city_matches.truncate(10);
+
+            if !city_matches.is_empty() {
+                let results: Vec<Value> = city_matches
+                    .into_iter()
+                    .map(|(idx, slug, ..)| {
+                        json!({
+                            "name": pd.name[idx],
+                            "slug": slug,
+                            "place_type": pd.place_type.get(idx).to_string(),
+                        })
+                    })
+                    .collect();
+
+                info!(
+                    query = query,
+                    city = city_name,
+                    results = results.len(),
+                    "Destination search fell back to city stations"
+                );
+
+                return json!({
+                    "results": results,
+                    "message": format!(
+                        "No travel data for '{}' directly. Pick one of these nearby stations:",
+                        city_name
+                    )
+                });
+            }
+        }
+
+        info!(
+            query = query,
+            mode = mode,
+            "Destination search returned no results"
+        );
+        return json!({
+            "results": [],
+            "message": format!("No travel time data available for '{}' by {}. This destination cannot be used as a travel time filter.", query, mode)
+        });
+    }
+
+    let results: Vec<Value> = matches
+        .into_iter()
+        .map(|(idx, slug, ..)| {
+            json!({
+                "name": pd.name[idx],
+                "slug": slug,
+                "place_type": pd.place_type.get(idx).to_string(),
+            })
+        })
+        .collect();
+
+    json!({ "results": results })
+}
diff --git a/server-rs/src/routes/ai_filters/usage.rs b/server-rs/src/routes/ai_filters/usage.rs
new file mode 100644
index 0000000..73ddf40
--- /dev/null
+++ b/server-rs/src/routes/ai_filters/usage.rs
@@ -0,0 +1,119 @@
+//! Weekly AI token usage tracking and rate limiting, persisted on the user's
+//! PocketBase record.
+
+use axum::http::StatusCode;
+use metrics::counter;
+use serde_json::{json, Value};
+use tracing::warn;
+
+use crate::pocketbase::get_superuser_token;
+use crate::state::AppState;
+
+/// Monotonically increasing week number derived from Unix epoch.
+/// Resets every 7 days (604800 seconds). Used for weekly rate limiting.
+pub(super) fn current_week_number() -> u64 {
+    let secs = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        // Only possible if the system clock is before 1970; fall back to
+        // week 0 rather than panicking inside a request handler.
+        .unwrap_or_default()
+        .as_secs();
+    secs / 604_800
+}
+
+/// Fetch the user's current AI token usage from PocketBase.
+/// Returns `(tokens_used, week_number)`.
+pub(super) async fn fetch_ai_usage(
+    state: &AppState,
+    user_id: &str,
+) -> Result<(u64, u64), (StatusCode, String)> {
+    let token = get_superuser_token(state).await.map_err(|err| {
+        warn!("Failed to auth superuser for AI usage check: {err}");
+        (StatusCode::BAD_GATEWAY, "Internal error".into())
+    })?;
+
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/users/records/{user_id}");
+    let resp = state
+        .http_client
+        .get(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .send()
+        .await
+        .map_err(|err| {
+            warn!("Failed to fetch user record for AI usage: {err}");
+            (StatusCode::BAD_GATEWAY, "Internal error".into())
+        })?;
+
+    if !resp.status().is_success() {
+        let status = resp.status();
+        warn!("PocketBase user fetch failed ({status})");
+        return Err((StatusCode::BAD_GATEWAY, "Internal error".into()));
+    }
+
+    let body: Value = resp.json().await.map_err(|err| {
+        warn!("Failed to parse user record: {err}");
+        (StatusCode::BAD_GATEWAY, "Internal error".into())
+    })?;
+
+    let tokens_used = body
+        .get("ai_tokens_used")
+        .and_then(|val| val.as_u64())
+        .unwrap_or(0);
+    let week = body
+        .get("ai_tokens_week")
+        .and_then(|val| val.as_u64())
+        .unwrap_or(0);
+
+    Ok((tokens_used, week))
+}
+
+/// Update the user's AI token usage in PocketBase.
+/// Best-effort — logs warnings on failure but does not propagate errors.
+async fn update_ai_usage(state: &AppState, user_id: &str, tokens_used: u64, week: u64) {
+    let token = match get_superuser_token(state).await {
+        Ok(tk) => tk,
+        Err(err) => {
+            warn!("Failed to auth superuser for AI usage update: {err}");
+            return;
+        }
+    };
+
+    let pb_url = state.pocketbase_url.trim_end_matches('/');
+    let url = format!("{pb_url}/api/collections/users/records/{user_id}");
+    let res = state
+        .http_client
+        .patch(&url)
+        .header("Authorization", format!("Bearer {token}"))
+        .json(&json!({
+            "ai_tokens_used": tokens_used,
+            "ai_tokens_week": week,
+        }))
+        .send()
+        .await;
+
+    match res {
+        Ok(resp) if resp.status().is_success() => {}
+        Ok(resp) => {
+            let status = resp.status();
+            warn!("Failed to update AI usage ({status})");
+        }
+        Err(err) => warn!("Failed to update AI usage: {err}"),
+    }
+}
+
+pub(super) async fn record_ai_request_usage(
+    state: &AppState,
+    user_id: &str,
+    existing_tokens_used: u64,
+    week: u64,
+    request_tokens_used: u64,
+    status: &'static str,
+) {
+    if request_tokens_used > 0 {
+        let new_total = existing_tokens_used.saturating_add(request_tokens_used);
+        update_ai_usage(state, user_id, new_total, week).await;
+        counter!("ai_tokens_total").increment(request_tokens_used);
+    }
+    counter!("ai_requests_total", "status" => status).increment(1);
+}
diff --git a/server-rs/src/routes/export.rs b/server-rs/src/routes/export.rs
index f36687e..b057285 100644
--- a/server-rs/src/routes/export.rs
+++ b/server-rs/src/routes/export.rs
@@ -780,31 +780,28 @@ pub async fn get_export(
         // groups themselves; postcodes within a group are sorted alphabetically.
         // Each group carries a rolled-up summary aggregate for its header row.
         let outcode_groups: Vec<OutcodeGroup> = {
-            let mut order: Vec<String> = Vec::new();
-            let mut by_outcode: FxHashMap<String, OutcodeGroup> = FxHashMap::default();
+            let mut groups: Vec<OutcodeGroup> = Vec::new();
+            let mut idx_by_outcode: FxHashMap<String, usize> = FxHashMap::default();
             for (i, (pc_idx, agg)) in postcode_aggs.iter().enumerate() {
                 let outcode = outcode_of(&postcode_data.postcodes[*pc_idx]).to_string();
-                let group = by_outcode.entry(outcode.clone()).or_insert_with(|| {
-                    order.push(outcode.clone());
-                    OutcodeGroup {
-                        outcode: outcode.clone(),
+                let idx = *idx_by_outcode.entry(outcode.clone()).or_insert_with(|| {
+                    groups.push(OutcodeGroup {
+                        outcode,
                         members: Vec::new(),
                         summary: PostcodeExportAgg::new(total_export_features),
-                    }
+                    });
+                    groups.len() - 1
                 });
-                group.members.push(i);
-                group.summary.merge_from(agg);
+                groups[idx].members.push(i);
+                groups[idx].summary.merge_from(agg);
             }
-            for group in by_outcode.values_mut() {
+            for group in &mut groups {
                 group.members.sort_by(|&a, &b| {
                     postcode_data.postcodes[postcode_aggs[a].0]
                         .cmp(&postcode_data.postcodes[postcode_aggs[b].0])
                 });
             }
-            order
-                .into_iter()
-                .map(|outcode| by_outcode.remove(&outcode).unwrap())
-                .collect()
+            groups
         };
 
         // Build Excel workbook with two sheets
diff --git a/server-rs/src/routes/hexagon_stats.rs b/server-rs/src/routes/hexagon_stats.rs
index 736b233..844175c 100644
--- a/server-rs/src/routes/hexagon_stats.rs
+++ b/server-rs/src/routes/hexagon_stats.rs
@@ -130,6 +130,11 @@ pub struct HexagonStatsResponse {
     pub price_history: Vec<PricePoint>,
     #[serde(skip_serializing_if = "Vec::is_empty")]
     pub crime_by_year: Vec<CrimeYearStats>,
+    /// Latest year in the crime dataset as a whole. When a selection's series
+    /// end earlier (force-level publication gap, e.g. Greater Manchester),
+    /// the client captions the data as stale.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub crime_latest_year: Option<i32>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub central_postcode: Option<String>,
     #[serde(skip_serializing_if = "Vec::is_empty")]
@@ -645,12 +650,19 @@ pub async fn get_hexagon_stats(
             "GET /api/hexagon-stats"
         );
 
+        let crime_latest_year = if crime_by_year.is_empty() {
+            None
+        } else {
+            stats::crime_latest_available_year(&state.crime_by_year)
+        };
+
         Ok(HexagonStatsResponse {
             count: total_count,
             numeric_features,
             enum_features: enum_features_out,
             price_history,
             crime_by_year,
+            crime_latest_year,
             central_postcode,
             filter_exclusions,
         })
diff --git a/server-rs/src/routes/pb_proxy.rs b/server-rs/src/routes/pb_proxy.rs
index 9c839a4..c654cef 100644
--- a/server-rs/src/routes/pb_proxy.rs
+++ b/server-rs/src/routes/pb_proxy.rs
@@ -36,8 +36,9 @@ fn is_allowed_pb_path(path: &str) -> bool {
 
 /// Dedicated HTTP client for proxying — does not follow redirects so 3xx
 /// responses are passed through to the browser (needed for OAuth flows).
-/// No overall timeout because SSE (Server-Sent Events) connections used by
-/// PocketBase realtime/OAuth2 are long-lived streams.
+/// No client-wide timeout because SSE (Server-Sent Events) connections used
+/// by PocketBase realtime/OAuth2 are long-lived streams; non-realtime
+/// requests get a per-request timeout instead (see PROXY_REQUEST_TIMEOUT).
 static PROXY_CLIENT: LazyLock<reqwest::Client> = LazyLock::new(|| {
     reqwest::Client::builder()
         .redirect(reqwest::redirect::Policy::none())
@@ -47,6 +48,11 @@ static PROXY_CLIENT: LazyLock<reqwest::Client> = LazyLock::new(|| {
         .expect("Failed to build proxy HTTP client")
 });
 
+/// Timeout for proxied requests other than the realtime SSE stream, so a hung
+/// PocketBase cannot pile up handlers indefinitely. Generous enough for file
+/// uploads/downloads over slow links.
+const PROXY_REQUEST_TIMEOUT: Duration = Duration::from_secs(30);
+
 pub async fn proxy_to_pocketbase(
     State(shared): State<Arc<SharedState>>,
     req: Request,
@@ -58,10 +64,7 @@ pub async fn proxy_to_pocketbase(
     let target_path = path.strip_prefix("/pb").unwrap_or(path);
     if !is_allowed_pb_path(target_path) {
         warn!(path = %target_path, "Rejected PocketBase proxy request to disallowed path");
-        return Response::builder()
-            .status(StatusCode::NOT_FOUND)
-            .body(Body::empty())
-            .unwrap();
+        return StatusCode::NOT_FOUND.into_response();
     }
     let query = req
         .uri()
@@ -73,6 +76,12 @@ pub async fn proxy_to_pocketbase(
     let method = req.method().clone();
     let mut builder = PROXY_CLIENT.request(method, &url);
 
+    // The realtime SSE stream is intentionally unbounded; everything else
+    // must complete within the timeout.
+    if target_path != "/api/realtime" {
+        builder = builder.timeout(PROXY_REQUEST_TIMEOUT);
+    }
+
     // Forward only safe headers (allowlist)
     const ALLOWED_HEADERS: &[&str] = &[
         "content-type",
@@ -96,10 +105,7 @@ pub async fn proxy_to_pocketbase(
         Ok(bytes) => bytes,
         Err(err) => {
             warn!("Failed to read request body: {err}");
-            return Response::builder()
-                .status(StatusCode::BAD_REQUEST)
-                .body(Body::from("Failed to read request body"))
-                .unwrap();
+            return (StatusCode::BAD_REQUEST, "Failed to read request body").into_response();
         }
     };
     builder = builder.body(body_bytes);
@@ -129,14 +135,14 @@ pub async fn proxy_to_pocketbase(
             // realtime system and OAuth2 flow — buffering would hang forever
             // since SSE responses never complete.
             let body = Body::from_stream(upstream.bytes_stream());
-            response.body(body).unwrap()
+            response.body(body).unwrap_or_else(|err| {
+                warn!("Failed to build proxied response: {err}");
+                (StatusCode::BAD_GATEWAY, "Invalid upstream response").into_response()
+            })
         }
         Err(err) => {
             warn!("PocketBase proxy error: {err}");
-            Response::builder()
-                .status(StatusCode::BAD_GATEWAY)
-                .body(Body::from("PocketBase unavailable"))
-                .unwrap()
+            (StatusCode::BAD_GATEWAY, "PocketBase unavailable").into_response()
         }
     }
 }
diff --git a/server-rs/src/routes/postcode_stats.rs b/server-rs/src/routes/postcode_stats.rs
index b51d172..dffc0d9 100644
--- a/server-rs/src/routes/postcode_stats.rs
+++ b/server-rs/src/routes/postcode_stats.rs
@@ -184,12 +184,19 @@ pub async fn get_postcode_stats(
             "GET /api/postcode-stats"
         );
 
+        let crime_latest_year = if crime_by_year.is_empty() {
+            None
+        } else {
+            stats::crime_latest_available_year(&state.crime_by_year)
+        };
+
         Ok(HexagonStatsResponse {
             count: total_count,
             numeric_features,
             enum_features: enum_features_out,
             price_history,
             crime_by_year,
+            crime_latest_year,
             central_postcode: None,
             filter_exclusions,
         })
diff --git a/server-rs/src/routes/stats.rs b/server-rs/src/routes/stats.rs
index 1a349a9..6b439c7 100644
--- a/server-rs/src/routes/stats.rs
+++ b/server-rs/src/routes/stats.rs
@@ -340,16 +340,14 @@ pub fn compute_crime_by_year(
         let points: Vec<CrimeYearPoint> = years
             .iter()
             .filter_map(|&year| {
-                let denom = fully_covered_rows
-                    + covered_counts.get(&year).copied().unwrap_or(0);
+                let denom = fully_covered_rows + covered_counts.get(&year).copied().unwrap_or(0);
                 if denom == 0 {
                     // No selected postcode has published data for this year.
                     return None;
                 }
                 Some(CrimeYearPoint {
                     year,
-                    count: (sums.get(&year).copied().unwrap_or(0.0) / denom as f64)
-                        as f32,
+                    count: (sums.get(&year).copied().unwrap_or(0.0) / denom as f64) as f32,
                 })
             })
             .collect();
@@ -365,6 +363,19 @@ pub fn compute_crime_by_year(
     out
 }
 
+/// Latest year present anywhere in the by-year crime dataset. The client
+/// compares each selection's last charted year against this to caption
+/// force-level publication gaps (e.g. Greater Manchester ends mid-2019) as
+/// stale data instead of letting old numbers read as current.
+pub fn crime_latest_available_year(crime_by_year: &CrimeByYearData) -> Option<i32> {
+    crime_by_year
+        .years_by_type
+        .iter()
+        .flatten()
+        .copied()
+        .max()
+}
+
 pub fn compute_poi_feature_stats(
     matching_rows: &[usize],
     poi_metrics: &PostcodePoiMetrics,
diff --git a/server-rs/src/state.rs b/server-rs/src/state.rs
index edd8ebb..1109ad8 100644
--- a/server-rs/src/state.rs
+++ b/server-rs/src/state.rs
@@ -87,6 +87,105 @@ pub struct AppState {
     pub bugsink_frontend_config: Option<BugsinkFrontendConfig>,
 }
 
+#[cfg(test)]
+impl AppState {
+    /// Minimal AppState for integration tests of the PocketBase/Stripe money
+    /// paths (checkout, webhook, licensing, invites). All map/property data is
+    /// empty; only the HTTP-facing config (PocketBase URL, Stripe secrets,
+    /// caches) carries meaningful values.
+    pub(crate) fn for_tests(pocketbase_url: String) -> Self {
+        use std::time::Duration;
+
+        use crate::data::{
+            ActualListingData, CrimeByYearData, OutcodeData, POIData, PlaceData, PostcodeData,
+            PropertyData, TravelTimeStore,
+        };
+        use crate::utils::InternedColumn;
+
+        let http_client = reqwest::Client::builder()
+            .timeout(Duration::from_secs(5))
+            .connect_timeout(Duration::from_secs(2))
+            .build()
+            .expect("test HTTP client should build");
+
+        AppState {
+            data: PropertyData::empty_for_tests(),
+            grid: GridIndex::build(&[], &[], 0.01),
+            h3_cells: Vec::new(),
+            feature_name_to_index: FxHashMap::default(),
+            min_keys: Vec::new(),
+            max_keys: Vec::new(),
+            avg_keys: Vec::new(),
+            features_response: FeaturesResponse { groups: Vec::new() },
+            ai_filters_system_prompt: String::new(),
+            poi_data: Arc::new(POIData::empty_for_tests()),
+            poi_grid: Arc::new(GridIndex::build(&[], &[], 0.01)),
+            place_data: Arc::new(PlaceData::empty_for_tests()),
+            postcode_data: Arc::new(PostcodeData {
+                postcodes: Vec::new(),
+                centroids: Vec::new(),
+                aabbs: Vec::new(),
+                polygons: Vec::new(),
+                postcode_to_idx: FxHashMap::default(),
+            }),
+            outcode_data: Arc::new(OutcodeData {
+                names: Vec::new(),
+                name_lower: Vec::new(),
+                centroids: Vec::new(),
+                cities: Vec::new(),
+            }),
+            poi_category_groups: Arc::new(Vec::new()),
+            travel_time_store: Arc::new(TravelTimeStore::empty_for_tests()),
+            actual_listings: Arc::new(ActualListingData {
+                lat: Vec::new(),
+                lon: Vec::new(),
+                postcode: Vec::new(),
+                address: Vec::new(),
+                property_type: InternedColumn::build(&[]),
+                property_sub_type: InternedColumn::build(&[]),
+                leasehold_freehold: InternedColumn::build(&[]),
+                price_qualifier: InternedColumn::build(&[]),
+                bedrooms: Vec::new(),
+                bathrooms: Vec::new(),
+                rooms_total: Vec::new(),
+                floor_area_sqm: Vec::new(),
+                asking_price: Vec::new(),
+                asking_price_per_sqm: Vec::new(),
+                listing_url: Vec::new(),
+                listing_status: InternedColumn::build(&[]),
+                listing_date_iso: Vec::new(),
+                features: Vec::new(),
+                filter_feature_data: Vec::new(),
+                poi_filter_feature_data: Vec::new(),
+                grid: GridIndex::build(&[], &[], 0.01),
+            }),
+            crime_by_year: Arc::new(CrimeByYearData {
+                crime_types: Vec::new(),
+                years_by_type: Vec::new(),
+                series_by_postcode: FxHashMap::default(),
+                covered_years_by_postcode: FxHashMap::default(),
+            }),
+            token_cache: Arc::new(TokenCache::new()),
+            superuser_token_cache: Arc::new(SuperuserTokenCache::new()),
+            share_cache: Arc::new(ShareBoundsCache::new()),
+            screenshot_url: "http://127.0.0.1:1/screenshot".to_string(),
+            public_url: "https://test.example".to_string(),
+            is_dev: false,
+            http_client,
+            pocketbase_url,
+            pocketbase_admin_email: "admin@test.example".to_string(),
+            pocketbase_admin_password: "test-admin-password".to_string(),
+            gemini_api_key: "test-gemini-key".to_string(),
+            gemini_model: "test-model".to_string(),
+            google_maps_api_key: "test-maps-key".to_string(),
+            stripe_secret_key: "sk_test_dummy".to_string(),
+            stripe_webhook_secret: "whsec_test_secret".to_string(),
+            stripe_referral_coupon_id: "couponTest30".to_string(),
+            bugsink_frontend_config: None,
+        }
+    }
+}
+
 /// Wraps AppState for shared access across route handlers.
 /// Route handlers call `load_state()` to get the current snapshot.
 pub struct SharedState {