..

frontend/src/components/account/AccountPage.tsx

@@ -20,6 +20,7 @@ import { BookmarkIcon } from '../ui/icons/BookmarkIcon';
 import { TrashIcon } from '../ui/icons/TrashIcon';
 import { CloseIcon } from '../ui/icons/CloseIcon';
 import { useLicense } from '../../hooks/useLicense';
+import { useModalA11y } from '../../hooks/useModalA11y';
 
 function PageLayout({ children }: { children: React.ReactNode }) {
   return (
@@ -43,17 +44,43 @@ function DeleteDialog({
   onConfirm: () => void;
 }) {
   const { t } = useTranslation();
+  const dialogRef = useModalA11y();
+
+  useEffect(() => {
+    const handleKeyDown = (e: KeyboardEvent) => {
+      if (e.key === 'Escape') onCancel();
+    };
+    document.addEventListener('keydown', handleKeyDown);
+    return () => document.removeEventListener('keydown', handleKeyDown);
+  }, [onCancel]);
+
   return (
-    <div className="fixed inset-0 z-50 flex items-center justify-center" onClick={onCancel}>
-      <div className="absolute inset-0 bg-black/50 dark:bg-black/70" />
     <div
-        className="relative w-full max-w-sm mx-4 bg-white dark:bg-warm-800 rounded-lg shadow-xl border border-warm-200 dark:border-warm-700"
+      className="fixed inset-0 z-50 flex items-center justify-center"
+      onClick={onCancel}
+      role="presentation"
+    >
+      <div className="absolute inset-0 bg-black/50 dark:bg-black/70" aria-hidden="true" />
+      <div
+        ref={dialogRef}
+        role="dialog"
+        aria-modal="true"
+        aria-labelledby="delete-search-dialog-title"
+        tabIndex={-1}
+        className="relative w-full max-w-sm mx-4 bg-white dark:bg-warm-800 rounded-lg shadow-xl border border-warm-200 dark:border-warm-700 outline-none"
         onClick={(e) => e.stopPropagation()}
       >
         <div className="flex items-center justify-between px-5 pt-5 pb-3">
-          <h2 className="text-lg font-semibold text-navy-950 dark:text-white">{title}</h2>
+          <h2
+            id="delete-search-dialog-title"
+            className="text-lg font-semibold text-navy-950 dark:text-white"
+          >
+            {title}
+          </h2>
           <button
+            type="button"
             onClick={onCancel}
+            aria-label={t('common.close')}
             className="text-warm-400 hover:text-warm-700 dark:text-warm-400 dark:hover:text-warm-200"
           >
             <CloseIcon className="w-5 h-5" />
@@ -62,12 +89,14 @@ function DeleteDialog({
         <p className="px-5 pb-4 text-sm text-warm-700 dark:text-warm-300">{message}</p>
         <div className="flex gap-3 justify-end px-5 pb-5">
           <button
+            type="button"
             onClick={onCancel}
             className="px-4 py-2 text-sm rounded border border-warm-200 dark:border-warm-700 text-warm-700 dark:text-warm-300 hover:bg-warm-50 dark:hover:bg-warm-700"
           >
             {t('common.cancel')}
           </button>
           <button
+            type="button"
             onClick={onConfirm}
             className="px-4 py-2 text-sm rounded bg-red-600 text-white font-medium hover:bg-red-700"
           >
@@ -124,6 +153,7 @@ function NotesInput({ value, onSave }: { value: string; onSave: (notes: string)
       value={text}
       onChange={handleChange}
       onBlur={handleBlur}
+      aria-label={t('savedPage.notesPlaceholder')}
       placeholder={t('savedPage.notesPlaceholder')}
       rows={1}
       className="w-full resize-none overflow-hidden rounded border border-warm-200 dark:border-warm-700 bg-warm-50 dark:bg-warm-900 px-3 py-1.5 text-sm text-warm-700 dark:text-warm-300 placeholder-warm-400 dark:placeholder-warm-500 focus:outline-none focus:ring-1 focus:ring-teal-400"
@@ -160,6 +190,7 @@ function EditableName({ value, onSave }: { value: string; onSave: (name: string)
       <input
         ref={inputRef}
         value={text}
+        aria-label={t('savedPage.clickToRename')}
         onChange={(e) => setText(e.target.value)}
         onKeyDown={(e) => {
           if (e.key === 'Enter') commit();
@@ -175,12 +206,15 @@ function EditableName({ value, onSave }: { value: string; onSave: (name: string)
   }
 
   return (
-    <h3
+    <h3 className="font-medium text-navy-950 dark:text-warm-100 truncate">
+      <button
+        type="button"
         onClick={() => setEditing(true)}
-      className="font-medium text-navy-950 dark:text-warm-100 truncate cursor-pointer hover:text-teal-600 dark:hover:text-teal-400 border-b border-dotted border-transparent hover:border-warm-400 dark:hover:border-warm-500"
         title={t('savedPage.clickToRename')}
+        className="w-full truncate text-left cursor-pointer hover:text-teal-600 dark:hover:text-teal-400 border-b border-dotted border-transparent hover:border-warm-400 dark:hover:border-warm-500 focus:outline-none focus:ring-1 focus:ring-teal-400 rounded"
       >
         {value}
+      </button>
     </h3>
   );
 }
@@ -704,7 +738,7 @@ function InviteSection({ user }: { user: AuthUser }) {
         return next;
       });
     } catch {
-      // Silent — non-critical
+      // Silent, non-critical
     } finally {
       setInviteHistoryLoading(false);
     }
@@ -853,6 +887,14 @@ export default function AccountPage({
   const { t } = useTranslation();
   const [newsletterSaving, setNewsletterSaving] = useState(false);
   const [newsletterError, setNewsletterError] = useState<string | null>(null);
+  // Mirror the loaded account record's value locally so the checkbox always
+  // reflects the stored value and updates instantly on toggle, instead of the
+  // possibly-stale global auth-store value the async refresh hasn't reconciled
+  // yet (the source of the "checked while stored value is false" flicker).
+  const [newsletterChecked, setNewsletterChecked] = useState(user.newsletter);
+  useEffect(() => {
+    setNewsletterChecked(user.newsletter);
+  }, [user.newsletter]);
   const { startCheckout, checkingOut, error: checkoutError } = useLicense();
   const isLicensed = user.subscription === 'licensed' || user.isAdmin;
 
@@ -916,10 +958,11 @@ export default function AccountPage({
             <label className="flex items-center gap-3 cursor-pointer">
               <input
                 type="checkbox"
-                checked={user.newsletter}
+                checked={newsletterChecked}
                 disabled={newsletterSaving}
                 onChange={async (e) => {
                   const checked = e.target.checked;
+                  setNewsletterChecked(checked); // optimistic; reconciled by onRefreshAuth
                   setNewsletterSaving(true);
                   setNewsletterError(null);
                   try {
@@ -933,6 +976,7 @@ export default function AccountPage({
                     assertOk(res, 'Update newsletter');
                     await onRefreshAuth();
                   } catch (err) {
+                    setNewsletterChecked(!checked); // revert the optimistic toggle on failure
                     const msg =
                       err instanceof Error ? err.message : t('accountPage.updateNewsletterError');
                     setNewsletterError(msg);

frontend/src/components/invite/InvitePage.test.tsx

@@ -0,0 +1,63 @@
+// @vitest-environment jsdom
+import { cleanup, render, screen } from '@testing-library/react';
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import InvitePage from './InvitePage';
+
+vi.mock('react-i18next', async (importOriginal) => ({
+  ...(await importOriginal<typeof import('react-i18next')>()),
+  useTranslation: () => ({ t: (key: string) => key, i18n: { language: 'en' } }),
+}));
+
+function mockFetch(invite: { ok: boolean; status: number; body?: unknown }) {
+  return vi.fn((input: RequestInfo | URL) => {
+    const url = String(input);
+    if (url.includes('/api/invite/')) {
+      return Promise.resolve({
+        ok: invite.ok,
+        status: invite.status,
+        json: () => Promise.resolve(invite.body ?? {}),
+      } as Response);
+    }
+    return Promise.resolve({
+      ok: true,
+      status: 200,
+      json: () => Promise.resolve({ current_price_pence: 4999 }),
+    } as Response);
+  });
+}
+
+const baseProps = {
+  user: null,
+  theme: 'light' as const,
+  onLoginClick: vi.fn(),
+  onRegisterClick: vi.fn(),
+  onLicenseGranted: vi.fn(),
+};
+
+afterEach(() => {
+  cleanup();
+  vi.unstubAllGlobals();
+  vi.restoreAllMocks();
+});
+
+describe('InvitePage invalid-invite messaging', () => {
+  it('shows the unified copy for an unknown code (200 valid:false)', async () => {
+    vi.stubGlobal(
+      'fetch',
+      mockFetch({
+        ok: true,
+        status: 200,
+        body: { valid: false, invite_type: '', used: false, invited_by: null },
+      })
+    );
+    render(<InvitePage code="BOGUS" {...baseProps} />);
+    expect(await screen.findByText('invitePage.invalidInviteLink')).toBeTruthy();
+  });
+
+  it('shows the SAME copy for a malformed code rejected with 400', async () => {
+    vi.stubGlobal('fetch', mockFetch({ ok: false, status: 400 }));
+    render(<InvitePage code="%20" {...baseProps} />);
+    expect(await screen.findByText('invitePage.invalidInviteLink')).toBeTruthy();
+    expect(screen.queryByText('invitePage.invalidInvite')).toBeNull();
+  });
+});

frontend/src/components/invite/InvitePage.tsx

@@ -115,7 +115,20 @@ export default function InvitePage({
           fetch(apiUrl(`invite/${encodeURIComponent(code)}`)),
           fetch(apiUrl('pricing')),
         ]);
-        if (!inviteRes.ok) throw new Error('Failed to validate invite');
+        if (!inviteRes.ok) {
+          // A 4xx means the code itself is malformed or unknown, e.g. /invite/%20
+          // is rejected with 400 while /invite/BOGUS returns 200 {valid:false}.
+          // Render the SAME 'invalid invite link' copy for both so messaging is
+          // consistent; only 5xx / network errors fall through to the transient
+          // error state below.
+          if (inviteRes.status >= 400 && inviteRes.status < 500) {
+            if (!cancelled) {
+              setInvite({ valid: false, invite_type: '', used: false, invited_by: null });
+            }
+            return;
+          }
+          throw new Error('Failed to validate invite');
+        }
         const data: InviteInfo = await inviteRes.json();
         if (!cancelled) setInvite(data);
         if (pricingRes.ok) {
@@ -230,7 +243,9 @@ export default function InvitePage({
       <div className="flex-1 flex items-center justify-center bg-gradient-to-b from-navy-950 via-navy-900 to-navy-900 relative overflow-hidden">
         <HexCanvas isDark={isDark} />
         <div className="text-center relative z-10">
-          <p className="text-lg font-medium text-white mb-2">{t('invitePage.invalidInvite')}</p>
+          <p className="text-lg font-medium text-white mb-2">
+            {t('invitePage.couldNotValidateTitle')}
+          </p>
           <p className="text-warm-400">{error}</p>
         </div>
       </div>

frontend/src/components/landing/SeoContentPage.tsx

@@ -13,7 +13,6 @@ import {
 } from '../../lib/seoLandingPages';
 import { safeJsonLd } from '../../lib/json-ld';
 
-const PUBLIC_URL = 'https://perfect-postcode.co.uk';
 const ProductShowcase = lazy(() => import('../home/ProductShowcase'));
 
 function ProductShowcaseFallback() {
@@ -102,32 +101,11 @@ export default function SeoContentPage({
 }) {
   const { t, i18n } = useTranslation();
   const page = getLocalizedSeoContentPage(pageKey, i18n.language);
-  const url = `${PUBLIC_URL}${page.path}`;
   usePageMeta(page.metaTitle, page.metaDescription);
 
   return (
     <main className="flex-1 overflow-y-auto bg-warm-50 text-navy-950 dark:bg-navy-950 dark:text-warm-100">
       <FaqJsonLd faq={page.faq} />
-      <JsonLd
-        data={{
-          '@context': 'https://schema.org',
-          '@type': 'BreadcrumbList',
-          itemListElement: [
-            {
-              '@type': 'ListItem',
-              position: 1,
-              name: 'Perfect Postcode',
-              item: `${PUBLIC_URL}/`,
-            },
-            {
-              '@type': 'ListItem',
-              position: 2,
-              name: page.title,
-              item: url,
-            },
-          ],
-        }}
-      />
 
       <section className="bg-navy-950 text-white">
         <div className="mx-auto max-w-5xl px-6 py-16 md:px-10 md:py-20">

frontend/src/components/map/HoverCard.tsx

@@ -10,6 +10,7 @@ import { getElectionVoteShareFeatureName } from '../../lib/election-filter';
 import { getEthnicityFeatureName } from '../../lib/ethnicity-filter';
 import { getQualificationFeatureName } from '../../lib/qualification-filter';
 import { getTenureFeatureName } from '../../lib/tenure-filter';
+import { getCouncilFeatureName } from '../../lib/council-filter';
 import {
   POI_DISTANCE_FILTER_NAME,
   getPoiDistanceFeatureName,
@@ -60,6 +61,7 @@ export default memo(function HoverCard({
       const ethnicityFeatureName = getEthnicityFeatureName(name);
       const qualificationFeatureName = getQualificationFeatureName(name);
       const tenureFeatureName = getTenureFeatureName(name);
+      const councilFeatureName = getCouncilFeatureName(name);
       const poiDistanceFeatureName = getPoiDistanceFeatureName(name);
       const backendName =
         schoolBackendName ??
@@ -69,6 +71,7 @@ export default memo(function HoverCard({
         ethnicityFeatureName ??
         qualificationFeatureName ??
         tenureFeatureName ??
+        councilFeatureName ??
         poiDistanceFeatureName ??
         name;
       const val = data[`avg_${backendName}`] ?? data[`min_${backendName}`];

frontend/src/components/map/LocationSearch.tsx

@@ -231,7 +231,7 @@ export default function LocationSearch({
         return;
       }
 
-      // Postcode — fetch geometry. Unlike place/address results, a postcode
+      // Postcode: fetch geometry. Unlike place/address results, a postcode
       // result carries no coordinates, so the camera move genuinely depends on
       // this response and stays gated by isCurrentLookup.
       try {

frontend/src/components/map/MapErrorBoundary.tsx

@@ -9,11 +9,11 @@ import { MapFallback } from './map-page/Fallbacks';
  *
  * deck.gl runs in interleaved mode, sharing MapLibre's WebGL context. When that
  * shared context gets into a bad state (lost context, a buffer/texture finalized
- * while still referenced — the "bufferSubData: no buffer" / "bindTexture: deleted
+ * while still referenced, the "bufferSubData: no buffer" / "bindTexture: deleted
  * object" storm), the next interleaved draw can throw. deck.gl's own animation
  * loop swallows render errors via its `onError` prop, but interleaved draws fire
- * from MapLibre's synchronous `render` event, which can run inside a React commit
+ * from MapLibre's synchronous `render` event, which can run inside a React commit,
- * — so the throw bubbles to the nearest React error boundary.
+ * so the throw bubbles to the nearest React error boundary.
  *
  * Without this, that throw reaches the single top-level boundary and blanks the
  * entire app ("Something went wrong / Refresh the page"). This boundary contains
@@ -26,7 +26,7 @@ const MAX_AUTO_RECOVERIES = 2;
 // Remount after a short delay so a transient GL state can settle first.
 const RECOVERY_DELAY_MS = 400;
 // Crashes spaced further apart than this are treated as independent, so the
-// auto-recovery budget resets — only a tight crash loop escalates to manual retry.
+// auto-recovery budget resets. Only a tight crash loop escalates to manual retry.
 const STABILITY_WINDOW_MS = 30_000;
 
 interface MapErrorBoundaryProps {

frontend/src/components/map/MobileDrawer.tsx

@@ -113,7 +113,7 @@ export default function MobileDrawer({
     >
       <div className="h-[10%] shrink-0" aria-hidden="true" />
 
-      {/* Panel — bottom 90% */}
+      {/* Panel: bottom 90% */}
       <div
         ref={panelRef}
         className="pointer-events-auto h-[90%] bg-white dark:bg-navy-950 rounded-t-2xl flex flex-col border-t border-x border-warm-300 ring-1 ring-navy-950/10 shadow-2xl shadow-navy-950/45 dark:border-navy-600 dark:ring-white/10 dark:shadow-black/60 overflow-hidden"

frontend/src/components/map/NumberLine.test.ts

@@ -40,8 +40,8 @@ describe('computeNumberLineLayout', () => {
       300,
       fmt
     )!;
-    const low = layout.items.find((i) => i.kind === 'national')!; // 30 — lowest
+    const low = layout.items.find((i) => i.kind === 'national')!; // 30: lowest
-    const high = layout.items.find((i) => i.kind === 'area')!; // 50 — highest
+    const high = layout.items.find((i) => i.kind === 'area')!; // 50: highest
     expect(low.tickX).toBeCloseTo(layout.plotLeft, 5);
     expect(high.tickX).toBeCloseTo(layout.plotRight, 5);
   });

frontend/src/components/map/NumberLine.tsx

@@ -120,7 +120,7 @@ export function computeNumberLineLayout(
  * spanning the lowest→highest value, so the selection can be read against its
  * reference points (national / outcode / sector) at a glance. Labels are spread
  * to avoid overlap
- * — common since the nested area/sector/outcode values cluster — and connected
+ * (common since the nested area/sector/outcode values cluster) and connected
  * back to their tick with a thin leader line.
  */
 export default function NumberLine({ points, format }: NumberLineProps) {

frontend/src/components/map/map-page/derivedState.ts

@@ -11,6 +11,7 @@ import { getElectionVoteShareFeatureName } from '../../../lib/election-filter';
 import { getEthnicityFeatureName } from '../../../lib/ethnicity-filter';
 import { getQualificationFeatureName } from '../../../lib/qualification-filter';
 import { getTenureFeatureName } from '../../../lib/tenure-filter';
+import { getCouncilFeatureName } from '../../../lib/council-filter';
 import { getPoiDistanceFeatureName } from '../../../lib/poi-distance-filter';
 import { getSchoolBackendFeatureName } from '../../../lib/school-filter';
 
@@ -31,6 +32,7 @@ export function getMapPageBackendFeatureName(featureName: string): string {
     getEthnicityFeatureName(featureName) ??
     getQualificationFeatureName(featureName) ??
     getTenureFeatureName(featureName) ??
+    getCouncilFeatureName(featureName) ??
     getPoiDistanceFeatureName(featureName) ??
     featureName
   );

frontend/src/components/map/map-page/lazyComponents.ts

@@ -4,6 +4,7 @@ export const Map = lazy(() => import('../Map'));
 export const Filters = lazy(() => import('../Filters'));
 export const POIPane = lazy(() => import('../POIPane'));
 export const OverlayPane = lazy(() => import('../OverlayPane'));
+export const ListingPane = lazy(() => import('../ListingPane'));
 export const AreaPane = lazy(() => import('../AreaPane'));
 export const PropertiesPane = lazy(() =>
   import('../PropertiesPane').then((module) => ({ default: module.PropertiesPane }))

pipeline/check_school_cutoffs.py

@@ -107,7 +107,7 @@ def match_schools(truth: pl.DataFrame, gias: pl.DataFrame) -> pl.DataFrame:
         .alias("_pc"),
     ).with_row_index("_row_id")
 
-    # 1. Exact postcode match (unique postcodes only — site-sharing schools
+    # 1. Exact postcode match (unique postcodes only: site-sharing schools
     #    would mismatch phases otherwise; those fall through to name matching).
     pc_unique = gias.filter(pl.col("_pc").is_not_null()).unique(
         subset="_pc", keep="none"
@@ -291,7 +291,7 @@ def main() -> None:
         print(f"\nWrote matched rows to {args.matched_out}")
 
     if binding.is_empty():
-        raise SystemExit("No binding, matchable cutoffs — nothing to calibrate on")
+        raise SystemExit("No binding, matchable cutoffs: nothing to calibrate on")
 
 
 if __name__ == "__main__":

pipeline/check_travel_times.py

@@ -5,11 +5,11 @@ computation failed or was interrupted, leaving either zero rows or only
 the origin postcode.  We detect this by an absolute, structural criterion:
 a file is corrupt only when it is unreadable or has a row count at or below
 CORRUPT_ROW_FLOOR.  Per-mode percentile/median/range figures are reported
-for context only — they never drive the deletable set, so repeated runs
+for context only. They never drive the deletable set, so repeated runs
 (including with --delete) are idempotent and never erode legitimate
 small-catchment (rural/island) origins.
 
-Duplicates arise when places.parquet is rebuilt between R5 runs — each
+Duplicates arise when places.parquet is rebuilt between R5 runs. Each
 place gets a new numeric index prefix, so the skip-completed logic
 doesn't recognize previous results.  --dedup keeps only the largest
 file per slug and removes the rest.
@@ -105,7 +105,7 @@ def find_bad_files(base_dir: Path) -> tuple[list[BadFile], dict[str, dict]]:
         if not row_counts:
             continue
 
-        # Reporting statistics only — these never decide what gets deleted.
+        # Reporting statistics only; these never decide what gets deleted.
         p5 = percentile(row_counts, 5)
         median = percentile(row_counts, 50)
 

pipeline/download/broadband.py

@@ -16,7 +16,7 @@ PERFORMANCE_URL = "https://www.ofcom.org.uk/siteassets/resources/documents/resea
 
 # Pre-staged file path. Ofcom put the entire ofcom.org.uk domain behind
 # Cloudflare's Managed Challenge in 2026, which requires a JS-executing
-# browser to pass — no amount of User-Agent / TLS-impersonation spoofing
+# browser to pass. No amount of User-Agent / TLS-impersonation spoofing
 # (curl_cffi chrome120..131, safari17, firefox133, chrome_android) gets
 # past it. When the automated download fails, the user must download the
 # zip manually from the Source URL above and place it at this path.

pipeline/download/census_population.py

@@ -63,7 +63,7 @@ def _fetch_csv(
     """Download a CSV, defending against silent truncation.
 
     The NOMIS bulk files are served with chunked transfer encoding and no
-    Content-Length, so a dropped connection ends the stream without raising —
+    Content-Length, so a dropped connection ends the stream without raising,
     yielding a short file. We retry until the parsed row count reaches the
     file's known approximate size, keeping the largest parse seen.
     """
@@ -72,7 +72,7 @@ def _fetch_csv(
         try:
             download(url, dest)
             df = pl.read_csv(dest)
-        except Exception as exc:  # noqa: BLE001 — retry any transport/parse error
+        except Exception as exc:  # noqa: BLE001: retry any transport/parse error
             print(f"  {url} attempt {attempt}: error {exc}")
             time.sleep(3)
             continue

pipeline/download/crime.py

@@ -193,7 +193,7 @@ def select_coverage_archives(
     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
+    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.
     """

pipeline/download/development_sites.py

@@ -143,8 +143,8 @@ def _has_dwellings(min_d: int | None, max_d: int | None) -> bool:
 def _brownfield_url(entity) -> str | None:
     """Canonical per-site page on the Planning Data platform.
 
-    The register's own ``site-plan-url`` is unreliable — many LPAs point every
+    The register's own ``site-plan-url`` is unreliable (many LPAs point every
-    row at a single generic register landing page — so we link to the stable,
+    row at a single generic register landing page), so we link to the stable,
     always-per-site entity page instead.
     """
     entity_id = _to_int(entity)

pipeline/download/education.py

@@ -10,13 +10,13 @@ rather than a single headline number.
 We give the ONS bands colloquial labels (the census's "Level 1/2/3/4+" jargon
 means little to a homebuyer): No qualifications / Some GCSEs / Good GCSEs /
 Apprenticeship / A-levels / Degree or higher / Other qualifications. NOTE the
-census does NOT split undergraduate from postgraduate — "Level 4 and above" is a
+census does NOT split undergraduate from postgraduate. "Level 4 and above" is a
 single bucket ("Degree or higher").
 
 The join key downstream (merge.py) is `lsoa21`, the same key used for ethnicity,
 median age, and IoD.
 
-Source: NOMIS (ONS Census 2021 — TS067 dataset, NM_2084_1)
+Source: NOMIS (ONS Census 2021, TS067 dataset, NM_2084_1)
 License: Open Government Licence v3.0
 """
 
@@ -42,7 +42,7 @@ BASE_URL = (
 )
 
 # Map each canonical NOMIS C2021_HIQUAL_8 band to our colloquial output bucket.
-# 1:1 mapping (no folding) — keyed on the exact NOMIS label so a relabelled or
+# 1:1 mapping (no folding): keyed on the exact NOMIS label so a relabelled or
 # missing band fails loudly in validation instead of silently dropping people.
 BAND_MAP = {
     "No qualifications": "No qualifications",

pipeline/download/election_results.py

@@ -5,7 +5,7 @@ import httpx
 import polars as pl
 
 # UK Parliament publishes candidate-level results for the 2024 General Election.
-# One row per candidate per constituency — we aggregate to per-constituency stats.
+# One row per candidate per constituency. We aggregate to per-constituency stats.
 URL = "https://electionresults.parliament.uk/general-elections/6/candidacies.csv"
 
 # Map party names to a smaller set for vote share columns.

pipeline/download/england_boundary.py

@@ -9,7 +9,7 @@ from pathlib import Path
 
 import httpx
 
-# ArcGIS REST API — query for England only, generalised (BGC) resolution
+# ArcGIS REST API: query for England only, generalised (BGC) resolution
 URL = (
     "https://services1.arcgis.com/ESMARspQHYMw9BZ9/arcgis/rest/services/"
     "Countries_December_2024_Boundaries_UK_BGC/FeatureServer/0/query"

pipeline/download/ethnicity.py

@@ -11,7 +11,7 @@ different neighbourhoods in one borough no longer share an identical ethnicity
 profile. The join key downstream (merge.py) is `lsoa21`, the same key already
 used for median age and IoD.
 
-Source: NOMIS (ONS Census 2021 — TS021 dataset, NM_2041_1)
+Source: NOMIS (ONS Census 2021, TS021 dataset, NM_2041_1)
 License: Open Government Licence v3.0
 """
 

pipeline/download/gias.py

@@ -4,12 +4,12 @@ GIAS is the DfE register of all educational establishments in England, updated
 daily. The CSV is generated on-demand via a four-step interaction with the
 public Downloads page (there is no static URL):
 
-1.  GET /Downloads — extract anti-forgery token, the `all.edubase.data` tag,
+1.  GET /Downloads: extract anti-forgery token, the `all.edubase.data` tag,
     and the FileGeneratedDate that the server expects for that tag today.
-2.  POST /Downloads/Collate — submit the form to start file generation. The
+2.  POST /Downloads/Collate: submit the form to start file generation. The
     redirect URL contains a generation UUID.
 3.  Poll /Downloads/GenerateAjax/{id} until status:true.
-4.  GET the Azure blob URL with ?id={id} — returns a ZIP containing
+4.  GET the Azure blob URL with ?id={id}: returns a ZIP containing
     `edubasealldataYYYYMMDD.csv`.
 
 The CSV is cp1252-encoded with 135 columns. We keep the fields useful for a
@@ -262,7 +262,7 @@ def transform(zip_bytes: bytes) -> pl.DataFrame:
         .cast(pl.Float32, strict=False),
     )
 
-    # Drop rows without coordinates — a small number of historic/dummy entries
+    # Drop rows without coordinates: a small number of historic/dummy entries
     # have Easting=0 which would map to the Atlantic.
     df = df.filter(
         pl.col("Easting").is_not_null()

pipeline/download/greenspace_water.py

@@ -21,7 +21,7 @@ from tqdm import tqdm
 logger = logging.getLogger(__name__)
 
 
-MIN_AREA_SQM = 5_000  # ~70m x 70m — skip pocket parks and small ponds
+MIN_AREA_SQM = 5_000  # ~70m x 70m: skip pocket parks and small ponds
 
 # OSM tags that indicate non-residential green/water areas
 GREENSPACE_TAGS = {
@@ -108,7 +108,7 @@ class GreenspaceHandler(osmium.SimpleHandler):
             return
 
         # Invalid geometries are often the largest, most complex park/water
-        # multipolygons (self-touching rings from OSM) — repair like pois.py
+        # 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:
@@ -161,7 +161,7 @@ def main():
         df.write_parquet(args.output)
         print(f"Saved to {args.output}")
     else:
-        print("No geometries found — skipping output")
+        print("No geometries found, skipping output")
 
 
 if __name__ == "__main__":

pipeline/download/lsoa_children.py

@@ -1,6 +1,6 @@
 """Download Census 2021 children by five-year age band per LSOA.
 
-Source: NOMIS (ONS Census 2021 — TS007A dataset, age by five-year bands)
+Source: NOMIS (ONS Census 2021, TS007A dataset, age by five-year bands)
 License: Open Government Licence v3.0
 
 Used to estimate how many primary-age (4-10) and secondary-age (11-15)

pipeline/download/lsoa_population.py

@@ -1,6 +1,6 @@
 """Download Census 2021 usual resident population by LSOA.
 
-Source: NOMIS (ONS Census 2021 — TS001 dataset)
+Source: NOMIS (ONS Census 2021, TS001 dataset)
 License: Open Government Licence v3.0
 """
 

pipeline/download/test_tenure.py

@@ -76,7 +76,7 @@ def test_tenure_lives_rent_free_rolls_into_private_rent():
 
 
 def test_tenure_percentages_independent_per_lsoa():
-    """Two LSOAs get independent profiles — the LSOA granularity is the point."""
+    """Two LSOAs get independent profiles: the LSOA granularity is the point."""
     df = pl.concat(
         [
             pl.DataFrame(_long_rows("E01000010", {"Owned: Owns outright": 100})),

pipeline/download/test_transit_network.py

@@ -76,7 +76,7 @@ def test_clean_national_rail_gtfs_orders_by_stop_sequence_not_file_order(
     """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
+    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"

pipeline/transform/area_crime_averages.py

@@ -5,7 +5,7 @@ The right pane shows each crime metric next to its area context: the mean
 average-annual count (``"X (/yr, 7y)"``) across the selection's postcode sector (e.g.
 ``"E14 2"``), its outcode (e.g. ``"E14"``), and the nation. Crime is constant
 within a postcode (the merge keys it on the postcode), so each postcode
-contributes its single value weighted by how many properties sit in it — keeping
+contributes its single value weighted by how many properties sit in it, keeping
 every scope on the same property-weighted basis as the per-selection mean, so the
 four numbers (this selection / sector / outcode / nation) are directly
 comparable. The national figure here is an EXACT property-weighted mean, which is
@@ -18,7 +18,7 @@ crime values from ``postcode.parquet`` and the per-postcode property weights fro
 ``properties.parquet`` mirrors exactly the two inputs the server loads, so the
 result matches what the server used to compute (minus its u16 quantization loss).
 
-Output schema — one row per area:
+Output schema, one row per area:
 
     scope            : ``"national"`` | ``"outcode"`` | ``"sector"``
     area             : the outcode (``"E14"``) / sector (``"E14 2"``);
@@ -43,7 +43,7 @@ SCOPE_NATIONAL = "national"
 SCOPE_OUTCODE = "outcode"
 SCOPE_SECTOR = "sector"
 
-# Area label on the national row — it spans the whole country, so it has no code.
+# Area label on the national row. It spans the whole country, so it has no code.
 NATIONAL_AREA = ""
 
 # Both merge outputs key on the canonical NSPL `pcds` postcode (spaced, e.g.
@@ -71,7 +71,7 @@ def _weighted_mean(column: str) -> pl.Expr:
 
     A null crime value is a genuine gap (the postcode's police force published no
     usable data), not zero crime, so it must dilute neither the numerator nor the
-    denominator — exactly as the server's former estimator skipped NaN values.
+    denominator, exactly as the server's former estimator skipped NaN values.
     Yields null when no postcode in the group has data for this type.
     """
     weight = pl.col(_WEIGHT_COLUMN)

pipeline/transform/crime.py

@@ -126,7 +126,7 @@ def transform_crime(
         # 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,
+        # 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"))
@@ -194,7 +194,7 @@ def _write_crime_by_year(
     )
 
     yearly_per_type = (
-        # Per-incident weight sum, not `_weight.first() * len` — see the
+        # 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"))

pipeline/transform/crime_spatial.py

@@ -92,7 +92,7 @@ STREET_CSV_NAME_RE = re.compile(r"^(\d{4}-\d{2})-(.+)-street\.csv$")
 
 # Trailing-window definitions, in (label, years) form. Each window's average is
 # pooled over the force's covered months inside the window; one average-annual-
-# count column (`(/yr, <label>)`) — the filterable crime feature — is emitted per
+# count column (`(/yr, <label>)`), the filterable crime feature, is emitted per
 # window.
 WINDOWS: tuple[tuple[str, int], ...] = (("7y", 7), ("2y", 2))
 

pipeline/transform/join_epc_pp.py

@@ -22,20 +22,20 @@ pl.Config.set_tbl_cols(-1)
 
 RATING_RANK = {"A": 1, "B": 2, "C": 3, "D": 4, "E": 5, "F": 6, "G": 7}
 # Value-quality floor for price aggregations. A flat nominal floor is a blunt
-# tool against a deflating threshold — £50k was completely normal for a 1990s
+# tool against a deflating threshold: £50k was completely normal for a 1990s
 # house, so a 50k floor wrongly discarded ~a third of legitimate 1990s
 # open-market sales (and deleted properties whose only sales were old/cheap),
 # biasing early-year price history upward. 10k recovers the large [10k,50k)
 # band of genuine cheaper sales while still excluding the nominal/junk transfers
-# (£1 etc.). A small tail of real sub-10k sales is still dropped — a deliberate
+# (£1 etc.). A small tail of real sub-10k sales is still dropped, a deliberate
 # conservative tradeoff to keep clearly-implausible transfers out.
 MIN_PRICE = 10_000
 
 # Time-aware consecutive-sale jump guard. Price-paid contains keyed-in price
-# errors that pass the MIN_PRICE/category filters — e.g. 13 QUICKSETTS HR2 7PP,
+# errors that pass the MIN_PRICE/category filters and would otherwise become
-# a 93 m² terrace, sold £140,000 in 2016 then "£207,500,000" in 2026 (clearly
+# latest_price. For example, 13 QUICKSETTS HR2 7PP, a 93 m² terrace, sold
-# £207,500 with extra digits, lodged as category A) — and would otherwise
+# £140,000 in 2016 then "£207,500,000" in 2026 (clearly £207,500 with extra
-# become latest_price. A quality sale is flagged when it exceeds its
+# digits, lodged as category A). A quality sale is flagged when it exceeds its
 # neighbouring sale by more than JUMP_TOLERANCE * JUMP_GROWTH_PER_YEAR ** years
 # between the two sales. Calibration: genuine extreme appreciation (prime
 # London 1995->2026 is roughly x50 over 31 years) stays comfortably under
@@ -174,8 +174,8 @@ def _clean_number(column: str, dtype: pl.DataType) -> pl.Expr:
 def _join_address_parts(*columns: str) -> pl.Expr:
     """Join address components into one display address, single-spaced.
 
-    Price-paid SAON/PAON/STREET are EMPTY STRINGS (not null) when absent —
+    Price-paid SAON/PAON/STREET are EMPTY STRINGS (not null) when absent
-    saon is "" on ~88% of rows — and ``concat_str(..., ignore_nulls=True)``
+    (saon is "" on ~88% of rows), and ``concat_str(..., ignore_nulls=True)``
     skips only nulls, so empty components still contributed their separator
     (``' 10 PALACE GREEN'``, doubled spaces when a middle part was empty).
     Convert ``''``→null per component so ignore_nulls works as intended, then
@@ -444,6 +444,13 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
         epc_base.sort("inspection_date", descending=True, nulls_last=True)
         .group_by("_epc_match_address", "_epc_match_postcode")
         .first()
+        # The deduped row carries the most-recent certificate (sorted newest
+        # first, .first() keeps it), so normalising its raw tenure here yields
+        # the LATEST certificate's coarse tenure. Kept past the .drop("tenure")
+        # so downstream (merge) can tell a still-social dwelling from one that
+        # was social once but whose latest cert is no longer social. Computed
+        # before .drop("tenure") because it reads that column.
+        .with_columns(tenure_status(pl.col("tenure")).alias("latest_tenure_status"))
         .drop("tenure")
     )
 
@@ -520,7 +527,7 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
     #
     # Emission rule (walking certificates oldest-first, ignoring unknown-tenure
     # ones so they neither appear nor break the chain):
-    #   - the first known status is emitted only when it is a rental — an
+    #   - the first known status is emitted only when it is a rental: an
     #     owner-occupied baseline is the unremarkable default for a property
     #     that has changed hands and would only clutter the timeline;
     #   - every later certificate is emitted when its status differs from the
@@ -658,8 +665,8 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
         )
         .filter(pl.col("pp_address").is_not_null())
         # Price-paid carries ~72k duplicate (address, postcode, date, price)
-        # transaction groups with DISTINCT transaction ids — the same completed
+        # transaction groups with DISTINCT transaction ids (the same completed
-        # sale lodged twice — which double-counted sales in historical_prices.
+        # sale lodged twice), which double-counted sales in historical_prices.
         # Collapse each to one row. ppd_category stays in the subset so an
         # A/B-categorised pair of the same sale survives as two rows; only the
         # A row feeds the price aggregations (quality_ok), which is intentional.
@@ -714,6 +721,10 @@ def _run(epc_path: Path, price_paid_path: Path, output_path: Path, temp_dir: Pat
                 pl.col("date_of_transfer").dt.year().alias("year"),
                 pl.col("date_of_transfer").dt.month().cast(pl.UInt8).alias("month"),
                 "price",
+                # Per-sale new-build flag straight from the PPD "old_new" field
+                # (Y = newly built on first transfer, N = established). Kept on
+                # each transaction so the timeline can mark the new-build sale.
+                (pl.col("old_new") == "Y").alias("is_new"),
             )
             .filter(quality_ok)
             .alias("historical_prices"),

pipeline/transform/merge.py

@@ -83,7 +83,7 @@ _AREA_COLUMNS = [
     "% Mixed",
     "% White",
     "% Other",
-    # Crime — average annual recorded incident count (incidents/yr), 7-year and
+    # Crime: average annual recorded incident count (incidents/yr), 7-year and
     # 2-year windows. These are the filterable crime features; the per-incident
     # records live in a separate side table the server loads directly (it bypasses
     # the merge).
@@ -145,11 +145,18 @@ _AREA_COLUMNS = [
     "% A-levels",
     "% Degree or higher",
     "% Other qualifications",
-    # Tenure (Census 2021 TS054, % of households by tenure) — unlike ethnicity &
+    # Tenure (Census 2021 TS054, % of households by tenure). Unlike ethnicity &
     # education these three percentages are user-filterable, not display-only.
     "% Owner occupied",
     "% Social rent",
     "% Private rent",
+    # Council housing (EPC-derived, NOT Census): share of dwellings in the
+    # postcode ever recorded as social housing per EPC, and the ever-social
+    # subset whose latest EPC certificate is no longer social rented (sold off).
+    # Aggregated from the per-property "was_council_house" / "latest_tenure_status"
+    # flags in _epc_council_by_postcode and joined onto the AREA frame only.
+    "% Council housing",
+    "% Ex-council",
     # Politics
     "Voter turnout (%)",
     "% Labour",
@@ -254,7 +261,7 @@ def _subset_numbers_compatible(left: str, right: str) -> bool:
     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
+    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))
@@ -574,7 +581,7 @@ def _is_current_planning_record(end_date: object) -> bool:
     """A planning record is current when it has no end-date OR its end-date is
     still in the future. The planning.data.gov.uk `end-date` field marks when a
     designation is RETIRED, so a future date (e.g. 2029-12-31) is a still-current
-    area and must NOT be dropped — the previous "any non-empty date = ended"
+    area and must NOT be dropped. The previous "any non-empty date = ended"
     logic wrongly excluded those (e.g. 22 current Gateshead conservation areas)."""
     if end_date is None:
         return True
@@ -864,7 +871,7 @@ def _join_area_side_tables(
 ) -> pl.LazyFrame:
     base = base.join(iod, left_on="lsoa21", right_on="LSOA code (2021)", how="left")
     # Ethnicity is Census 2021 TS021 at LSOA (~33,755 areas), joined on the same
-    # `lsoa21` key as median age and IoD — a ~100x granularity gain over the old
+    # `lsoa21` key as median age and IoD, a ~100x granularity gain over the old
     # Local-Authority broadcast, with no change to the 6-bucket output schema.
     base = base.join(ethnicity, on="lsoa21", how="left")
     # Education (Census 2021 TS067 "highest level of qualification") is sourced at
@@ -1140,9 +1147,9 @@ def _address_score(query: str, candidate: str | None, *, allow_token_set: bool)
     if not candidate:
         return 0
     # token_set_ratio returns 100 whenever the shorter token set is a subset of
-    # the longer. For a NUMBER-LESS query that is unsafe — a single locality
+    # the longer. For a NUMBER-LESS query that is unsafe: a single locality
     # token (e.g. "KINGSWOOD") subsets to 100 against any long address that
-    # merely contains it — so number-less queries score with token_sort_ratio
+    # 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 fuzzy_join._numbers_compatible gate has already
     # guaranteed the candidate carries identical house numbers, so token_set
@@ -1241,14 +1248,14 @@ 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 — `fuzzy_join._numbers_compatible`
+    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.
     The direct-EPC path layers a street-level fallback on top of this strict
-    matcher — see `_best_street_epc_fallback`.
+    matcher. See `_best_street_epc_fallback`.
 
     ``addressed_fields`` names the candidate columns to fuzzy-match against (a
     candidate may carry both a register and an EPC address). Returns
@@ -1301,7 +1308,7 @@ def _best_listing_match(
 # the listing's own postcode unit is the nearest segment of the street, and a
 # certificate sharing a house-number token with the listing (e.g. listing
 # "751 753 Cranbrook Road" vs certificate "751 Cranbrook Road", which fails the
-# strict set-equality gate) is almost certainly the right property — both
+# strict set-equality gate) is almost certainly the right property. Both
 # should beat a bare attribute-agreement win.
 _STREET_FALLBACK_SAME_POSTCODE_BONUS = 3.0
 _STREET_FALLBACK_NUMBER_OVERLAP_BONUS = 8.0
@@ -1309,8 +1316,8 @@ _STREET_FALLBACK_NUMBER_OVERLAP_BONUS = 8.0
 # is era-homogeneous (a single development). When the same-street certificates
 # span more than this many years between their oldest and newest build, the
 # street mixes construction eras (a Victorian terrace with modern infill, say),
-# so no single year represents the unidentified number-less listing property —
+# so no single year represents the unidentified number-less listing property.
-# the fallback then publishes a null construction year rather than an arbitrary,
+# The fallback then publishes a null construction year rather than an arbitrary,
 # often wrong-by-a-century guess. Two adjacent EPC age bands (one development
 # straddling a band boundary) span at most ~26 representative years, so this
 # threshold keeps genuinely-uniform streets while rejecting mixed ones.
@@ -1328,7 +1335,7 @@ def _street_match_with_reliable_construction_year(
     ``_STREET_FALLBACK_CONSTRUCTION_SPAN_YEARS`` between their oldest and newest
     build, the street mixes construction eras and the matched certificate's year
     (e.g. an 1890 Victorian house on a street with 2007 infill) would otherwise
-    be presented as the listing property's own — so the year and its
+    be presented as the listing property's own, so the year and its
     approximate-date flag are nulled, leaving an honest "unknown" rather than a
     wrong-by-a-century value. Other street-representative EPC facts (energy
     rating, floor area) are inherently per-property approximations the fallback
@@ -1357,7 +1364,7 @@ def _best_street_epc_fallback(
     """Street-level direct-EPC fallback for listings the strict matcher missed.
 
     ~90% of scraped listings publish a street-level address only ("Oldstead
-    Road, Bromley" — Rightmove never exposes the house number or UPRN), so the
+    Road, Bromley", since Rightmove never exposes the house number or UPRN), so the
     strict matcher in `_best_listing_match` can never match them against the
     virtually-always-numbered EPC register and their EPC-derived fields
     (energy rating, interior height, former-council-house flag, construction
@@ -1371,7 +1378,7 @@ def _best_street_epc_fallback(
     same-postcode-unit preference and a house-number-overlap bonus (a
     numbered listing that failed the strict set-equality gate, e.g. a
     "751 753" range vs "751", still lands on the right property). The result
-    is street-representative rather than property-exact — hence the distinct
+    is street-representative rather than property-exact, hence the distinct
     "street" method label so downstream consumers can tell the two confidence
     levels apart. The matched certificate's construction year is kept only when
     the street is era-homogeneous (see
@@ -1439,7 +1446,7 @@ def _best_street_epc_fallback(
             # bathrooms (the upstream storage.py defect noted in
             # `_finalize_listings`). It systematically over-counts, so comparing
             # it to the EPC habitable-room count biases selection toward larger,
-            # typically older certificates — the opposite of a useful signal —
+            # typically older certificates (the opposite of a useful signal),
             # and there is no clean listing-side habitable-room count to use.
             if (
                 listing_postcode
@@ -1472,9 +1479,9 @@ def _load_listings_for_merge(listings_path: Path, arcgis_path: Path) -> pl.DataF
     """Read the listings parquet and prepare it for the wide-frame merge.
 
     Output is keyed by `_listing_idx` and carries:
-      * `postcode` — canonical (NSPL `pcds`) form, with terminated postcodes
+      * `postcode`: canonical (NSPL `pcds`) form, with terminated postcodes
         remapped to their nearest active successor;
-      * `pp_address` — the listing's raw register address (used as the
+      * `pp_address`: the listing's raw register address (used as the
         address half of the fuzzy match);
       * one `_actual_*` overlay column per `_LISTING_OVERLAY_SOURCES` entry.
     """
@@ -1684,7 +1691,7 @@ def _listing_match_frame(listings: pl.DataFrame) -> pl.DataFrame:
 
     Listings are matched to EPC certificates and properties by UPRN and by
     fuzzy street address within their (now accurate, detail-page-sourced)
-    postcode — never by coordinate proximity — so no projected easting/northing
+    postcode (never by coordinate proximity), so no projected easting/northing
     is computed here. `_listing_uprn` flows through from the loaded listings.
     """
     return listings.with_columns(
@@ -1763,8 +1770,8 @@ def _index_candidates(
     The EPC register's UPRN is NOT unique: a single building/parent UPRN fans
     across many distinct flats (up to 58 distinct (address, postcode) rows in
     the 2026-06 data; ~9k UPRNs collide, touching ~20k epc_pp rows). Such a
-    UPRN cannot serve as a 1:1 exact-match key — it would mis-link a listing to
+    UPRN cannot serve as a 1:1 exact-match key (it would mis-link a listing to
-    one arbitrary flat — so any UPRN that resolves to more than one distinct
+    one arbitrary flat), so any UPRN that resolves to more than one distinct
     ``(postcode_key, address_key)`` identity is dropped from ``uprn_index``;
     those listings fall back to the fuzzy street-address matcher, which
     disambiguates the specific flat. A UPRN repeated for the SAME identity
@@ -1878,7 +1885,7 @@ def _index_epc_streets(
     maps outcode -> the tokens appearing in at least a quarter of that
     outcode's street keys. Those are locality suffixes (LONDON, SURREY, the
     town name) rather than street names, and a fallback match must be anchored
-    by at least one token that is NOT one of them — otherwise a town-only
+    by at least one token that is NOT one of them. Otherwise a town-only
     listing address ("COULSDON SURREY") token_set-inflates to 100 against any
     street key carrying the same locality suffix and matches an arbitrary
     street in the outcode.
@@ -2260,7 +2267,7 @@ def _finalize_listings(df: pl.DataFrame) -> pl.DataFrame:
 
 @dataclass
 class _BuildResult:
-    """Outputs of `_build` — exactly one of the two slot pairs is populated."""
+    """Outputs of `_build`: exactly one of the two slot pairs is populated."""
 
     postcode: pl.DataFrame | None = None
     properties: pl.DataFrame | None = None
@@ -2286,6 +2293,38 @@ def _fill_property_level_no_defaults(frame: pl.LazyFrame) -> pl.LazyFrame:
     )
 
 
+def _epc_council_by_postcode(wide: pl.LazyFrame) -> pl.LazyFrame:
+    """Aggregate the per-property EPC social-tenure flags to POSTCODE percentages.
+
+    Two EPC-derived AREA columns, each a share of *all* deduped dwellings in the
+    postcode (one row per dwelling in ``wide``). This denominator is the dwelling
+    universe, NOT Census households: a dwelling with no EPC is ``was_council_house
+    == "No"`` and so counts against the share, making these EPC-coverage-limited
+    lower bounds that are not directly comparable to the Census ``% Social rent``
+    (which stays at LSOA grain). A postcode holds relatively few dwellings, so
+    these shares are coarse and noisier than an LSOA average.
+
+    * ``% Council housing``: dwellings ever recorded as council/social housing
+      per EPC (``was_council_house == "Yes"``).
+    * ``% Ex-council``: ever-council dwellings whose LATEST EPC certificate is no
+      longer social rented (sold off / no longer social):
+      ``was_council_house == "Yes" AND latest_tenure_status != "Rented (social)"``.
+      A null ``latest_tenure_status`` counts as not-currently-social.
+
+    ``was_council_house`` is already "Yes"/"No" filled for every row (see
+    ``_fill_property_level_no_defaults``), so the means are over the full postcode.
+    Returns a postcode-keyed LazyFrame to left-join onto the AREA frame only.
+    """
+    currently_social = (pl.col("latest_tenure_status") == "Rented (social)").fill_null(
+        False
+    )
+    ever_social = pl.col("was_council_house") == "Yes"
+    return wide.group_by("postcode").agg(
+        (ever_social.mean() * 100).round(1).alias("% Council housing"),
+        ((ever_social & ~currently_social).mean() * 100).round(1).alias("% Ex-council"),
+    )
+
+
 def _build(
     epc_pp_path: Path,
     arcgis_path: Path,
@@ -2311,9 +2350,9 @@ def _build(
     """Build postcode/properties dataframes (or enriched listings) from epc_pp + auxiliary data.
 
     Modes:
-      * `normal` — produces (postcode_df, properties_df) as before. Ignores
+      * `normal`: produces (postcode_df, properties_df) as before. Ignores
         `actual_listings_path` if supplied.
-      * `listings` — requires `actual_listings_path`; produces a single
+      * `listings`: requires `actual_listings_path`; produces a single
         enriched-listings DataFrame and skips the postcode/properties outputs.
         Listings flow through the same enrichment joins as historical rows,
         so postcode-scoped features (tree density, crime, deprivation, …) end
@@ -2331,8 +2370,8 @@ def _build(
     )
     _validate_lad_source_coverage(iod_path, rental_prices_path)
 
-    # The dwelling universe — floor filter, terminated-postcode remap,
+    # The dwelling universe (floor filter, terminated-postcode remap,
-    # collapse-dedupe, restrict to active English postcodes — is shared with
+    # collapse-dedupe, restrict to active English postcodes) is shared with
     # price estimation so estimates line up 1:1 with these rows. See
     # pipeline.transform.property_base.
     wide = build_property_base(epc_pp_path, arcgis_path)
@@ -2460,6 +2499,17 @@ def _build(
     wide = _join_area_side_tables(wide, **area_side_tables)
     postcode_area = _join_area_side_tables(postcode_area, **area_side_tables)
 
+    # EPC-derived council/ex-council shares: aggregate the per-property social
+    # tenure flags to POSTCODE percentages and attach to the AREA frame only
+    # (these are area columns, like the Census tenure block, not per-property).
+    # Built before dropping latest_tenure_status, which is its only consumer.
+    epc_council_by_postcode = _epc_council_by_postcode(wide)
+    postcode_area = postcode_area.join(epc_council_by_postcode, on="postcode", how="left")
+    # latest_tenure_status is property-grain and not in _AREA_COLUMNS, so the
+    # split would otherwise leak it into properties.parquet. It has served its
+    # purpose (the postcode aggregate above), so drop it from the property frame.
+    wide = wide.drop("latest_tenure_status", strict=False)
+
     # Derive bedroom count: habitable rooms - 1 (assuming 1 reception room), clipped to 0..4
     wide = wide.with_columns(
         (pl.col("number_habitable_rooms") - 1)
@@ -2649,7 +2699,7 @@ def main():
         required=False,
         help=(
             "Optional scraped-listings parquet. When provided, listings flow "
-            "through the same merge pipeline as historical properties — set "
+            "through the same merge pipeline as historical properties. Set "
             "--output-listings to write the enriched-listings file instead "
             "of the postcode/properties files."
         ),

pipeline/transform/poi_proximity.py

@@ -43,7 +43,7 @@ GROCERY_STATIC_EXCLUDED_CATEGORIES = {
 # Scope: "Public Park Or Garden" is the core park function. "Playing Field"
 # (open public recreation grounds) is borderline but kept: outside big cities
 # the local rec ground is the de facto park. "Play Space" (playgrounds) is
-# excluded — a playground is not a park, and "Playground" is already its own
+# excluded: a playground is not a park, and "Playground" is already its own
 # OSM-derived category. The remaining functions (Religious Grounds, Golf
 # Course, Cemetery, Allotments, Bowling Green, Tennis Court, Other Sports
 # Facility) are clearly not parks.
@@ -76,7 +76,7 @@ def _groceries_categories(pois: pl.DataFrame) -> list[str]:
     with group "Groceries"; it never emits the literal "Supermarket". Collecting
     every Groceries category captures both the OSM strings and the brand names.
     Speciality food retail (bakeries, butchers, delis, off-licences) is
-    excluded — see GROCERY_STATIC_EXCLUDED_CATEGORIES.
+    excluded. See GROCERY_STATIC_EXCLUDED_CATEGORIES.
     """
     if "group" not in pois.columns:
         raise ValueError("POI dataframe must include a 'group' column")
@@ -138,7 +138,7 @@ def _greenspace_count_frame(greenspace: pl.DataFrame) -> pl.DataFrame:
 
     os_greenspace.parquet is one row per ACCESS POINT (park gate), which is the
     right grain for nearest-distance (the nearest gate is what matters) but
-    wildly over-counts "Number of amenities (Park) within Xkm" — a large park
+    wildly over-counts "Number of amenities (Park) within Xkm": a large park
     with 30 gates counted as 30 parks. Counting uses one row per site at the
     site centroid (falling back to the first access point when no centroid is
     available). Degrades gracefully: a legacy parquet without `site_id` is

server-rs/src/main.rs

@@ -125,12 +125,7 @@ async fn static_cache_headers(
     response
 }
 
-/// Add baseline security headers to every response. These are deliberately the
+
-/// low-risk, broadly-compatible set: a Content-Security-Policy and
-/// Permissions-Policy are intentionally left out because this app loads many
-/// cross-origin resources (maplibre/deck.gl, Stripe, Google Street View,
-/// analytics, the error sink) and a mis-scoped policy would break the map or
-/// checkout. They should be added later as report-only first.
 async fn security_headers(
     request: axum::extract::Request,
     next: middleware::Next,

server-rs/src/routes/invites.rs

@@ -1,6 +1,7 @@
 use std::collections::HashSet;
 use std::sync::{Arc, LazyLock, Mutex};
 
+use axum::extract::rejection::JsonRejection;
 use axum::extract::{Path, State};
 use axum::http::StatusCode;
 use axum::response::{IntoResponse, Response};
@@ -153,7 +154,7 @@ fn current_unix_secs_string() -> String {
 
 /// Fetch the live `is_admin` flag for a user, bypassing any cached token
 /// claims. Returns Err with an HTTP response if PocketBase is unreachable
-/// or returns an unexpected payload — the caller should propagate that.
+/// or returns an unexpected payload. The caller should propagate that.
 async fn verify_is_admin(
     state: &AppState,
     pb_url: &str,
@@ -295,7 +296,7 @@ async fn mark_invite_used(
     // Defense in depth: PocketBase has no atomic compare-and-swap for record
     // updates, and our local + distributed locks could in principle fail (lock
     // server timeout, server restart mid-redemption). Re-read the record and
-    // confirm WE actually own it — if a concurrent redemption beat us to the
+    // confirm WE actually own it. If a concurrent redemption beat us to the
     // PATCH, both writes succeeded but the loser's user_id is overwritten and
     // we must NOT grant a license.
     let verify_url = format!("{pb_url}/api/collections/invites/records/{invite_id}");
@@ -328,7 +329,7 @@ async fn mark_invite_used(
             invite_id,
             expected = user_id,
             actual = actual_user,
-            "Invite redemption race lost — invite already claimed by another user"
+            "Invite redemption race lost. Invite already claimed by another user"
         );
         return Err((StatusCode::CONFLICT, "Invite was already redeemed").into_response());
     }
@@ -472,7 +473,7 @@ pub async fn post_invites(
 /// Only recognized when `--dist` is not set (i.e., dev mode).
 const DEV_INVITE_CODE: &str = "devdevdevdev";
 
-/// Validate an invite code. Public endpoint — codes are 12-char random alphanumeric
+/// Validate an invite code. Public endpoint: codes are 12-char random alphanumeric
 /// so enumeration is impractical, and the response only reveals valid/invalid + type.
 pub async fn get_invite(
     State(shared): State<Arc<SharedState>>,
@@ -542,7 +543,7 @@ pub async fn get_invite(
             let used = !used_by.is_empty();
             let created_by = invite["created_by"].as_str().unwrap_or("");
 
-            // Look up inviter's name (email local part) — sanitized before returning.
+            // Look up inviter's name (email local part), sanitized before returning.
             let invited_by = if !created_by.is_empty() {
                 let user_url = format!("{pb_url}/api/collections/users/records/{created_by}");
                 match state
@@ -594,7 +595,7 @@ pub async fn get_invite(
 pub async fn post_redeem_invite(
     State(shared): State<Arc<SharedState>>,
     Extension(user): Extension<OptionalUser>,
-    Json(req): Json<RedeemRequest>,
+    body: Result<Json<RedeemRequest>, JsonRejection>,
 ) -> Response {
     let state = shared.load_state();
     let user = match user.0 {
@@ -602,11 +603,20 @@ pub async fn post_redeem_invite(
         None => return StatusCode::UNAUTHORIZED.into_response(),
     };
 
+    // Parse the body only after authenticating, and return a clean message
+    // instead of leaking the raw serde rejection (e.g. "missing field `code`").
+    let req = match body {
+        Ok(Json(req)) => req,
+        Err(_) => {
+            return (StatusCode::BAD_REQUEST, "A valid invite code is required.").into_response()
+        }
+    };
+
     if let Err(msg) = validate_invite_code(&req.code) {
         return (StatusCode::BAD_REQUEST, msg).into_response();
     }
 
-    // Dev-only: fake redeem — just return "licensed" without touching PocketBase
+    // Dev-only: fake redeem, just return "licensed" without touching PocketBase
     if state.is_dev && req.code == DEV_INVITE_CODE {
         info!(user_id = %user.id, "Dev invite redeemed (no-op)");
         return Json(RedeemResponse {