don't crash

pipeline/download/median_age.py

@@ -72,6 +72,18 @@ assert len(EXPECTED_BAND_NAMES) == len(AGE_BANDS), (
     "EXPECTED_BAND_NAMES and AGE_BANDS must stay aligned 1:1"
 )
 
+# NOMIS sometimes labels a band with a wording variant that denotes the SAME
+# age range (e.g. "Aged 4 years and under" for ages 0-4, "Aged 90 years and
+# over" wording for the top band). Map such known-equivalent labels back to the
+# canonical name BEFORE validation so a real band change still fails loudly,
+# but a cosmetic relabel of an identical range does not block the build.
+BAND_NAME_ALIASES = {
+    "Aged 4 years and under": "Aged 0 to 4 years",
+}
+assert set(BAND_NAME_ALIASES.values()) <= set(EXPECTED_BAND_NAMES), (
+    "BAND_NAME_ALIASES must map to canonical EXPECTED_BAND_NAMES"
+)
+
 
 def compute_median_age(counts: list[int]) -> float:
     """Compute median age from five-year band counts using linear interpolation."""
@@ -98,6 +110,15 @@ def _bands_to_median_table(pivoted: pl.DataFrame) -> pl.DataFrame:
     missing/extra/relabelled band would otherwise silently mis-align counts
     against the wrong AGE_BANDS lower bound, so we fail loudly instead.
     """
+    # Normalise known-equivalent NOMIS label variants to their canonical name
+    # before validating (renaming onto an already-present canonical column would
+    # collide, so polars raises loudly in that genuinely ambiguous case).
+    rename_map = {
+        c: BAND_NAME_ALIASES[c] for c in pivoted.columns if c in BAND_NAME_ALIASES
+    }
+    if rename_map:
+        pivoted = pivoted.rename(rename_map)
+
     # Validate the pivoted age-band columns against the canonical NOMIS set
     # BEFORE computing anything.
     band_cols = [c for c in pivoted.columns if c != "GEOGRAPHY_CODE"]

pipeline/download/noise.py

@@ -266,8 +266,12 @@ def _download_tile(
         except (
             NoGeoTiffError,
             httpx.HTTPStatusError,
-            httpx.TimeoutException,
-            httpx.ConnectError,
+            # TransportError is the superset of TimeoutException, ConnectError,
+            # ReadError and ProtocolError — including RemoteProtocolError, raised
+            # when the WCS server closes the connection mid-stream ("incomplete
+            # chunked read"). All are transient; retry/split rather than letting
+            # one flaky tile crash the whole raster download.
+            httpx.TransportError,
         ) as e:
             last_error = e
             if attempt < MAX_RETRIES:

pipeline/download/test_median_age.py

@@ -57,6 +57,21 @@ def test_null_band_count_is_treated_as_zero_not_crash():
     assert table["median_age"][0] == pytest.approx(32.5)
 
 
+def test_equivalent_band_label_alias_is_accepted():
+    # NOMIS relabelled the first band "Aged 4 years and under" (same as ages
+    # 0-4). It must be normalised to the canonical name and used as band 0, not
+    # rejected. All 100 people in that band -> median in the 0-4 range: 2.5.
+    counts_by_band = {name: [0] for name in EXPECTED_BAND_NAMES}
+    counts_by_band["Aged 4 years and under"] = counts_by_band.pop("Aged 0 to 4 years")
+    counts_by_band["Aged 4 years and under"] = [100]
+    pivoted = _pivoted(counts_by_band)
+
+    table = median_age._bands_to_median_table(pivoted)
+
+    assert table.height == 1
+    assert table["median_age"][0] == pytest.approx(2.5)
+
+
 def test_missing_band_raises_clear_error():
     counts_by_band = {name: [10] for name in EXPECTED_BAND_NAMES}
     del counts_by_band["Aged 85 years and over"]