Score heating-oil dwellings via SAP main-fuel code 28 🟩

#1632: LRHA WAVE 3 45→51 computed. Newly-unblocked oil configs land
outside 0.5 (coverage growth per module convention) — within-0.5 floor
re-baselined 0.46→0.41 at 41.2%/MAE 2.948.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

backend/documents_parser/tests/test_pashub_sap_accuracy_lrha_wave3.py

@@ -48,10 +48,14 @@ _MANIFEST_PATH = _FIXTURES_DIR / "manifest.json"
 # '...2002-2021', cylinder size 'Exact cylinder volume' / 'Medium (131-170
 # litres)', 'Heating oil' fuel, community-heating charging controls, 'MV - Heat
 # Recovery' — each its own field-by-field follow-up, not tuning.
+# 2026-07-16 (#1622/#1632): 'Heating oil' main fuel → 28 unblocked the oil
+# dwellings → **51 computed / 53 blocked** (within-0.5 41.2% = 21/51, MAE
+# 2.948). The newly-computable oil configs land outside 0.5 per the
+# coverage-growth convention above — floor re-baselined 0.46 → 0.41.
 # Treat the constants as a regression tripwire, NOT a cohort accuracy figure;
 # re-baseline (coverage growth is not loosening) as further mapper fixes unblock
 # fixtures.
-_MIN_WITHIN_HALF: float = 0.46
+_MIN_WITHIN_HALF: float = 0.41
 _MAX_SAP_MAE: float = 3.0
 
 _KNOWN_GAP_REASON = (

datatypes/epc/domain/mapper.py

@@ -7817,6 +7817,10 @@ _PASHUB_MAIN_FUEL_TO_SAP10: Dict[str, int] = {
     # once the extractor captured the `Fuel:`-form label it had been dropping.
     "Dual fuel appliance (mineral and wood)": 10,
     "Electricity": 30,  # the standard-electricity fuel code
+    # Oil boiler/heater fuel. 28 = epc_codes `main_fuel` "oil (not community)",
+    # the same code the Elmhurst map pins for the identical label; routes via
+    # `API_FUEL_TO_TABLE_32` → Table 32 code 4 (heating oil) for cost/CO2/PE.
+    "Heating oil": 28,
 }
 
 # PasHub surveyed secondary-heating `secondary_fuel` label → SAP 10.2 Table 32