fix(sap): (203)-weight the dual-main Table 4e MIT adjustment

For a dwelling with two main systems heating different parts, the (93)m
Table 4e mean-internal-temperature adjustment was taken from system 1 alone
(`details[0]`). SAP 10.2 p.186 requires the (203)-weighted mean of each
system's control adjustment — (1 - (203)) for system 1, (203) for system 2 —
mirroring the Table 9b responsiveness weighting already applied two lines
above. Keyed to system 1, a dwelling led by a small-fraction system (e.g. a
20% HHRSH, adj 0.0 °C, ahead of an 80% manual-charge storage heater, adj
+0.7 °C) was modelled ~0.7 °C too cool → space heat under-counted → over-rated.

Compute the weighted adjustment at the call site. Shared `cert_to_inputs`
path, so it also lifts the accredited gov-API RdSAP corpus (81.5% → 81.7%,
guardrail). 5 Edmund Close +2.74 → +0.59; 4 Edmund Close -2.51 → -0.55.
LRHA cohort within-0.5 60.2% → 62.1%, MAE 0.496 → 0.445; ratchets re-based to
0.62 / 0.45. Also corrects an oil-combi docstring (462051619031 is a Grant
Vortex 10599, not the Worcester 18415).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

backend/documents_parser/tests/test_pashub_sap_accuracy_lrha_wave3.py

@@ -128,8 +128,24 @@ _MANIFEST_PATH = _FIXTURES_DIR / "manifest.json"
 # P960 worksheet exactly (5.4175 kWh/day). The 5 oil storage-combis collapse
 # from +2.0..+4.9 into the ±1.5 spread: within-0.5 55.3% → 58.3%, MAE
 # 0.675 → 0.531.
-_MIN_WITHIN_HALF: float = 0.58
-_MAX_SAP_MAE: float = 0.54
+# 2026-07-27 (fan-out audit — 3 spec-correct fixes, each with a named test
+# above): (1) the window channel collapsed MIXED per-window glazing onto
+# windows[0]'s U (SAP 10.2 (26)-(27) sums each window's own Table 24 U); (2) a
+# surveyed measured cylinder heat-loss factor was dropped on the PasHub path (an
+# extractor label mismatch + the mapper never wiring `cylinder_heat_loss`), so
+# the calc fell back to the Table 2 insulation default (SAP 10.2 §4 (48)-(50)
+# uses the declared loss); (3) the dual-main Table 4e MIT adjustment was keyed
+# to system 1 alone, not (203)-weighted across both mains (SAP 10.2 p.186).
+# Fixes (1)/(3) are symmetric — they correct over- AND under-raters — and (3)
+# lives in the shared `cert_to_inputs` path, improving the accredited gov-API
+# RdSAP corpus too (81.5% → 81.7%, guardrail). Combined: within-0.5 58.3% →
+# 62.1%, MAE 0.531 → 0.445. The residual top-of-cohort is now entirely the
+# Varsity Rise community-heating cluster (a solar-DHW over-credit + an immersion
+# fuel-collision), deferred to Elmhurst adjudication (per-cert build sheets
+# prepared). Floor/ceiling re-baselined to the corrected cohort — coverage
+# correction is not loosening.
+_MIN_WITHIN_HALF: float = 0.62
+_MAX_SAP_MAE: float = 0.45
 
 _KNOWN_GAP_REASON = (
     "pashub `from_site_notes` mapper does not int-code a main-heating "
@@ -229,7 +245,11 @@ def test_oil_storage_combi_applies_pcdb_store_loss() -> None:
     The five oil storage-combis (461996268736, 461989738731, 497712315613,
     497602074839, 462051619031) all collapse from +2..+4.9 into the cohort's
     normal ±1.5 spread; only this one is pinned (the others' residuals are
-    PasHub-vs-Elmhurst divergence, not the store-loss bug)."""
+    PasHub-vs-Elmhurst divergence, not the store-loss bug). NB the five share
+    the store_type-1 storage-combi treatment but NOT one PCDB product — e.g.
+    462051619031 is a Grant Vortex Outdoor Condensing Combi (PCDB 10599), not
+    the Worcester Greenstar Heatslave II (18415) pinned here; the Table 2 note-b
+    store loss uses each boiler's own PCDB store volume/insulation."""
     outcome = _evaluate("461996268736")
     assert outcome.diff is not None
     assert outcome.diff < 0.5, f"oil storage-combi store loss not applied: diff {outcome.diff:.2f}"
@@ -321,6 +341,25 @@ def test_measured_cylinder_loss_overrides_table2() -> None:
     assert outcome.diff < 0.5, f"measured cylinder loss not applied: diff {outcome.diff:.2f}"
 
 
+@pytest.mark.skipif(
+    "462019682527" not in _BY_ID, reason="dual-main Table-4e fixture not in manifest"
+)
+def test_dual_main_table_4e_adjustment_is_fraction_weighted() -> None:
+    """Regression: for a dwelling with two main systems heating different parts
+    (5 Edmund Close, 462019682527 — a 20% HHRSH, Table 4e adj 0.0 °C, + an 80%
+    manual-charge storage heater, adj +0.7 °C), the (93)m Table 4e mean-internal-
+    temperature adjustment must be the (203)-weighted mean of both systems'
+    control adjustments per SAP 10.2 p.186, not system 1's alone. Keyed to the
+    20% HHRSH the adjustment was 0.0 (dwelling modelled ~0.7 °C too cool → space
+    heat under-counted) → over-rated +2.74. The (203)-weighted +0.56 °C brings
+    it to +0.59. Bound 1.0 guards against regression to the +2.74 mis-weight; the
+    residual is normal cohort spread. NB this fix lives in the shared
+    `cert_to_inputs` path and improves the gov-API RdSAP corpus too (81.5→81.7%)."""
+    outcome = _evaluate("462019682527")
+    assert outcome.diff is not None
+    assert outcome.diff < 1.0, f"Table 4e adjustment not (203)-weighted: diff {outcome.diff:.2f}"
+
+
 @pytest.mark.skipif(not _FIXTURES, reason="no pashub_accuracy_lrha_wave3 fixtures/manifest")
 @pytest.mark.parametrize("deal_id", _IDS)
 def test_pashub_fixture_computes(deal_id: str) -> None:

domain/sap10_calculator/rdsap/cert_to_inputs.py

@@ -8405,6 +8405,18 @@ def cert_to_inputs(
         main_2_responsiveness_value = _responsiveness(
             _mit_main_2, tariff=_mit_tariff, epc=epc
         )
+    # SAP 10.2 p.186: the (93)m Table 4e temperature adjustment for a dwelling
+    # with two main systems heating different parts is the (203)-weighted mean
+    # of each system's control adjustment — (1 - (203)) for system 1, (203) for
+    # system 2 — mirroring the Table 9b responsiveness weighting above. Keyed to
+    # system 1 alone it ignores the dominant system's control when the two carry
+    # different Table 4e controls (e.g. a 20% HHRSH + 80% manual-charge storage).
+    control_temperature_adjustment_value = _control_temperature_adjustment_c(main)
+    if _mit_main_2 is not None and main_2_control_type_value is not None:
+        control_temperature_adjustment_value = (
+            (1.0 - main_2_fraction_value) * _control_temperature_adjustment_c(main)
+            + main_2_fraction_value * _control_temperature_adjustment_c(_mit_main_2)
+        )
     monthly_total_gains_w = tuple(
         internal_gains_monthly_w[m] + solar_gains_monthly_w[m] for m in range(12)
     )
@@ -8429,7 +8441,7 @@ def cert_to_inputs(
         control_type=control_type_value,
         responsiveness=responsiveness_value,
         living_area_fraction=living_area_fraction_value,
-        control_temperature_adjustment_c=_control_temperature_adjustment_c(main),
+        control_temperature_adjustment_c=control_temperature_adjustment_value,
         main_2_control_type=main_2_control_type_value,
         main_2_fraction=main_2_fraction_value,
         main_2_responsiveness=main_2_responsiveness_value,