Treat a lodged party ceiling as U=0 instead of a pitched roof 🟩

RdSAP 10 item 2-3(h) "another dwelling above" is U=0 (spec p.45: "There is no
heat loss through the roof of a building part that has the same dwelling above
or another dwelling above"). gov-API `roof_construction = 3` is that code, but
the mapper labelled it "Pitched (slates/tiles), no access to loft" (which is
code 5), so the per-part party override at heat_transmission.py:1172 never
matched and the part fell through to the dwelling_type-label-only
has_exposed_roof heuristic.

190 of the 200 corpus code-3 parts sit at index 0 and are already suppressed by
`_cert_lodges_exposed_roof` off the roof_insulation_location == "ND" signal. The
unmitigated set is the 10 parts at index > 0, where extensions default to
exposed and were billed a phantom pitched roof. Also removes a latent PV defect:
the bogus "Pitched" prefix satisfied `is_pitched` and divided PV array area by
cos(35 deg), inflating kWp.

Shipped with the stairwell fix so the corpus pin moves once: within-0.5
78.8% -> 79.6%, SAP MAE 0.626 -> 0.611, PE MAE 2.9 -> 2.7.

The roof change alone trades 4 certs out of the +-0.5 band for 2 in (one moving
4.4 SAP closer). Those 4 were investigated and are NOT a masked bug: the corpus
error distribution is symmetric (mean +0.076, median +0.03, sigma 1.606; 112
certs above +0.5 vs 102 below), ground-floor flats show 7 positive / 8 negative
with no excess over the null, and closing them needs HLC changes of 2.3-5.6% —
below the ~8.8% noise floor of the RHI anchor, calibrated on the 76 certs whose
SAP is exact (space-heating ratio 1.062 +/- 0.088).

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

datatypes/epc/domain/mapper.py

@@ -4607,7 +4607,23 @@ _API_FLOOR_CONSTRUCTION_TO_STR: Dict[int, Optional[str]] = {
 # here too rather than left half-fixed.
 _API_ROOF_CONSTRUCTION_TO_STR: Dict[int, Optional[str]] = {
     1: "Flat",
-    3: "Pitched (slates/tiles), no access to loft",
+    # RdSAP 10 item 2-3(h) "another dwelling above" — U=0 (spec p.45: "There is
+    # no heat loss through the roof of a building part that has the same
+    # dwelling above or another dwelling above"). Previously mislabelled
+    # "Pitched (slates/tiles), no access to loft" (that is code 5), so the
+    # per-part party override at `heat_transmission.py:1172` never matched.
+    # 190 of the 200 corpus code-3 parts sit at index 0 and are already
+    # suppressed by `_cert_lodges_exposed_roof` off the `roof_insulation_location
+    # == "ND"` signal; the unmitigated set is the 10 parts at index > 0, where
+    # extensions default to exposed and were billed a phantom pitched roof.
+    # Also removes a latent PV defect: the bogus "Pitched" prefix satisfied
+    # `is_pitched` (cert_to_inputs.py:3559) and divided PV array area by
+    # cos(35°), inflating kWp. Corpus evidence (1000 certs, 200 code-3 parts):
+    # 200/200 lodge roof_insulation_location "ND", 186/200 lodge the roof
+    # description literally "(another dwelling above)", zero lodge a "no access
+    # to loft" description, and no top-floor dwelling ever lodges it. Codes 7
+    # and 9 were corrected by an earlier slice; 3 was left behind.
+    3: "(another dwelling above)",
     4: "Pitched (slates/tiles), access to loft",
     5: "Pitched (vaulted ceiling)",
     6: None,

tests/datatypes/epc/domain/test_mapper_roof_party_ceiling.py

@@ -0,0 +1,78 @@
+"""Regression: gov-API `roof_construction = 3` is "another dwelling above"
+(RdSAP 10 item 2-3(h), U=0), NOT a pitched roof.
+
+RdSAP 10 item 2-3 "Above the building part" (spec PDF p.72, printed p.71) lists
+the codes and their treatment; (g) "same dwelling above" and (h) "another
+dwelling above" are both U=0, and p.45 states it plainly: "There is no heat loss
+through the roof of a building part that has the same dwelling above or another
+dwelling above."
+
+The mapper labelled code 3 "Pitched (slates/tiles), no access to loft", so the
+per-part party override in `heat_transmission.py` — which matches on the literal
+string "another dwelling above" in `roof_construction_type` — never fired. The
+building part then fell through to the dwelling-level `has_exposed_roof`
+heuristic, which is keyed only on the `dwelling_type` LABEL. Where that label
+says top-floor (or the part is an extension, billed exposed unconditionally), we
+invented an entire pitched roof over a lodged party ceiling.
+
+Evidence that code 3 is a party ceiling, from the committed RdSAP-21.0.1 corpus
+(1000 certs, 200 code-3 building parts):
+  - 200/200 lodge `roof_insulation_location == "ND"` (not determined — there is
+    no insulation to determine on a party ceiling);
+  - 186/200 lodge the cert's roof description as literally
+    "(another dwelling above)";
+  - dwelling types are exclusively non-top-floor (Mid-floor flat 104,
+    Ground-floor flat 86, Ground-floor maisonette 7, Mid-floor maisonette 1,
+    Mid-terrace house 2) — no top-floor dwelling ever lodges it.
+
+Codes 7 and 9 were corrected to the party string by an earlier slice; code 3 was
+left behind.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any
+
+from datatypes.epc.domain.mapper import EpcPropertyDataMapper
+
+_FIXTURE = Path("backend/epc_api/json_samples/RdSAP-Schema-21.0.1/epc.json")
+
+_PARTY_CEILING = "another dwelling above"
+
+
+def _load_with_roof_construction(code: int) -> dict[str, Any]:
+    data: dict[str, Any] = json.loads(_FIXTURE.read_text())
+    for part in data["sap_building_parts"]:
+        part["roof_construction"] = code
+    return data
+
+
+def test_roof_construction_3_maps_to_a_party_ceiling() -> None:
+    # Arrange — the lodged part has another dwelling above it (API code 3).
+    data = _load_with_roof_construction(3)
+
+    # Act
+    epc = EpcPropertyDataMapper.from_api_response(data)
+
+    # Assert — carries the party-ceiling marker the per-part override matches on,
+    # so the roof is billed at U=0 rather than as an invented pitched roof.
+    assert epc is not None
+    assert epc.sap_building_parts
+    for part in epc.sap_building_parts:
+        assert _PARTY_CEILING in (part.roof_construction_type or "").lower()
+
+
+def test_roof_construction_4_remains_a_real_pitched_roof() -> None:
+    # Arrange — code 4 is 2-3(a) "pitched roof, access to loft", a REAL roof.
+    data = _load_with_roof_construction(4)
+
+    # Act
+    epc = EpcPropertyDataMapper.from_api_response(data)
+
+    # Assert — must not be swept up by the code-3 correction.
+    assert epc is not None
+    assert epc.sap_building_parts
+    for part in epc.sap_building_parts:
+        assert _PARTY_CEILING not in (part.roof_construction_type or "").lower()

tests/infrastructure/epc_client/test_sap_accuracy_corpus.py

@@ -267,7 +267,7 @@ _CORPUS = Path(
 # unheated corridor, 3 = unheated stairwell), which was read NOWHERE under
 # domain/, so every sheltered wall took the corridor 0.5 and every stairwell
 # flat over-billed its sheltered wall. 23 corpus certs; within-0.5 +1.0pp.
-_MIN_WITHIN_HALF_SAP = 0.798
+_MIN_WITHIN_HALF_SAP = 0.796
 # 0.793 -> 0.789 via the §12 Unknown-meter + dual-electric-immersion off-peak
 # trigger (RdSAP 10 PDF p.62): Apartment 241 (main 691 + 903 dual immersion)
 # -5.38 -> -1.05. Worksheet-validated on "simulated case 48" (Elmhurst SAP 57,
@@ -393,7 +393,7 @@ _MIN_WITHIN_HALF_SAP = 0.798
 # The -0.672 overshoot is expected to move again when the C4 plant-
 # efficiency and C3.2 pumping siblings land. Unit-pinned in
 # test_cert_to_inputs.
-_MAX_SAP_MAE = 0.614
+_MAX_SAP_MAE = 0.611
 _MAX_CO2_MAE_TONNES = 0.072     # t CO2 / yr vs co2_emissions_current
 _MAX_PE_PER_M2_MAE = 3.0        # kWh / m2 / yr vs energy_consumption_current