Re-baseline pins for full-SAP PV + tariff mapper fixes 🟩

The PV-array (ea7f4f43) and electricity-tariff (6ec09892) mapper fixes shifted
the observed output of five frozen gates that weren't updated alongside:

- has_pv component-accuracy floor 0.9798 -> 0.9697: carrying full-SAP lodged PV
  now reads the true has_pv=True for full-SAP PV dwellings, so the leave-one-out
  scorer's actual changes (ground-truth-method shift, ADR-0037 pattern).
- uprn_10093116528 80->... pin 82 -> 83: tariff=1 (standard) was wrongly read as
  dual/Economy 7; translating to "single" re-prices the gas semi's electricity.
- uprn_10096028301 82 -> 84, uprn_10023444324 80 -> 82 (== lodged 82),
  uprn_10023444320 81 -> 83: now credit the lodged sap_energy_source.pv_arrays
  the schema previously dropped. Comments document the per-cert PV/Elmhurst
  relationship (incl. the mid-floor sibling landing +2 over its lodged integer).

Pre-existing, unrelated failures untouched: the missing
sap_16_0_full_no_floor_dims.json fixture and the RdSAP-21 floor-area test (both
reproduce on origin/main).

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

tests/domain/epc_prediction/test_component_accuracy_gate.py

@@ -56,6 +56,15 @@ _FIXTURE = Path(__file__).parents[3] / "tests" / "fixtures" / "epc_prediction"
 # new-build-vs-old-stock service mismatch on 1-2 targets each (heating_main_fuel
 # 0.9722->0.9394, water_heating_fuel ->0.9495, cylinder_insulation_type 0.6667->
 # 0.3333) plus floor_area (+0.31 MAE). Tighten-only resumes from these values.
+#
+# has_pv re-baselined 0.9798->0.9697 when full-SAP lodged PV mapping landed
+# (datatypes/epc/domain/mapper.py `_sap_17_1_pv_arrays`): full-SAP certs lodge
+# their measured array under `sap_energy_source.pv_arrays`, which the schema
+# dropped at parse, so the leave-one-out scorer's *actual* has_pv read False for
+# every full-SAP PV dwelling. Carrying the array now reads the true has_pv=True,
+# and one full-SAP target the similarity-weighted donors don't predict as PV
+# tips the agreement 32/33 (the held-out actual is now correct — a ground-truth-
+# method change, not a prediction-logic loosening). Tighten-only resumes here.
 _RATE_FLOORS: dict[str, float] = {
     "wall_construction": 0.9091,
     "wall_insulation_type": 0.8687,
@@ -76,7 +85,7 @@ _RATE_FLOORS: dict[str, float] = {
     "floor_insulation": 0.9375,
     "has_room_in_roof": 0.9495,
     "modal_glazing_type": 0.8384,
-    "has_pv": 0.9798,
+    "has_pv": 0.9697,
     "solar_water_heating": 1.0000,
 }
 

tests/domain/sap10_calculator/test_real_cert_sap_accuracy.py

@@ -122,12 +122,18 @@ _EXPECTATIONS: Final[tuple[RealCertExpectation, ...]] = (
     # (engine uses the cert's measured 0.19/0.11/0.11 U-values; Elmhurst uses
     # age-band L proxies + party-wall default) plus FGHRS (cert idx 60031) omitted
     # on BOTH sides (the engine can't yet model full-SAP FGHRS). PINNED TO THE
-    # OBSERVED 82, not lodged 84 — mapping deliberately untuned.
+    # OBSERVED 83 (was 82), not lodged 84 — mapping deliberately untuned.
+    # WAS 82 until the full-SAP electricity-tariff → RdSAP meter_type fix: this
+    # cert lodges energy_tariff=1 (standard), which the mapper previously passed
+    # through untranslated as RdSAP meter_type "1" — wrongly read as dual/Economy 7
+    # and priced on the off-peak high/low split. Translating it to "single" (the
+    # correct standard tariff) re-prices its electricity at the flat rate, lifting
+    # this gas semi 82→83. No PV (sap_energy_source.pv_arrays absent).
     RealCertExpectation(
         schema="SAP-Schema-17.1",
         sample="uprn_10093116528",
         cert_num="8000-8495-2839-2607-9683",
-        sap_score=82,
+        sap_score=83,
     ),
     # UPRN 10093116543 → cert 8358-7436-5620-6889-0906. SAP-Schema-17.1 — a
     # FULL-SAP cert (2017 mains-gas COMBI semi, Emsworth), forced through the
@@ -290,13 +296,18 @@ _EXPECTATIONS: Final[tuple[RealCertExpectation, ...]] = (
     # control 2106 (CBE); water from primary (combi); MEV on; AP50 Blower Door 3.5.
     # The −3 vs lodged 85 is the documented full-SAP→RdSAP gap: the engine uses the
     # cert's MEASURED U (wall 0.24 / floor 0.13, WORSE than RdSAP band-M defaults)
-    # + MEV priced as extract loss not heat recovery. PINNED to the observed 82 —
-    # mapping untuned; engine == Elmhurst.
+    # + MEV priced as extract loss not heat recovery. PINNED to the observed 84
+    # (was 82), still −1 vs lodged 85 — mapping untuned.
+    # WAS 82 until full-SAP lodged PV mapping landed: this cert lodges a 0.38 kWp
+    # array under sap_energy_source.pv_arrays (SE-facing, pitch 30°, unshaded) that
+    # the schema dropped at parse, so the Appendix-M generation credit was lost.
+    # Carrying it (mapper `_sap_17_1_pv_arrays`) credits the generation and lifts
+    # this flat 82→84, closing most of the gap to the lodged 85 the array explains.
     RealCertExpectation(
         schema="SAP-Schema-19.1.0",
         sample="uprn_10096028301",
         cert_num="0390-3321-6060-2405-7985",
-        sap_score=82,
+        sap_score=84,
     ),
     # UPRN 44012843 → cert 0775-2898-6628-9594-8005. SAP-Schema-16.3 — a
     # reduced-field (RdSAP-shaped) ground-floor FLAT, band K (2007-2011), cavity
@@ -326,14 +337,20 @@ _EXPECTATIONS: Final[tuple[RealCertExpectation, ...]] = (
     # worksheet SAP 80 — engine EXACTLY matches (80.13 vs 80); engine-on-Elmhurst's-
     # own-parsed-inputs 81.03 ≈ 80 → calculator faithful. Boiler set to the cert's
     # exact PCDB 16211 via the search dialog; control 2106 (CBE); water from primary
-    # (combi); MEV on; AP50 Blower Door 3.2; party wall 6.43 m entered. The −2 vs
-    # lodged 82 is the documented full-SAP→RdSAP gap (measured U 0.2/0.1 + MEV
-    # extract loss). PINNED to the observed 80 — mapping untuned; engine == Elmhurst.
+    # (combi); MEV on; AP50 Blower Door 3.2; party wall 6.43 m entered.
+    # WAS 80 (engine == Elmhurst, both built WITHOUT PV) until full-SAP lodged PV
+    # mapping landed: this cert lodges a 0.48 kWp array under
+    # sap_energy_source.pv_arrays (SE-facing, pitch 30°, unshaded) the schema
+    # dropped at parse. Crediting it (mapper `_sap_17_1_pv_arrays`) closes the
+    # −2 gap exactly — the engine now reproduces the accredited lodged 82. The
+    # Elmhurst worksheet (80) omitted the PV (not entered in the RdSAP build), so
+    # the +2 over Elmhurst is the now-credited array, not a calculator drift.
+    # PINNED to the observed 82 == lodged 82 — mapping untuned.
     RealCertExpectation(
         schema="SAP-Schema-17.0",
         sample="uprn_10023444324",
         cert_num="8501-5064-6739-1407-0163",
-        sap_score=80,
+        sap_score=82,
     ),
     # UPRN 10023444320 → cert 0868-6045-7331-4376-0914. SAP-Schema-17.0 — FULL-SAP
     # MID-FLOOR FLAT (sibling of 10023444324, same block / combi PCDB 16211 / MEV),
@@ -342,12 +359,24 @@ _EXPECTATIONS: Final[tuple[RealCertExpectation, ...]] = (
     # worksheet 82 — engine within ~1 (81.38 vs 82); engine-on-Elmhurst-inputs 82.46
     # ≈ 82 → calculator faithful. Boiler PCDB 16211 via search; control 2106 (CBE);
     # water from primary (combi); MEV on; AP50 Blower Door 3.09; mid-floor (floor =
-    # another dwelling below). PINNED to the observed 81 — mapping untuned.
+    # another dwelling below).
+    # WAS 81 until full-SAP lodged PV mapping landed: this cert lodges the SAME
+    # 0.48 kWp array as its ground-floor sibling 10023444324 under
+    # sap_energy_source.pv_arrays (the block's roof PV apportioned to the flat on
+    # the lodged cert). Crediting it faithfully (mapper `_sap_17_1_pv_arrays`)
+    # lifts this flat 81→83. NOTE this lands +2 OVER the lodged 81 (and +1 over the
+    # Elmhurst worksheet 82) — unlike the ground-floor sibling whose pre-PV engine
+    # was 2 UNDER lodged so the same array closed the gap exactly. The mid-floor's
+    # pre-PV engine already matched lodged, so the credited array now overshoots:
+    # the lodged 81 does not appear to carry the array's full generation credit
+    # that SAP Appendix-M awards it. This is the documented full-SAP→RdSAP residual
+    # (faithful to the cert's lodged PV, not tuned to the lodged integer). PINNED
+    # to the observed 83 — mapping untuned.
     RealCertExpectation(
         schema="SAP-Schema-17.0",
         sample="uprn_10023444320",
         cert_num="0868-6045-7331-4376-0914",
-        sap_score=81,
+        sap_score=83,
     ),
     # UPRN 10090844932 → cert 0646-3008-6208-0619-6204. RdSAP-Schema-20.0.0 —
     # END-TERRACE HOUSE, 2-storey, band L (2012-2022), cavity insulated, pitched