Bill an electric second main at its own §12/Table-12a rate 🟩

Closes #1684. `_main_2_space_heating_fuel_cost_gbp_per_kwh` billed an
electric Main 2 at Main 1's space-heating rate. That was a no-op on a
homogeneous off-peak cohort (storage+storage, direct+direct → equal
rates) but mis-billed the HETEROGENEOUS case: an off-peak STORAGE Main 1
(5.5 p/kWh low) + an on-peak DIRECT-ACTING panel Main 2 (15.29 p/kWh
high) rode the storage low rate, under-costing the panel half and
over-rating SAP.

Flip it to bill Main 2 at its OWN Table 12a Grid 1 SH row, the same
per-system resolution Main 1 and a non-electric Main 2 already use — and
the same one `_main_2_high_rate_fraction` already applies, so the scalar
cost and the high-rate split are now consistent.

The docstring's "regresses certs 13/34 (Parkers Hill / Dunley Road)"
warning was stale: those names appear nowhere but that docstring, and
the full guard suite shows zero new regressions (only the two
pre-existing `test_heating_systems_corpus` fails on main). LRHA WAVE 3
fixture 497655371974 (storage+panel) moves +15.3 → -2.5 vs PasHub; cohort
MAE 0.799 → 0.675 (ratchet re-baselined 0.80 → 0.68). within-0.5 holds at
55.3% — the -2.5 residual is PasHub-vs-Elmhurst divergence, not a bug, so
the fixture stays out of the within-0.5 pin. A ground-truth Elmhurst pin
is queued (see docs/HANDOVER_1684_ELMHURST_INPUTS_497655371974.md).

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

backend/documents_parser/tests/test_pashub_sap_accuracy_lrha_wave3.py

@@ -109,13 +109,18 @@ _MANIFEST_PATH = _FIXTURES_DIR / "manifest.json"
 # holds `main_heating_2` + `main_heating_1_percent`, the extractor slices per
 # system, and the mapper emits two MainHeatingDetails with the fraction → 5 of 6
 # converge (within-0.5 53.4% → 55.3%, MAE 0.799). The 6th (497655371974,
-# storage+panel) regresses because the calculator bills an electric Main 2 at
-# Main 1's rate (deferred §10a slice) — the correct mapping exposes that gap.
-# 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.
+# storage+panel) still missed because the calculator billed an electric Main 2
+# at Main 1's rate (deferred §10a slice) — the correct mapping exposed that gap.
+# 2026-07-27 (#1684): the calculator now bills an electric Main 2 at its OWN
+# Table 12a Grid 1 SH rate (not Main 1's), so the storage+panel Main 2's on-peak
+# half is costed correctly: fixture 497655371974 +15.3 → -2.5. within-0.5 holds
+# at 55.3% (still >0.5 vs PasHub's hybrid figure — genuine divergence, not a
+# bug), MAE 0.799 → 0.675. Homogeneous off-peak cohorts (storage+storage) are a
+# no-op. 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 = 0.55
-_MAX_SAP_MAE: float = 0.80
+_MAX_SAP_MAE: float = 0.68
 
 _KNOWN_GAP_REASON = (
     "pashub `from_site_notes` mapper does not int-code a main-heating "
@@ -187,11 +192,12 @@ def test_dual_main_heating_models_both_systems() -> None:
     MainHeatingDetail with the split (Main 2 fraction = 100 - Main 1 %) closes it
     to 0.0.
 
-    NB the sibling storage+PANEL dual-main (497655371974) is NOT pinned here: the
-    calculator deliberately bills an electric Main 2 at Main 1's rate
-    (`_main_2_space_heating_fuel_cost_gbp_per_kwh`, deferred §10a off-peak slice),
-    so its on-peak panel half is under-costed — a documented calculator gap the
-    correct two-main mapping now exposes, tracked separately."""
+    NB the sibling storage+PANEL dual-main (497655371974) is NOT pinned here.
+    Its Main 2 on-peak panel half is now costed correctly (#1684:
+    `_main_2_space_heating_fuel_cost_gbp_per_kwh` bills an electric Main 2 at its
+    OWN Table 12a rate), moving it +15.3 → -2.5 vs PasHub. The residual is
+    PasHub-vs-Elmhurst divergence (>0.5), not a bug — a ground-truth Elmhurst
+    pin is tracked separately, so it is left out of the within-0.5 assertion."""
     outcome = _evaluate("461029305556")
     assert outcome.diff is not None
     assert outcome.diff < 0.5, f"dual-main not modelled: diff {outcome.diff:.2f}"

docs/HANDOVER_1684_ELMHURST_INPUTS_497655371974.md

@@ -0,0 +1,130 @@
+# Elmhurst RdSAP input sheet — fixture 497655371974 (4 Edmund Close)
+
+Purpose: reproduce LRHA WAVE 3 fixture **497655371974** in the Elmhurst RdSAP 10
+entry tool, download its **Input Summary + SAP Worksheets**, and give us the
+**accredited SAP score** + the **space/water-heating fuel-cost worksheet lines**.
+This anchors #1684 to accredited ground truth (per the A5 discipline: Elmhurst,
+not PasHub's hybrid `pre_sap`), so we can pin it as a `RealCertExpectation`
+instead of relying on the PasHub-relative residual.
+
+The #1684 fix is already shipped and green; this worksheet is the **confirmatory
+ground-truth anchor**, not a blocker.
+
+## What this cert is (and why it matters for #1684)
+A detached house with **two electric main heating systems on Economy 7**:
+- **Main 1** — modern-slimline electric **storage** heaters (charged off-peak,
+  ~5.5 p/kWh low rate).
+- **Main 2** — electric **direct-acting panel/convector** heaters (on-peak,
+  ~15.29 p/kWh high rate), 50% split.
+
+The bug (#1684) billed the on-peak panel Main 2 at Main 1's off-peak storage
+rate → over-rated SAP by **+15.3 vs PasHub**. The fix bills Main 2 at its own
+Table 12a rate → **SAP 40.5** (−2.5 vs PasHub 43). **The load-bearing worksheet
+lines to check are the space-heating fuel-cost rows: Main 2's kWh must bill at
+the HIGH (on-peak) rate, not Main 1's low rate.**
+
+## Our engine's output (to compare against Elmhurst)
+| Quantity | Our engine |
+|---|---|
+| SAP score (continuous / band) | **40.49** (E) |
+| Space heating demand | 9 166 kWh/yr |
+| Main 1 (storage) fuel | 7 791 kWh/yr |
+| Main 2 (panel) fuel | 3 896 kWh/yr |
+| Hot water | 1 459 kWh/yr |
+| CO₂ | 1 555 kg/yr |
+
+If Elmhurst lands materially away from ~40.5, that is the signal to dig — but
+the direction (panel billed on-peak) is not in doubt (SAP 10.2 Table 12a Grid 1,
+`OTHER_DIRECT_ACTING_ELECTRIC` = 100% high rate on a 7-hour tariff).
+
+---
+
+## Page-by-page inputs
+
+### Property / general
+- Address: **4 Edmund Close, Gainsborough, DN21 5RN**
+- Property type: **House** · Built form: **Detached**
+- Dwelling type: Detached house · Tenure: Rented (social)
+- Transaction type: None of the above
+- Inspection date: 2026-04-30
+- Age band: **F** (England & Wales 1976–1982)
+- Storeys: **2** · Habitable rooms: **5** · Heated habitable rooms: 5
+- Total floor area: **78.3 m²**  (ground 38.24 + first 38.24 + extension 1.82)
+- Extensions: **1**
+- Country: England
+
+### Dimensions
+| Level | Storey height (m) | Floor area (m²) | Heat-loss perimeter (m) |
+|---|---|---|---|
+| Ground (main) | 2.62 | 38.24 | 25.64 |
+| First (main) | 2.30 | 38.24 | 25.64 |
+| Extension (ground) | 2.30 | 1.82 | 4.02 |
+
+Party-wall length: 0 (detached).
+
+### Walls
+- Construction: **Cavity**
+- Insulation: **Filled cavity** (retrofit CWI)
+- Thickness: **300 mm** (measured)
+- Dry-lined: No
+
+### Roof
+- Main: **Pitched, slates/tiles, access to loft** · Insulation at **joists**,
+  **250 mm**
+- Extension: Pitched, slates/tiles, **no access** · Insulation: unknown
+  (RdSAP default for age band)
+
+### Floor
+- **Solid** ground floor · Insulation: **as built** (none assumed) · U unknown
+
+### Windows / doors
+- Glazing / low-E: RdSAP defaults for age band F (no override surveyed)
+- Doors: **3** (all draught-proofed) · Insulated doors: 0
+- Draught-proofing: **100%**
+
+### Ventilation
+- **Natural** ventilation · No pressure test
+- **Draught lobby: present**
+- Open chimneys: 0 · Blocked chimneys: **1** · Open/closed/boiler/other flues: 0
+- Extract fans: 0 · Passive vents: 0 · Flueless gas fires: 0
+- Sheltered sides: 0
+
+### Lighting
+- Low-energy fixed lighting: **11** bulbs (of 11 → 100% low-energy)
+
+### Electricity meter / tariff
+- Meter type: **Dual (Economy 7 / 7-hour off-peak)**  ← load-bearing for #1684
+- Gas connection: **No** (all-electric dwelling)
+- PV / wind / battery: none · Export-capable: yes (no generation)
+
+### Main heating system 1
+- **Electric storage heaters — modern slimline** (SAP Table 4a **402**)
+- Fuel: **Electricity** · Charge control: **Manual** (SAP control 2401)
+- Fraction: **50%**
+
+### Main heating system 2
+- **Electric direct-acting — panel, convector or radiant heaters**
+  (SAP Table 4a **691**)
+- Fuel: **Electricity (on-peak)** · Control: **Appliance thermostats** (2602)
+- Fraction: **50%**
+
+### Secondary heating
+- **None**
+
+### Water heating
+- **Electric immersion** cylinder (SAP WHC **903**), fuel electricity
+- Immersion: **single** · on the Economy-7 (off-peak) meter
+- Cylinder size: **Small band ≈110 L** (RdSAP Table 28 code 2)
+- Insulation: **factory-fitted foam, 50 mm** · Cylinder thermostat: **No**
+- Electric shower ×1 · Bath ×1 · Mixer shower ×0
+- Solar water heating: No · WWHRS: None
+
+---
+
+## What to send back
+1. **SAP score** (continuous + band) from the Elmhurst Input Summary.
+2. The **SAP Worksheet** pages covering the **§10a/§12 fuel-cost lines** — the
+   space-heating cost split showing Main 1 (storage, low rate) vs Main 2 (panel,
+   high rate), and the per-fuel p/kWh applied.
+3. (Nice to have) the per-end-use kWh (space heating, water) to pin alongside the
+   score as `space_heating_kwh_per_yr` / `hot_water_kwh_per_yr`.

domain/sap10_calculator/rdsap/cert_to_inputs.py

@@ -2883,24 +2883,25 @@ def _main_2_space_heating_fuel_cost_gbp_per_kwh(
     prices: PriceTable,
 ) -> float:
     """Main heating system 2's space-heating fuel rate (£/kWh) for the
-    legacy/off-peak scalar cost path. Keyed on main 2's OWN fuel via the same
-    `_space_heating_fuel_cost_gbp_per_kwh` logic (off-peak Table 12a split when
-    main 2 is electric, flat Table 32 rate otherwise) — so a wood-log or other
-    non-electric second main is not billed at main 1's electric rate. Returns
-    0.0 when no second main is lodged (multiplied by main 2's 0 kWh).
+    legacy/off-peak scalar cost path. Keyed on main 2's OWN system via the same
+    `_space_heating_fuel_cost_gbp_per_kwh` logic Main 1 uses (off-peak Table 12a
+    Grid 1 SH high/low split when main 2 is electric, flat Table 32 rate
+    otherwise). Returns 0.0 when no second main is lodged (multiplied by main
+    2's 0 kWh).
 
-    Scoped to a NON-electric second main (the unambiguous correction): a wood/
-    oil/coal second main billed at main 1's rate is plainly wrong. An ELECTRIC
-    second main keeps main 1's space-heating scalar — its off-peak Table 12a
-    high/low split is the deferred §10a off-peak slice, and applying the split
-    per-system here regresses the off-peak electric cohort (certs 13 Parkers
-    Hill / 34 Dunley Road)."""
+    An ELECTRIC main 2 is billed at its OWN Table 12a Grid 1 SH row, not main
+    1's rate (#1684). On a HOMOGENEOUS off-peak cohort (storage+storage,
+    direct+direct) main 2's own rate already equals main 1's, so this is a
+    no-op; it only corrects the HETEROGENEOUS case — e.g. off-peak STORAGE main
+    1 (5.5 p/kWh low) + on-peak DIRECT-ACTING panel main 2 (15.29 p/kWh high),
+    where inheriting main 1's storage low rate under-costs the on-peak panel
+    half and over-rates SAP (LRHA WAVE 3 fixture 497655371974 read +15.3). This
+    keeps the scalar cost consistent with `_main_2_high_rate_fraction`, which
+    already resolves per-system."""
     details = epc.sap_heating.main_heating_details if epc.sap_heating else None
     if not details or len(details) < 2:
         return 0.0
     main_2 = details[1]
-    if _is_electric_main(main_2):
-        return _space_heating_fuel_cost_gbp_per_kwh(details[0], tariff, prices)
     return _space_heating_fuel_cost_gbp_per_kwh(main_2, tariff, prices)
 
 

tests/domain/sap10_calculator/rdsap/test_cert_to_inputs.py

@@ -620,31 +620,81 @@ def test_off_peak_main_2_non_electric_priced_at_own_fuel() -> None:
     assert abs(rate - 0.0423) <= 1e-9
 
 
-def test_off_peak_main_2_electric_keeps_main_1_scalar() -> None:
-    # Arrange — both mains electric on an off-peak tariff. The electric
-    # second-main off-peak Table 12a split is the deferred §10a slice, so the
-    # helper keeps main 1's space-heating scalar (no per-system split here) to
-    # avoid regressing the off-peak electric cohort.
+def test_off_peak_main_2_electric_billed_at_own_table_12a_rate() -> None:
+    # Arrange — HETEROGENEOUS electric mains on an off-peak (7-hour) tariff:
+    # Main 1 = modern-slimline STORAGE heaters (SAP 402, Table 12a
+    # OTHER_STORAGE_HEATERS → charged wholly off-peak → 5.5 p/kWh low rate);
+    # Main 2 = DIRECT-ACTING panel/convector heaters (SAP 691, Table 12a
+    # OTHER_DIRECT_ACTING_ELECTRIC → 100% high rate on 7-hour → 15.29 p/kWh).
+    # An electric Main 2 must be billed at its OWN Table 12a Grid 1 SH row —
+    # the on-peak panel half cannot ride Main 1's off-peak storage rate (§1684).
+    # This is the same per-system resolution Main 1 and a non-electric Main 2
+    # already use; the earlier "keep Main 1's scalar" deferral was a no-op on
+    # the HOMOGENEOUS cohort (storage+storage, direct+direct → equal rates) and
+    # only mis-billed the heterogeneous case, over-rating SAP (LRHA WAVE 3
+    # fixture 497655371974 storage+panel read +15.3 vs PasHub).
     from domain.sap10_calculator.tables.table_12a import Tariff
 
-    main_1 = MainHeatingDetail(
-        has_fghrs=False, main_fuel_type=29, heat_emitter_type=0,
-        emitter_temperature="NA", main_heating_control=2106,
-        main_heating_category=10, sap_main_heating_code=691,
+    main_1_storage = MainHeatingDetail(
+        has_fghrs=False, main_fuel_type=30, heat_emitter_type=0,
+        emitter_temperature="NA", main_heating_control=2401,
+        main_heating_category=10, sap_main_heating_code=402,
         main_heating_fraction=50,
     )
-    main_2 = MainHeatingDetail(
-        has_fghrs=False, main_fuel_type=29, heat_emitter_type=0,
-        emitter_temperature="NA", main_heating_control=2106,
+    main_2_direct = MainHeatingDetail(
+        has_fghrs=False, main_fuel_type=30, heat_emitter_type=0,
+        emitter_temperature="NA", main_heating_control=2602,
         main_heating_category=10, sap_main_heating_code=691,
         main_heating_fraction=50,
     )
+    epc = make_minimal_sap10_epc(
+        sap_building_parts=[make_building_part(construction_age_band="D")],
+        sap_heating=make_sap_heating(
+            main_heating_details=[main_1_storage, main_2_direct]
+        ),
+    )
+
+    # Act
+    main_2_rate = _main_2_space_heating_fuel_cost_gbp_per_kwh(
+        epc, Tariff.SEVEN_HOUR, SAP_10_2_SPEC_PRICES
+    )
+    main_2_own_rate = _space_heating_fuel_cost_gbp_per_kwh(
+        main_2_direct, Tariff.SEVEN_HOUR, SAP_10_2_SPEC_PRICES
+    )
+    main_1_storage_rate = _space_heating_fuel_cost_gbp_per_kwh(
+        main_1_storage, Tariff.SEVEN_HOUR, SAP_10_2_SPEC_PRICES
+    )
+
+    # Assert — Main 2 bills at its own direct-acting on-peak rate (15.29 p),
+    # NOT Main 1's storage low rate (5.5 p).
+    assert abs(main_2_rate - main_2_own_rate) <= 1e-9
+    assert abs(main_2_rate - 0.1529) <= 1e-4
+    assert main_2_rate > main_1_storage_rate + 0.09
+
+
+def test_off_peak_main_2_electric_homogeneous_storage_unchanged() -> None:
+    # Arrange — HOMOGENEOUS storage+storage on a 7-hour tariff. Per-system
+    # billing is a no-op here (both mains resolve to the same off-peak storage
+    # rate), which is why flipping the deferral (§1684) leaves the off-peak
+    # electric cohort untouched. Locks that invariant.
+    from domain.sap10_calculator.tables.table_12a import Tariff
+
+    def _storage() -> MainHeatingDetail:
+        return MainHeatingDetail(
+            has_fghrs=False, main_fuel_type=30, heat_emitter_type=0,
+            emitter_temperature="NA", main_heating_control=2401,
+            main_heating_category=10, sap_main_heating_code=402,
+            main_heating_fraction=50,
+        )
+
+    main_1 = _storage()
+    main_2 = _storage()
     epc = make_minimal_sap10_epc(
         sap_building_parts=[make_building_part(construction_age_band="D")],
         sap_heating=make_sap_heating(main_heating_details=[main_1, main_2]),
     )
 
-    # Act — main 2's rate equals main 1's space-heating scalar.
+    # Act
     main_2_rate = _main_2_space_heating_fuel_cost_gbp_per_kwh(
         epc, Tariff.SEVEN_HOUR, SAP_10_2_SPEC_PRICES
     )
@@ -652,7 +702,7 @@ def test_off_peak_main_2_electric_keeps_main_1_scalar() -> None:
         main_1, Tariff.SEVEN_HOUR, SAP_10_2_SPEC_PRICES
     )
 
-    # Assert
+    # Assert — identical storage rates; the flip does not disturb this cohort.
     assert abs(main_2_rate - main_1_rate) <= 1e-9