Model/domain/sap10_calculator/docs/HANDOVER_POST_S0380_140.md

Handover — post Slices S0380.138..140

Branch: feature/per-cert-mapper-validation. HEAD 068088bc. Predecessor: HANDOVER_POST_S0380_137.md.

TL;DR

Three slices landed on top of 3542186f this session, all concentrated on the §10a fuel-cost + §4 cylinder-storage-loss cascades. Each slice surfaced 1-2 spec-citable bugs hidden behind silent default fallbacks; together they shifted ~24 corpus residuals substantially.

Slice	Commit	Scope
S0380.138	a830e855	New _off_peak_low_rate_gbp_per_kwh(tariff) helper routing every off-peak callsite (_space_heating_fuel_cost_gbp_per_kwh, _hot_water_fuel_cost_gbp_per_kwh, _secondary_fuel_cost_gbp_per_kwh, _pv_dwelling_import_price_gbp_per_kwh) through the per-tariff Table 32 low-rate (codes 31/33/35/40) instead of hardcoded prices.e7_low_rate_p_per_kwh = 5.50. Companion _off_peak_low_rate_gbp_per_kwh_via_meter_heuristic covers the Unknown-meter path. PriceTable.e7_low_rate_p_per_kwh field deleted (dead code).
S0380.139	c4db37db	_is_off_peak_meter routed through canonical tariff_from_meter_type (the bare "18 Hour" lodging — all 41 corpus variants' surface form — is now recognised as off-peak; pre-slice only the long form "off-peak 18 hour" matched). Dead _RDSAP_DEFINITELY_OFF_PEAK frozenset deleted.
S0380.140	068088bc	§4 (56)m cylinder storage loss cascade closed via two compounding fixes: (a) extractor parses §16 "Cylinder thermostat (Already installed)" recommendation line (previously only §15.1 "Cylinder Thermostat" label was checked, so cylinder_thermostat=None for every variant on property 001431); (b) _separately_timed_dhw now excludes electric immersion per SAP 10.2 Table 2b note b (which restricts the ×0.9 multiplier to "boiler systems, warm air systems and heat pump systems"). Combined, cascade TF closes from 0.702 → 0.60 (matching worksheet); HW kWh closes to ±1e-3 of worksheet (2384.116 vs 2384.12).

Extended handover suite at HEAD: 883 pass, 0 fail.

Current residual state at HEAD 068088bc

Cascade-OK tier (25 variants on pin grid)

Variant	ΔSAP_c	Δcost	ΔPE
ashp	+0.24	-£5.57	-12
electric 1	-0.06	+£1.32	+94
electric 2	+0.47	-£10.92	+101
electric 3	+2.55	-£58.65	-1122
electric 5	+0.07	-£1.72	-161
electric 6	+1.33	-£30.60	-563
electric 7	+1.29	-£29.73	-498
electric 8	-0.26	+£5.92	+126
electric 9	-0.12	+£2.72	+91
gshp	+1.15	-£26.48	-455
oil 1	+2.66	-£61.24	-1050
oil pcdb 1/2	+0.42	-£9.77	-84
oil pcdb 3	+1.16	-£26.72	-271
pcdb 1	+6.95	-£157.61	-3135
solid fuel 2	+2.55	-£60.79	-1211
solid fuel 3	+1.24	-£28.31	-935
solid fuel 4-11 (×8)	-0.29..+0.10	small	±100

Blocked tier (16 variants)

Unchanged from previous handover — community heating × 5, electric storage 11/12/13/14, no system, oil 2-6, pcdb 3.

Next-slice candidates ranked by leverage

1. +2.5 SAP cluster (electric 3, oil 1, solid fuel 2) — open, HETEROGENEOUS

These three variants share ΔSAP_c ≈ +2.55 with Δcost ≈ −£60, but probing shows the residuals trace to DIFFERENT causes — not a single shared cascade gap. Slice attempted in this thread, abandoned after diagnosis showed each variant needs a different fix:

variant	SAP code	SH+Sec demand gap	HW kWh gap	Driver
electric 3	401	cascade UNDER by 1005 kWh	exact	§9 MIT for storage heaters
oil 1	127	cascade OVER by 104 kWh (small)	cascade UNDER by 854 kWh	HW efficiency 86% vs worksheet 65%
solid fuel 2	158	cascade OVER by 337 kWh	unclear (different lodging)	TBD

Combined-R hypothesis tested and rejected. SAP 10.2 §9b defines effective_R = (1−sec_frac) × R_main + sec_frac × R_sec, which the cascade is NOT applying (mean_internal_temperature_section_from_cert at L2543 and main orchestrator at L4490 both call mean_internal_temperature_monthly without passing secondary_fraction/secondary_responsiveness). A monkey-patch to inject these REGRESSES electric 3 (+2.55 → +3.17) because raising effective_R LOWERS cascade demand — the wrong direction. The cascade is "compensating by not using combined R" — fixing this in isolation will require fixing the demand undercount simultaneously.

Per-variant fix plans:

• electric 3: cascade total useful demand (10046 kWh) is ~10% below worksheet (11088 kWh). Not driven by R alone — even with combined-R fix it gets worse. Likely the cascade's Table 9 heating- hours-per-day for ctrl=3 storage heaters differs from worksheet (worksheet may assume 24-hr "always on" for off-peak storage; cascade uses the standard ctrl=3 pattern).
• oil 1: cascade water_efficiency for Table 4b oil boiler (code 127, eff=84%) produces HW kWh 2785; worksheet 3639. Worksheet effective HW eff ≈ 65% (suggests summer/winter blend or different Appendix D path). Per SAP 10.2 Appendix D §D2.1 — the cascade may not apply the right summer-eff override for non-PCDB oil boilers.
• solid fuel 2: anthracite (eff=65%). Cascade HW is 3599 kWh; worksheet HW likely lodged in a different line ref (the probe regex returned 0). Re-probe needed.

Recommended approach: take these as 3 separate per-variant slices in a fresh session. Each is its own diagnosis. The "+2.5" magnitude similarity is coincidence, not signal.

2. pcdb 1 PE -3135 (single variant, biggest residual) — open

Diagnosed during this session: cascade water_efficiency for PCDB boiler 716 (Potterton KOA 90/26) uses summer efficiency 53% → HW kWh 4269. Worksheet effective HW efficiency ≈ 34% → HW kWh 7064.

The diff is +2794 kWh HW × 5.44 p/kWh = +£152 cost gap.

Likely root: PCDB Appendix D §D2.1 Equation D1 monthly cascade isn't being invoked for this record. The record has both winter (65%) and summer (53%) but the cascade may default to summer scalar. The spec wants monthly weighted blend via Eq D1.

Suggested slice plan:

1. Check why Eq D1 monthly cascade isn't firing for PCDB 716
2. Spec citation: SAP 10.2 Appendix D §D2.1
3. Fix the dispatch + re-pin pcdb 1

3. solid fuel 2/3 PE -935..-1211 — open

Both anthracite. Same fuel + R as variants that closed (solid fuel 4-11 all ±170 PE). Distinct cause from #1 above.

Different main_heating_efficiency per cert lodgement? Different main_heating_control? Per-variant probe required.

4. Lighting/pumps rate 13.19 vs 13.67 on 18-hour — uniform but tiny

Worksheet bills lighting/pumps at 18-hour HIGH rate (Table 32 code 38 = 13.67 p/kWh). Cascade falls back to standard 13.19 because Table 12a Grid 2 has no EIGHTEEN_HOUR row.

Fix: add (OtherUse.ALL_OTHER_USES, Tariff.EIGHTEEN_HOUR): 1.0 to _OTHER_USE_HIGH_RATE_FRACTION. Empirical citation (no spec PDF verification).

Impact: ~+£2 cost / -0.086 SAP per variant × 25 variants = uniform small shift. Doesn't strongly close any single variant but cleans up cohort-wide.

5. Pumps overcount for electric storage — wrong direction

Cascade applies 130 kWh pumps default for any non-gas/non-HP main. Worksheet has 0 pumps for electric storage / underfloor / direct- acting (no wet pump). But the cascade is already UNDER-counting cost for these variants, so removing pumps would make residuals MORE negative. Defer until SH+Sec demand cluster (#1) closes.

6. Community heating unblocking (5 variants) — sizeable

Extend extractor to capture §14.1 Community Heating block (heat-network codes 41-58). Each cert maps to a Table 32 heat-network code via the lodged heat source type.

7. Electric storage unblocking (variants 11-14) — small

Extend _ELMHURST_MAIN_HEATING_EES_TO_FUEL_CODE for EES codes WEA, REA, OEA.

Standard slice workflow

1. Read spec page + identify rule
2. Probe one cluster variant; verify diagnosis via monkey-patch
3. Write failing AAA test (literal # Arrange / # Act / # Assert)
4. Implement helper / dispatch entry / mapper extension
5. Re-pin affected variants
6. Run extended handover suite (command in previous handover)
7. Pyright net-zero check (git stash → pyright → git stash pop → pyright)
8. Commit with spec citation + Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
9. Update project-heating-systems-corpus + MEMORY.md index

Memories to load (in order)

project-heating-systems-corpus     # HEAD 068088bc
feedback-sap-10-2-only-never-10-3  # CRITICAL — never reference SAP 10.3
feedback-worksheet-not-api-reference
feedback-spec-citation-in-commits
feedback-verify-handover-claims
feedback-zero-error-strict
feedback-commit-per-slice
feedback-aaa-test-convention
feedback-e2e-validation-philosophy
feedback-abs-diff-over-pytest-approx
feedback-spec-floor-skepticism
feedback-golden-residuals-near-zero
feedback-one-e-minus-4-across-the-board
reference-unmapped-sap-code
reference-unmapped-api-code
project-oil-price-spec-divergence

What NOT to do

• Don't reference SAP 10.3 — track 10.2 deliberately
• Don't widen pin tolerances to make pins pass — re-pin smaller or find the spec gap
• Don't re-investigate Slices .91..140 — all settled
• Don't add new helpers to domain/sap10_ml/ — on deprecation path
• Don't accept "spec-precision floor" framing without spec-citation verification (the +2.5 SAP cluster is NOT a precision floor; it's a diagnosable shared cascade gap)

Spec source quick-reference

All under domain/sap10_calculator/docs/specs/:

• SAP 10.2 full spec: sap-10-2-full-specification-2025-03-14.pdf
  • §4 (p.135-137) — water heating worksheet (45..65), Tables 2/2a/2b
  • §9 (p.155+) — MIT calc, Tables 9/9a/9b
  • Table 4a/4b/4d — heating systems + emitter responsiveness
  • Table 4f (p.174) — pumps + fans
  • Table 11 — secondary heating fraction by category
  • Table 12 (p.191) — SAP rating fuel prices
  • Table 12a (p.191) — high/low-rate fraction by system × tariff
• RdSAP 10 spec: RdSAP 10 Specification 10-06-2025.pdf
  • §19 Table 32 (p.95) — RdSAP10 fuel prices / CO2 / PE

Good luck.