Three silent-drop extraction fixes the gov-API/Elmhurst mappers already apply
but `from_site_notes` uniquely omitted — each let a favorable default reach the
SAP-10.2 calculator and over-rate the dwelling. Same calculator, so the whole
gap was extraction fidelity (gov-API corpus 78.6% within-0.5 vs pashub 23.9%).
#1598 — electric-shower + bath counts: `_map_sap_heating` built `shower_outlets`
but never set `electric_shower_count`/`mixer_shower_count`/`number_baths`, so
the cascade defaulted to 1 gas mixer / 0 electric, billing electric-shower hot
water off the cheap main system. 100/205 fixtures lodge "Electric Shower".
#1599 — floor label: pashub lodges "Ground Floor" (capital F); the calculator's
RdSAP §5(12) suspended-timber floor-infiltration gate is case-sensitive on
"Ground floor", so the 0.2-ACH term was silently dropped. Normalized at the
mapper boundary. 98/205 suspended-timber.
#1600 — water-heating "None": the RdSAP 10 §10.7 "no water heating system"
signal was discarded and HW inherited the gas combi (58 Hackle 67.7 vs
verified 59). Coded to WHC 999 so the existing electric-immersion + Table
28/29 default fires (also folds in #1594: blank cylinder size is no cylinder,
not a phantom one).
Guinness GMCA 205 cohort: within-0.5 23.9% -> 50.2% (103/205), MAE 1.410 ->
0.650. Ratchets tightened to 0.50 / 0.65. Verified vs Khalim's ground truth
(16 Stillwater 78.2 -> 75.7 vs 76). #1601 (upper-floor +0.25 m) held — it
empirically over-corrects this cohort (internal-vs-external dimension question,
under investigation). pyright: 0 new errors.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Read stored lodged EPCs whatever refetch_epc says, so a stored assessment
can win the ADR-0001 Recency Tie-Break against the gov register.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Backs out the WHC-950/951/952 primary-loss (d3c0b9e4) and flat-rate-charging
(fecf4f62) changes to cert_to_inputs.py: they fire on ANY cert with those
water-heating codes, regressing the gov-API RdSAP corpus
(test_api_path_sap_accuracy_on_rdsap_21_0_1_corpus MAE 0.627 -> 0.630), and an
Opus audit found the water-only branch has a pre-existing plant-efficiency
double count they flow more kWh through. The calculator work + the correct
fix (water_eff = 1.0/(DLF*1.05)) + an Elmhurst-worksheet regression are
tracked in #1592, isolated from the extractor accuracy PR. The cylinder-
insulation MAPPER fix (f5e4814b) is retained.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Khalim-authorised MAE loosening: coding window orientation exposes the known
SAP-10.2-vs-lodged offset the zeroed solar gains were masking; within-0.5 is
the north-star metric and jumps 20.0% -> 23.9%.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Manually verified in pashub (Khalim, 2026-07-14): deal 499584755922
(9 Council Houses, CW5 8AP) is F33, not the stale D67 — our computed 34.0
was right within a point; deal 499617935574 (Brightholme, 14 North Road)
is the accredited E43, not the preliminary 53. 58 Hackle Street (D58) and
16 Bingley Close (E52) were confirmed correct, so our +7.8/+5.7 deviations
there are extraction bugs to hunt next. Harness MAE 1.684 -> 1.551; ceiling
ratcheted to 1.56.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The extractor now captures "Heating System (Other):" and the mapper resolves
it to the Table 4a code (301, community boilers) + Table 12 community fuel
(51 via the shared _resolve_community_heating_fuel_code), so
is_heat_network_main routes the DLF / standing-charge / 80%-efficiency
branch instead of pricing the dwelling as an ordinary mains-gas boiler
(fixture 507644414148: +14.3 over-rate closes by ~8.6). Harness MAE
1.726 -> 1.684; ceiling ratcheted to 1.70.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two dig-confirmed defects in the fresh RIR mapping (499617935574 overshot
-9): the roof-space-level insulation (Joists/200mm) was never threaded into
the RIR surfaces, so slopes/commons/flat ceilings fell to the Table 18
"unknown" full-uninsulated 2.30 default; and common walls were mislabeled
"common_wall" (a Simplified-Type-2-only concept) instead of the Detailed-mode
stud_wall every worksheet-validated fixture uses. 44.0 -> ~50.3 on the target
fixture; harness MAE 1.748 -> 1.726.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
