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>
_parse_room_in_roof captures the RdSAP §3.10 block (age range, floor area,
gables with Exposed/Party type, slopes, common walls, flat ceiling as
length x height) on main and extension roof spaces; both part builders now
carry it to sap_room_in_roof (issue #1590 bug 5 — closes the last #1590
subtask). Harness MAE 1.748 (754917 corrects to pashub's own pre-SAP; RIR
surfaces now billed per-surface instead of as insulated loft).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The surveyed "Ventilation type" label never reached
mechanical_ventilation_kind, so ventilation_from_cert silently defaulted every
mechanical system to NATURAL (24d), dropping its ACH heat loss — 76/205
cohort fixtures lodge dMEV/MEV/PIV (issue #1590 bug 4). Mapping mirrors the
gov-API _API_MECHANICAL_VENTILATION_TO_KIND table (PIV-from-loft is
"as natural"). Harness: within-0.5 12.2% -> 18.5%, MAE 2.538 -> 1.730 — the
campaign's largest single move; floors ratcheted to 0.18 / 1.75.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"Not connected to electricity meter" -> 1 (zero Appendix M credit), "Connected
to dwellings electricity meter" -> 2; the raw string fell through the credit
gate's non-int branch to True, crediting a separately-metered array (issue
#1590 bug 3, fixture 499516101839 -6.9 SAP). Harness MAE 2.571 -> 2.538;
ceiling ratcheted to 2.55.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"Photovoltaic array kWp Known? Yes" lodges a per-system block (kWp, pitch,
orientation, overshading) instead of the percent-roof estimate; the extractor
only read the estimate, so measured PV never reached the mapper and the
Appendix M credit was silently zero (issue #1590 bug 1). Harness MAE
2.632 -> 2.571 (fixture 507639151843's 13.5-SAP under-rate closes to +0.5);
ceiling ratcheted to 2.58.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"Insulation At: None" is the assessor recording zero loft insulation, not an
unknown — with thickness left None, u_roof fell through to the age-band
"assumed insulated" default (0.16 W/m²K at band J/K) instead of the Table 16
uninsulated row (2.30), understating roof loss ~14x (issue #1590 bug 2).
Harness MAE 2.701 -> 2.632 (the two ~7-SAP roof over-raters correct exactly);
ceiling ratcheted to 2.64.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two hubspot_deal_data corrections: 499539910905 D58->B82 (confirmed bad row)
and one C73->C75. Rebuilt via scripts/build_pashub_accuracy_fixtures.py
(no --force; PDFs unchanged).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Secondary fuel 'Mains gas'/'Mains Gas'->26 and 'House Coal'->11 (Table 32);
main-heating control resolution becomes group-aware with a Table 4e Group 3
map ('Charging system linked to use of community heating, room thermostat
only' -> 2306, the code the accredited Elmhurst certs lodge for this cohort;
all four linked-to-use codes are calculation-identical). All 205 Guinness
fixtures now compute (was 201/4 xfails); ratchet re-baselined to the
full-coverage aggregate (12.2% within-0.5 floor, MAE ceiling 2.71) with the
two newly-computable outliers documented for verification.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The control gap surfaces as UnmappedSapCode on main and as UnmappedPasHubLabel
once the #1557 control mapper lands (unmapped community-heating control), so the
harness must swallow both; only the closed fuel-specific MissingMainFuelType is
dropped as a fuel-drop tripwire.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Foundation for PRD #1555: runs each PAS Hub site-note PDF through the
extractor -> EpcPropertyData -> Sap10Calculator and gauges the computed
SAP against pashub's own SAP-10.2 `pre_sap` (from hubspot_deal_data).
- test_pashub_sap_accuracy.py: hybrid gate. Per-fixture "must compute"
(xfail on the known in-progress mapper gaps MissingMainFuelType /
UnmappedSapCode / UnmappedPasHubLabel) + aggregate within-0.5 ratchet
floor, mirroring test_sap_accuracy_corpus.py.
- 205 image-stripped site-note PDFs + manifest.json. Images stripped so
the repo footprint stays ~52MB while the text layer the extractor reads
is byte-identical.
- build_pashub_accuracy_fixtures.py: provenance/rebuild from S3 +
hubspot_deal_data.
All 206 currently xfail on the known heating-string mapper gaps; each fix
(#1556-1568) flips its fixtures to computing and ratchets the floor.
Refs #1555#1568
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Thread the UPRN the PashubService has already resolved for a job into the
PasHub site-notes mapper so the EpcPropertyData aggregate is born with its
uprn set, and the existing save_epc_property_data path persists it.
- EpcPropertyDataMapper.from_site_notes gains uprn: Optional[int] = None and
sets it unconditionally (site notes never carry a UPRN natively).
- parse_site_notes_pdf / _parse_pashub forward the uprn; the Elmhurst branch
is untouched.
- PashubService coerces uprn str -> int in one place, carries it on the
internal upload record, and passes it into parse_site_notes_pdf.
- No new lookups; jobs with no known UPRN still persist null, unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Both flagged mismatches were Elmhurst input errors (same silent-stale-value
contamination class as the earlier chimneys/wall-thickness bugs), not
parser or calculator bugs:
- cylinder_size: build_100010086084.py's water_heating() selected
DropDownListCylinderSize by raw value "2", but this DOM's option values
ARE their visible litre-band text (no "2" among them) — Playwright
silently no-ops on a non-matching value, leaving a prior cert's "Medium
(131-170L)" selection in place. Fixed to match by text ("Normal"), and
added the missing mapper dict entries (Normal/Large litre-suffixed
labels) to _ELMHURST_CYLINDER_SIZE_LABEL_TO_SAP10 (a real mapper-coverage
gap — the calculator raises UnmappedElmhurstLabel rather than silently
mis-mapping).
- boiler_flue_type: the boiler-code search dialog's combined "Balanced/Open
Flue" Table 4b category doesn't drive the separate, independently
selectable RadioButtonListFlueType field, which was left at an inherited
"Balanced" from a prior cert. Fixed space_heating() to explicitly select
"Open" to match the lodged gov-API code. This field isn't consumed by
Sap10Calculator (ML/generator-only), so it had zero effect on the SAP
score.
Re-downloaded elmhurst_summary.pdf/elmhurst_worksheet.pdf after the fixes
(Recommendations page confirmed clean). Elmhurst-PDF-inputs path moved
46->47 SAP from the cylinder-volume correction; gov-API SAP (53) and
Elmhurst's own worksheet (51) are unchanged. Full accuracy suites re-run
clean (67 passed, 67 skipped, 1 xfailed, same as before).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The Elmhurst rebuild for this cert had two carryover bugs from the shared
Playwright session: TextBoxOpenChimneys silently inherited "2" from a prior
cert's build (this cert's register lodges open_fireplaces_count=0, which
the gov-API mapper already reads correctly), and wall thickness inherited
a stale 280mm. Also fixed a real bug in the space_heating() PCDB-ref-clearing
logic: input_value() returns the string "0" when already cleared, which is
truthy in Python, causing a spurious navigate-away that broke the next
E.goto() call.
Rebuilt cleanly (chimneys=0, wall thickness=220mm per RdSAP convention for
unrecorded solid-wall thickness, since CheckBoxWallThicknessUnknown does not
persist via Playwright in this tool version). Elmhurst's own worksheet score
moves from a contaminated 46/48 to 51, closing the gap against this engine's
53-on-gov-API-inputs from ~5-7pt to ~2pt. No mapper.py change was needed --
the gov-API mapper's open_chimneys_count=0 was correct all along.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Built UPRN 100010086084 (the largest gov-API-vs-lodged outlier from the
20-cert batch) in Elmhurst's accredited RdSAP10 tool to get real ground
truth, per feedback that lodged rating isn't a valid comparison target.
Elmhurst worksheet: SAP 46-48 (orientation-dependent), vs this engine's
53 on gov-API inputs -- a real ~5-7pt residual gap. Ruled out with
direct empirical tests: total_floor_area, missing party_wall_length,
region/climate (SAP rating always uses UK-average weather per Appendix
U -- confirmed inert), boiler/secondary efficiency (exact match, 66%/
63% both sides). Confirmed real but NOT blind-fixed (need more than one
cert's evidence, and a wall-U-value patch tested the wrong direction in
isolation): a wall-thickness-unknown U-value fallback mismatch (this
engine 1.70, Elmhurst's own "unknown" resolution 1.40) and a window
U-value divergence (1.85 vs 2.52 W/m2K for the same nominal glazing
pick). Single-window orientation confirmed a ~2pt partial contributor.
Also fixes a small, separate, evidenced gap the build surfaced: the
Elmhurst summary parser didn't recognise the live tool's "Medium
(131-170L)" cylinder-size label (only the bare "Medium").
Full write-up is in the RealCertExpectation comment for this cert in
test_real_cert_sap_accuracy.py and the worklist. Residual gap is
intentionally left open for follow-up, not force-closed.
Sampled all 20 unique property_ids that failed with "Unsupported EPC
schema: 'SAP-Schema-15.0'" across the sub_task history (114 total across
72 subtask rows), captured 20 real live certs, and pinned each as a
RealCertExpectation regression in the accuracy corpus.
16/20 land exact or within 1-3 SAP of the lodged rating, confirming the
door_count/WWHRS/FGHRS/draught-proofing defaults generalise beyond the
original two properties. The other 4 (all solid-brick, uninsulated,
age band C) diverge +6 to +12 — traced to a legitimate SAP-2009-vintage-
lodgement vs RdSAP10-recalculation methodology gap (u_wall() checked
directly against RdSAP10's own Table 6, returns the correct 1.70 — no
mapper defect), not a mapper bug, so pinned to the engine's observed
value rather than tuned to lodged.
Sync Planning Authority, Designated Area, Article 4 PD Rights, and
Listed Building from HubSpot into hubspot_deal_data so downstream
consumers of the existing HubSpot ETL process can read them.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Built the property-753950 cert (worklist P4, PR #1503) in the live RdSAP-10
Online tool via Playwright: SAP 44 (E), vs our engine's 42 and the lodged
2014 cert's 45 — a normal RdSAP-2012->SAP-10.2 recalculation spread, not a
defect surfaced by the multiple_glazed_proportion fix.
Also surfaced (and documented as a caveat, not fixed here) a pre-existing
SAP-16.2 gap: the cert lodges party_wall_length_m=0 on both floors despite
being semi-detached, which Elmhurst's validation rejects and this PR's
engine change doesn't touch.
uploaded_files has no property_id — matching is on landlord_property_id — so the
property_ids path was pointless indirection (look property.id up in property just
to translate back to landlord_property_id). The FE holds landlord_property_id per
row anyway.
task.inputs hand-pick key is now landlord_property_ids: str[] (was property_ids:
int[]). resolve_selection unions two landlord_property_id sources — project_codes
expanded via hubspot_deal_data, and the hand-picked ids taken as given — and no
longer queries the property table at all. Trigger-only change; the domain plan,
orchestrator and matching already work in landlord_property_id.
ADR-0060, CONTEXT.md and the request schema updated to match.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Property 753950 (uprn 100021969385, cert 0141-2860-6891-9124-5625,
SAP-Schema-16.2) hard-failed modelling_e2e: RdSapSchema17_1 requires
multiple_glazed_proportion as a non-optional int, and this 16.x cert omits
the field entirely, lodging only the multiple_glazing_type="ND" sentinel.
The cert's own window.description ("Fully double glazed") states the
glazing extent unambiguously, so _normalize_sap_schema_16_x now derives
0/100 from "single"/"double" wording when the field is absent, mirroring
the existing single-glazed multiple_glazing_type cascade. This is
worklist P4 (.claude/skills/expand-sap-accuracy-corpus/worklist.md) —
a flat default was tried previously and reverted because making such
certs mappable at all pulls them into the EPC-prediction donor pool and
tips near-tie similarity matches; deriving from explicit text (rather
than a blind default) was the suggested unblock.
Re-measured the component-accuracy gate as the worklist asked: it drops
(has_hot_water_cylinder 0.8687->0.8586, cylinder_insulation_type
0.3333->0.1667, door_count residual 0.3131->0.3333) via the same
donor-pool-composition mechanism as the prior #1245/ADR-0037
re-baselines, not a prediction-logic loosening. Re-baselined the floors
with that rationale recorded inline.
The 'all properties' selection resolved against property.portfolio_id, but a
portfolio spans multiple HubSpot projects — so 'all' pulled the whole portfolio,
not the project the user was looking at. The project↔property grain lives on
hubspot_deal_data, not property.
task.inputs is now {project_codes?: str[], property_ids?: int[], portfolio_id?}:
the route resolves the distinct landlord_property_id set as the union of every
property in the named project_codes (from hubspot_deal_data) and the hand-picked
property_ids; portfolio_id is optional and only names the package. Drops the
portfolio-scoped select_all. Cap now applies to the resolved set size.
ADR-0060, CONTEXT.md and the request schema updated to match. New router tests
cover project-code resolution, the union+dedup, null landlord_property_id drop,
and the empty-project rejection.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Addresses PR #1498 review:
- A per-file read failure (missing/deleted object, or a bucket the role can't
reach) is now a SkippedDocument(reason=unreadable), not a whole-run abort;
an all-unreadable selection still fails (no empty package). Best-effort on
the read path (ADR-0060).
- Each ZIP member is streamed to a temp file (S3DocumentDownloader.download,
boto download_file) and added from disk, so a single multi-GB member never
hits the heap — the 'never held whole in memory' claim is now true.
- The 409 double-submit guard is DB-arbitrated: the sub_task insert is
conditional on the task having none, so a race creates only one.
- Local env gets a placeholder recipient email so the route is exercisable.
- PackageEntry carries landlord_property_id so a read failure can be reported.
- TODO noting the UploadedFile.s3_upload_timestamp typing fix.
Two new tests: per-file read failure skips-and-reports; address fallback flows
to the folder name.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
