Per the PR #1439 handoff: swept every cert now sitting at |Δ| in [1.0, 2.0)
(58 certs) using the cost-vs-demand signature bucketer (why_diverge.py's
logic) plus field_bias.py's full-corpus subgroup scan (re-run clean after
the two engine fixes above — no |meanΔ|>=0.5 systematic subgroup with
coherent median).
Breakdown: 17 room-in-roof (already spec-mandated/worksheet-validated per
PRs #1440/#1441), 7 electric cost/tariff variants (spec-faithful vs lodged,
consistent with the diagnosed electric cluster), 4 non-electric cost
residuals (spec-faithful), ~26 non-electric fabric/demand certs. Manually
inspected geometry (floor areas/perimeters/heights, wall/roof codes,
window/door counts) for the 4 largest fabric/demand outliers in the band
(100031768368, 100070450818, 100110097281, 100051051866) — no data-entry
anomaly comparable to the earlier 12109191 decimal-floor-area bug. This
confirms the handoff's "our calc faithful, lodged differs" pattern holds
across the band; no engine bug survived the pass. Notes appended per-cert
in worklist.md (`<-`, preserved by build_worklist.py reruns).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ventilation(): the cert lodges no extract-fan count and RdSAP 10 Table 5's
age-band B default is 0, but the build entered 1 — inflating Elmhurst's own
(8)/(21) infiltration lines relative to both the cert and our engine.
space_heating(): switched from the generic "Combi condens" SAP-Table dialog
code (84% efficiency, no PCDB record) to elmhurst_lib.set_pcdb_boiler(18908)
via clear_main_heating_code — resolves the exact lodged product (Worcester
Greenstar 4000, 88.70%) and its correct keep-hot classification (None),
matching the SAP engine's own PCDB lookup and removing the "Standard Combi"
keep-hot artifact the reused generic code carried.
Also corrected the file's stale docstring (described a different property
entirely — mid-terrace/250mm/DUAL meter/11 windows/lodged 63 — left over
from a copy-paste).
Verified against the refreshed worksheet: (21) shelter-adjusted ACH 0.6938
now EXACT vs our engine's ventilation_from_cert; (61) combi loss monthly
kWh matches ours to ~0.01 kWh (was carrying a ~360 kWh/yr keep-hot artifact).
Both open items from the PR #1439 handoff are resolved.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Cracks the 'PCDB disabled' blocker. The PCDF boiler reference resolves the exact
lodged boiler (e.g. 18908 -> Worcester Greenstar 4000, 88.70%) when: (1) the main
heating CODE is empty server-side (else the field + magnifying glass are
aspNetDisabled), and (2) the index is entered via REAL keyboard + Tab (a JS-set
.value doesn't survive the AutoPostBack). Now Elmhurst uses the same boiler
efficiency as the SAP engine, so worksheet vs calculator compare like-for-like.
Verified on 100061275133: Elmhurst now shows 'Efficiency of main space heating
system 1 = 88.70%' (was generic 84%). Includes updated worksheet evidence.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Third Playwright-entry error found via input reconciliation: my build left
draught-proofed doors at 0 (Elmhurst computed 33% draught proofing), but the
cert lodges percent_draughtproofed=100. Our calculator correctly uses 100% ->
infiltration 42.66 W/K; the ~19 W/K ventilation gap to Elmhurst was purely my
build error. On matched inputs our fabric+TB (109.38) equals Elmhurst's total
fabric (37) EXACTLY, confirming the calculator is RdSAP-faithful. Committed for
build-faithfulness record; the engine's +1.24 vs lodged 76 is not a bug.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Audits identity pass-through fields (dims, counts, thicknesses) raw cert vs
EpcPropertyData across all 1000 certs. All apparent mismatches are legitimate
mapper behavior the first-pass audit didn't model: window_wall_type=4 windows
correctly split to sap_roof_windows (not dropped); empty placeholder building
parts correctly discarded; 'measured' insulation-thickness sentinel resolved to
the wall_insulation_thickness_measured value. Remaining per-BP diffs were an
index-misalignment artifact (raw BP[i] vs epc BP[i] after empty BP0 dropped).
Conclusion: mapper is faithful; divergences are Playwright-entry or calc-side.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Input reconciliation (Elmhurst summary vs EpcPropertyData vs raw cert) exposed
two Playwright-entry errors: lighting entered as 6 incandescent (cert has 5 LED,
100% low-energy) and wall thickness left at Elmhurst's 260mm default (cert lodges
400mm). Fixing both moved Elmhurst 67->69. Confirmed the MAPPER is faithful
(every fabric/heating/lighting field matches the raw cert) and the engine uses
the correct PCDB 18908 efficiency (88.7%). The remaining Elmhurst gap to lodged
76 is the un-enterable PCDB boiler (automation falls back to generic BGW 84% +
keep-hot); the engine (77) is faithful. Not a mapping error.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Field-level bias scan (scripts/corpus_1000/field_bias.py) flagged solid-brick +
internal-insulation certs (wall_con+ins=(3,3)) systematically under-rating
(median -0.40, n=18). Root cause: _insulation_bucket routed insulation that is
KNOWN PRESENT but with a thin lodged thickness (5-24mm, below the smallest
tabulated 50mm column) to bucket 0 (uninsulated, U 1.7) - WORSE than the
present-but-unknown case, which the RdSAP Table 6 footnote correctly routes to
the 50mm row. A known 10mm shouldn't read as higher-U than not knowing at all.
Now present insulation never falls to bucket 0; thin/unknown thickness both
round up to the 50mm row. Only genuinely absent insulation maps to 0.
Corpus cert 5117280 (10mm internal, no wall thickness -> bucket path): Δ -4.44
-> +1.08. Corpus MAE 0.641 -> 0.637, PE MAE 3.0 -> 2.9. Gauge 77.7% holds; 44
RealCert pins + 342 fabric tests green. TDD: 2 new unit tests.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Added scripts/corpus_1000/demand_cohort.py (isolates fabric/demand-driven
divergences from cost noise). Investigation conclusion: the demand-side tail is
one-offs, not a cluster. 12109191 (+13.1) is a LODGED decimal error (365.67 m²
floor, 10x its ground floor). 217060420 (-4.4) is spec-faithful (36.6 m2 lodged
glazing at spec U). The only >=3 structural clusters among diverging certs are
electric flats (already diagnosed spec-faithful). No generic fabric bug remains;
gauge 77.7% is near its structural ceiling. Marked 12109191 unbuildable in ledger.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Electric-tariff cluster investigation. The 112-cert electric-main cohort is
mixed-sign within every category/meter - no systematic tariff bias. Root cause
of the biggest non-xfail divergences (10012334488 +13.2, 10091578598 +7.81):
SAP code 195 = 'Electric water storage boiler' (Table 4a p.170), which SAP 10.2
§12 Rule 2 correctly bills mostly at the 7-hour off-peak rate. Both are pure
cost gaps (PE/CO2 match lodged -> demand right); the lodged software over-billed
the storage boilers at peak rate. The engine is SPEC-FAITHFUL - matching lodged
would tune against §12/Table-4a. Pinned to the spec-faithful engine values.
Adds scripts/corpus_1000/elec_cohort.py (electric-main cohort analyzer).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2-storey semi, solid brick 400mm with EXTERNAL insulation, 300mm loft,
suspended uninsulated ground floor, gas combi, double glazing (glazed_type 13).
Built in accredited Elmhurst RdSAP10: roof (U 0.14, 5.459 W/K) and windows
(15.32 x eff U 1.3258 = 20.311 W/K) match the engine EXACTLY; the
externally-insulated wall is engine U 0.29 vs Elmhurst 0.30, validating the
slice-5 external-insulation default (100mm added over solid brick); suspended
ground floor engine 0.68 vs Elmhurst 0.70. Elmhurst's overall SAP 67 is not
comparable (reused-assessment Standard-Combi keep-hot loss, a build artifact).
Engine 77 vs lodged 76 (within 1.24). Pinned to the observed 77.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Top-floor flat, solid brick uninsulated, uninsulated flat roof, gas combi,
SINGLE glazing (glazed_type 5). Built in accredited Elmhurst RdSAP10 and every
heat-loss element matches the engine EXACTLY: walls 53.18, flat roof 145.57,
floor 0, party 0, and — with the faithful single glazing — windows 34.7114 W/K
(8.62 m2 x eff U 4.0268). Confirms the mapper's glazed_type 5 -> single (raw U
4.8) is spec-correct per epc_codes.csv, not the double it superficially
resembles. Engine 61 vs lodged 62; the -1.18 is a cost-side residual, not a
fabric bug. Pinned to the observed engine 61.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
SOLVED the window-grid blocker (per Junte's "keep one and edit, don't
delete all"): set_single_window now ADDS the fresh window (it inserts at
DOM index 0), then deletes the STALE rows BELOW it by index (high->low,
never index 0) via the per-row Delete image button + JS-clicked Yes modal.
The freshly-added row survives as the single window; no 0.00 junk strands
to block the Recommendations gate. Verified end-to-end: cert 100050881708
built, worksheet downloaded, walls (0.70) + roof (0.19) match the engine
EXACTLY. Combined with the earlier fill+Tab (area postback) + frame/gap
(validation) fixes, the per-cert loop now runs autonomously on clean
single-main certs.
Pin: 100050881708 (semi, cavity filled, pitched 225mm, suspended ground
floor, combi) engine 72 = lodged 70 +2; walls/roof accredited-Elmhurst-
matched. NB the build entered the suspended ground floor as an upper
"above unheated space" floor, so Elmhurst's worksheet SAP (61) is not a
clean comparison — the engine uses the correct §3.12 suspended-ground U
(0.74); the pin rests on the walls+roof match + lodged.
No engine change.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Continued hardening for the per-cert Elmhurst loop. Confirmed findings:
- The window Width/Height must be set via Playwright .fill() + Tab (fires
the ASP.NET AutoPostBack so the server commits the area); a JS-set of
.value bypasses the postback and Elmhurst persists a 0.00-area row.
- Frame Type (PVC) + Glazing Gap (12 mm) are REQUIRED per row or the
Recommendations gate blocks the worksheet ("Frame Type/Glazing Gap must
be entered"). add_combined_window now sets both.
KNOWN BLOCKER (documented for the next attempt): the window grid is not
yet reliably automatable when stale rows exist on the shared assessment.
New rows insert at index 0; the delete flow reliably removes only index 0
(the newest), so accumulated 0.00 junk rows BELOW cannot be targeted, and
grid reads are timing-inconsistent across postbacks. A robust handler
needs to either (a) drive the per-row Delete by matching the row's
server-side id rather than DOM index, or (b) start each build from a
freshly-created assessment (not the shared one) so the grid is empty.
Until then the per-cert loop runs only on the FIRST build after a manual
window reset. build_100050881708.py is the WIP template.
No engine change.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Support for the per-cert Elmhurst worksheet loop:
- goto()/save_close(): fall back networkidle -> domcontentloaded on
timeout so an incomplete-validation state (half-set RR, 0-area window)
no longer hangs navigation.
- add_combined_window(): select glazing first (AutoPostBack), then JS
input+change+blur the width/height with settle waits so the Area cell
computes before Add (a plain .fill() raced the postback -> 0.00 row).
- set_single_window(): leave the grid holding exactly one row, deleting
rows carried over from a prior build of the SHARED assessment (the
window grid is the only accumulating page).
- reset_transient_state(): return the shared assessment to neutral after
a run (clear RR age band, secondary = No, meter = Single).
KNOWN RESIDUAL: the window grid still intermittently persists a 0.00 area
through Save & Close (Elmhurst's "each Add wipes the previous row" +
last-row-undeletable behaviour), and the secondary-heating dialog's modal
background can intercept the cascade click. Reliable autonomous grinding of
per-cert builds needs these two hardened further; the loop currently
completes clean single-main gas certs (see the 47084930 flat-roof slice)
but stalls on secondary-heating + stale-window combinations. build_
100021924710.py is the WIP template for that harder shape.
No engine change.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A FLAT roof lodges its insulation thickness in the DEDICATED gov-EPC API
`flat_roof_insulation_thickness` field (e.g. "75mm"), leaving
`roof_insulation_thickness` None (that field is for pitched-loft joists).
`heat_transmission_from_cert` read only the latter, so a measured
flat-roof thickness was ignored and the roof billed at the uninsulated
age-band flat default (age E = 1.5) instead of its Table-16 insulated U.
Fixed by preferring `flat_roof_insulation_thickness` when the part is a
flat roof — the exact mirror of the existing rafter-thickness branch.
An "AB"/"NI" (as-built/unknown) value parses to None and keeps the
age-band default, unchanged; only measured thicknesses move.
PER-CERT ELMHURST VALIDATION (cert 47084930, top-floor flat, flat roof
75 mm): built on the lodged inputs in accredited Elmhurst RdSAP10-Online
(evidence saved: elmhurst_summary.pdf / elmhurst_worksheet.pdf). The
worksheet bills "insulated flat roof" at U 0.5 (floor 0.70 + wall 0.25
also matching the engine). The fix takes the engine roof 96.4 -> 32.1 W/K
(= 64.26 x 0.5, Elmhurst-exact), PE +47 -> +0.2, SAP 69.51 -> 74.34 =
lodged 74.
Gauge: within 77.3% -> 77.7%, SAP MAE 0.648 -> 0.641, CO2 0.074 -> 0.072,
PE 3.1 -> 2.97. Unit-pinned in test_heat_transmission
(flat_roof_insulation_thickness -> U 0.5); RealCertExpectation 47084930
= 74 (Elmhurst-validated). Also adds the build script build_47084930.py.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A hot-water cylinder lodging gov-API `cylinder_size=0` ("size not
determined", with has_hot_water_cylinder=true) resolved to a None volume
in `_cylinder_storage_loss_override` (which reads `_cylinder_volume_l_
from_code`, returning None for code 0). The subsequent `if volume_l is
None: return None` then DROPPED the cylinder's Table-2 storage loss
entirely — under-costing the DHW and over-rating the dwelling. Meanwhile
the Table-13 high-rate-fraction path already used `_hot_water_cylinder_
volume_l` (which defaults size 0 to 110 L), so the two DHW paths
disagreed on the same cylinder.
RdSAP 10 §10.5 Table 28 ("if the actual size is not determined, the size
is taken as according to Table 28") makes an unsized present cylinder the
110 L "Normal" baseline, which STILL incurs the storage loss. Fixed by
defaulting the storage-loss volume to 110 L for the explicit size-0 case.
Gated on the EXPLICIT 0 (not None): a full-SAP cert whose RdSAP
cylinder_size is simply unlodged (None, e.g. pinned 10091568921) keeps
its own cylinder handling rather than a forced 110 L RdSAP default.
7 corpus certs lodge cylinder_size=0 + has_cylinder=true. Gauge: within
77.0% -> 77.3%, SAP MAE ~flat (0.648), CO2 0.074 -> 0.073, PE 3.2 -> 3.05
(the previously-dropped storage loss now correctly counted in the demand
cascade). Unit-pinned in test_cert_to_inputs (storage loss non-None for
size 0); RealCertExpectation 200004017091 = 71.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Built the cert in accredited Elmhurst RdSAP10-Online on the lodged register
inputs; downloaded Input Summary + SAP Worksheets (committed alongside).
Confirmed:
- Slice 5 DIRECTION validated: the externally-insulated solid-brick wall is
billed insulated by both tools (Elmhurst U 0.48, engine U 0.29), far below
the uninsulated 1.40 the pre-fix §5.8 gate produced.
- ROOF matches EXACTLY (U 0.30, 9.35 W/K).
- Engine 71.74 = lodged 71; Elmhurst current SAP 66 (its "unknown external
insulation" default is more conservative + this build used a generic combi
and no keyed wall thickness, so 66 is not a clean engine-comparison).
Follow-up lead (documented in ELMHURST_FINDINGS.md): the engine feeds the
UNMEASURED lodged wall_thickness (360 mm, measured=N) into the §5.6 formula,
whereas RdSAP §5.3 uses measured thickness / Table-3 default otherwise. The
party-wall "difference" is NOT a bug (engine 0.25 is spec-correct, Table 15
row 4). Also adds the build script scripts/hyde/build_200004296092.py.
No engine change in this commit.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A SOLID-BRICK or STONE wall lodging External/Internal insulation
(wall_insulation_type 1/3) but no measured insulation thickness (gov-API
"NI" -> None) was billed at its UNINSULATED §5.6/Table-6 U-value: the §5.8
added-insulation R-value adjustment gates on `insulation_thickness_mm > 0`,
so an unknown thickness fell straight through to the raw solid-wall U
(~1.4-1.7). This over-counted wall heat loss and under-rated the dwelling.
RdSAP 10 §5.4 (PDF p.32): "the U-value with additional insulation is based
on the thickness of insulation of mineral wool type (assume 100 mm if
thickness is unknown)". Apply that 100 mm default so the §5.8 Table-14
R-value fires. Scoped to solid brick / stone (the only constructions with
a §5.8 R-value path in u_wall) so cavity (composite path) and timber /
system-built walls are untouched.
Localised by deep-diving corpus cert 200004296092 (end-terrace, solid
brick 360 mm "with external insulation", NI): walls 101.7 W/K (U~1.40,
HLP 3.46 = absurd for an insulated wall) -> ~0.29; PE +59.6 -> ~0; engine
64.35 (Δ -6.65) -> 71.74 = lodged 71. ~8 solid-brick/stone corpus certs
carry this NI-thickness-with-insulation shape.
Gauge: within-0.5 held at 77.0%, SAP MAE 0.658 -> 0.647, CO2 0.074 ->
0.073, PE 3.2 -> 3.1 (floor ratcheted). Unit-pinned in test_rdsap_uvalues
(brick 360 mm external NI -> 0.29) + RealCertExpectation 200004296092 = 72.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
An instantaneous point-of-use electric water heater (WHC 907/909, no
cylinder) on an off-peak tariff was billed 100% at the off-peak LOW rate
by the generic electric-off-peak else-branch. But SAP 10.2 §12 (PDF line
2680) computes the off-peak on-peak HW proportion via Table 13, "dependent
on the total floor area and the CYLINDER size" — it presumes a stored-water
cylinder charged overnight. An instantaneous heater has NO cylinder, heats
on demand, and cannot shift to the off-peak window, so 100% of its
consumption bills at the HIGH rate (Table 12a WH high-rate fraction 1.0).
Both the SAP-cost rate (`_hot_water_fuel_cost_gbp_per_kwh`) and the
ADR-0014 / CO2-PE fraction (`_hot_water_high_rate_fraction`) fixed
consistently (low-rate scalar -> 7-hour high rate; fraction 0.0 -> 1.0).
Localised by deep-diving corpus cert 74061136 (HHR-storage mid-floor flat,
WHC 909): PE matched lodged (+1.6, roof+floor zero-loss) while SAP over-rated
+7.72 — the cost-only signature. Its DHW was ours £59 vs lodged £344 (5.8x);
the tariff was the whole gap. Fix: +7.72 -> -1.25 (residual is separate small
fabric). 7 corpus certs carry electric-instantaneous DHW on off-peak; the 3
outside 0.5 all move sharply inward (MAE the win, not within-0.5 crossings).
RdSAP 10 §12 tariff routing confirmed spec-correct (Unknown meter + storage
409 -> off-peak 7-hour, Rule 2) — the bug was the DHW rate, not the tariff.
Unit-pinned in test_cert_to_inputs; RealCertExpectation 74061136 = 72.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The gov-API `secondary_fuel_type` enums 5 (anthracite) / 9 (dual fuel) /
33 (coal) collide in VALUE with same-valued RdSAP-10 Table-32 codes for
OTHER fuels (5=bulk LPG, 9=LPG SC11F, 33=elec 10h-low). The main-fuel
boundary and the CO2/display `_secondary_fuel_cost_gbp_per_kwh` already
canonicalise these via `canonical_fuel_code`, but the SAP-DRIVING
`_fuel_cost` cascade had a SEPARATE inline secondary rate that passed the
raw enum straight to the price lookup — so the two paths diverged and the
cost path silently mis-priced.
Anthracite (enum 5) was billed at the bulk-LPG code-5 rate 12.19 p/kWh
instead of 3.64 — a 3x over-cost on the secondary that under-rated every
solid-fuel-secondary dwelling. Worst: corpus cert 100050355518 (semi,
2 extensions, anthracite room-heater secondary) lodged 36 / engine 20.8
-> 35.5. Its CO2/PE already matched lodged (the demand cascade was right)
— a pure cost-side gap, which is what localised it.
Fix: canonicalise `secondary_fuel` before the Table-32 lookup in
`_fuel_cost`, mirroring the main-fuel boundary. 55 corpus certs carry a
colliding secondary.
Gauge: within-0.5 75.5% -> 77.0%, SAP MAE 0.708 -> 0.668 (floors
ratcheted). The prior enum-9 test only guarded the display helper; new
test_fuel_cost_secondary_colliding_fuel_priced_at_canonical_rate pins the
_fuel_cost cost path directly. RealCertExpectation pinned for
100050355518 (36 = lodged).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
RdSAP 10 §5.11.4: a pitched roof lodging thickness "NI" with description
"...insulated (assumed)" was billed at the §5.11.4 observed-retrofit
50 mm row (U 0.68) while the SAME description lodged with "ND" fell
through to the Table 18 age-band default (0.40 for bands A-G). The
ND-vs-NI sentinel is lodging-software noise; "(assumed)" means the
insulation PRESENCE is an age-band assumption, not an observed
retrofit, so Table 18's "thickness cannot be determined" clause
governs both forms. The bare "insulated" description (observed
retrofit, no qualifier) keeps the 50 mm row.
Evidence: the 27-cert loc4+NI corpus cohort was systematically
under-rated (~-1 SAP; zero positive movers among "(assumed)" certs),
vs the 102-cert loc4+ND cohort already on Table 18 at ~0 bias.
Gauge: within-0.5 74.2% -> 75.5%, SAP MAE 0.721 -> 0.708, CO2 MAE
0.09 -> 0.074, PE MAE 3.5 -> 3.2 (floors ratcheted). Unit-pinned in
test_rdsap_uvalues (assumed -> Table 18; bare-insulated -> 0.68
regression guard); integration pin in test_heat_transmission updated;
RealCertExpectation pinned for 10094975827 (66 = lodged, was -1.87).
Ledger: C002 103001004 deferred pending Elmhurst arbitration
(single-immersion off-peak pricing + declared cylinder loss vs
Table-2); C006 quadruplet marked ⚠ (LIG-21.0 software honours lodged
0 extract fans literally vs the Elmhurst-validated 0=unknown->default
convention, worksheet case 46).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Setup for the RdSAP-21.0.1 corpus gauge campaign (74.2% within-0.5,
SAP MAE 0.721 at HEAD, 1000/1000 computed):
- scripts/corpus_1000/build_worklist.py — runs all 1000 corpus certs
through the gauge's own path (from_api_response -> Sap10Calculator),
ranks |engine - lodged| descending, clusters by dwelling signature,
and (re)writes scripts/corpus_1000/worklist.md preserving per-cert
statuses and notes. 258 certs outside 0.5 in 249 clusters.
- C001 (worst cert, Δ +23.9): uprn 4510053280, ground-floor flat 47 m²
on ASHP PCDB 100053 (Mitsubishi Ecodan 5 kW, PSR table 0.2-2.0).
Dwelling PSR 2.031 -> SAP 10.2 N2 footnotes 44/45 heat-pump extension
(reciprocal interpolation toward 100% at 2x largest PSR) = 305%
space efficiency, which accredited Elmhurst also applies (golden
case 56, record 100061). The LODGED software instead treated the
out-of-range record as invalid and billed 100% direct electric +
standard schedule (reproduces lodged CO2 to 1%). Lodged-software
methodology gap, worklist ⚠, engine untouched; observed engine 75
pinned in test_real_cert_sap_accuracy.
Gauge floors unchanged (no engine change).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Portfolio 814 / scenario 1271 audit (2026-07-03): all 26 properties lodging
has_hot_water_cylinder with insulation type 0 (none) or NULL over-rate their
effective baseline by up to +18 SAP (avg +12.7) because
_cylinder_storage_loss_override returns None for any insulation label it
can't resolve and the worksheet keeps the zero-storage-loss combi default.
Same-lodged neighbours split by 19 effective points (WD5 0DP pids
742215/742216, both lodged 54, effective 50 vs 69) and the inflated members
distorted neighbour-divergence cohort medians (E17 6EB pid 742355 Δ+18).
New HIGH check cylinder-storage-loss-dropped fires on the deterministic
input pattern (cylinder present, insulation type ∉ {factory, loose jacket}
or thickness missing) — 28/338 on the motivating portfolio (the 26 above
plus 2 same-pattern rows with no lodged score). The calculator fix
(uninsulated Table 2 loss, or RdSAP Table 29 age-band default) is separate
follow-up work.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
39 Hyde-796 property_overrides rows ('100% secondary glazing (sap 9.94)')
plus the classifier-cache row; the cache phase self-gates on the
'Secondary glazing' pgEnum label landing (assessment-model#345).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Addresses PR #1402 review: the guard only claimed dominant-single, leaving
the symmetric bug open — the LLM can flatten a dominant-double/triple split
onto its minority single (e.g. "96% Double glazing 2002 or later, 4% Single").
A dominant double/triple whose era is explicit ("pre-2002" / "2002 or later")
is just as fully determined as era-free single, so the guard now claims it via
_DOMINANT_MEMBER. Only a genuinely ambiguous era ("unknown age", unstated) still
defers to the LLM — the "96% double -> None" contract now holds solely for the
era-unknown case, not the era-stated one.
Backfill script reuses the same guard, so it now corrects any dominant split;
renamed reclassify_dominant_single_glazing.py -> reclassify_dominant_glazing.py
to match. Tests cover double/triple x pre-2002/2002-or-later and the still-
deferred unknown-age case; 14 green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The landlord property-type description is a "<dwelling type>: <built form>: <floor>"
split whose leading token IS the dwelling type; the built-form tail is not. The
LLM occasionally over-read the tail and flipped the type — a handful of
"Bungalow: EndTerrace" / "Bungalow: MidTerrace" dwellings were stored as House.
Adds property_type_guard (claims the recognised leading token: House / Bungalow /
Flat / Maisonette / Park home; defers unrecognised phrasings to the LLM) and wires
property_type through a GuardedColumnClassifier, so the built-form tail can never
flip the type and the live path is deterministic.
Applied the scoped backfill to portfolio 796 (Hyde): 3 rows corrected from House
back to Bungalow. No enum migration needed — the targets are original members.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
`plan-stops-short-of-goal` fires on anything below the goal band — 322/338
on portfolio 814, mostly flats a single band short of B, which drowns the
real hits. Add a sharper, band-keyed check that isolates the genuinely-
stuck tail: a plan that leaves the dwelling still rated E/F/G with the band
UNCHANGED from baseline on an unlimited-budget scenario — the "E->E, no big
movement" worklist. On portfolio 814 / scenario 1271 it flags just 4 (pids
742121, 742210, 742265, 742347) vs 322. No SAP-points threshold; the
strictly-worse case is left to plan-below-baseline-band so the two
partition cleanly.
Also refresh the find-weird-recommendations skill's Phase 4 catalogue:
register the new check in Phase 1, correct the stale 742121 note (it is an
electric maisonette whose HHRSH IS offered but scores -6.3 SAP, so the
Optimiser correctly drops it — not the old mains-gas story), and add the
742265 community-flat findings (HHRSH/ASHP gated out by heat-network
topology; vaulted-ceiling roof with ND insulation).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
MainFuelType had no individual wood-logs member — only "biomass (community)" —
so the LLM classifier funnelled "Solid Fuel: Wood Logs" into the community fuel,
inventing a community heat network the dwelling isn't on (and mislabelling the
connection). main_fuel had no deterministic guard at all, so nothing caught it.
Verified against domain/sap10_calculator/docs/specs: RdSAP 10 Specification
Table 32 lists "wood logs" as a solid fuel (code 20, 0.028 kgCO2e/kWh); the
calculator's input scheme (the gov EPC API fuel enum) codes it 6 -> Table 32 20
(sap_efficiencies._API_TO_TABLE32), and water_heating_overlay already pins the
same fuel to 6. So _FUEL_CODES["wood logs"] = 6 is confirmed, not guessed.
Adds MainFuelType.WOOD_LOGS + the _FUEL_CODES entry, a main_fuel_guard mirroring
water_heating_guard (claims the "wood log" token; dual fuel keeps its own member
since it has no "wood log" substring), and wires main_fuel through a
GuardedColumnClassifier so the live path is deterministic.
Applied the scoped backfill to portfolio 796 (Hyde): 21 rows off
"biomass (community)" -> "wood logs". property_overrides (TEXT) only; the
classifier-cache pgEnum member is deferred to the FE Drizzle migration.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A landlord glazing split like "4% Double glazing 2002 or later, 96% Single
Glazing" is a near-uniform *single*-glazing dwelling, but the LLM classifier
latched onto the 4% minority and wrote a "Double glazing…" override — over-
crediting the dwelling.
glazing_mix_guard rescued genuine mixes (no type >= 90%) to MIXED but returned
None for near-uniform splits, trusting the LLM to have applied the dominant type.
It hadn't. Single glazing is era-free, so a dominant-single split is fully
determined: the guard now returns SINGLE for it (dominant double/triple still
carry era ambiguity, so they stay the LLM's job — the existing "96% double" case
still defers).
Adds a portfolio-scoped backfill (reuses the same guard, so it can't drift from
the live path) and applied it to portfolio 796 (Hyde): 115 glazing overrides
corrected from a spurious Double back to Single.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The neighbour SAP-divergence cohort was (postcode, property_type, built_form),
which mixed electric- and gas-heated dwellings. Electricity scores materially
lower in SAP than mains gas for identical fabric, so a mixed-fuel postcode
produced cross-fuel false outliers — an electric dwelling flagged only for being
electric among gas neighbours.
Add the main-heating fuel class to the cohort key (electricity variants — 29/30
+ off-peak 31-40 — collapse to one class; every other code is its own bucket).
Now every flagged divergence is a same-fuel comparison worth investigating. On
portfolio 814 this refined 20 raw divergences to 17 same-fuel ones while keeping
the genuine within-fuel outliers (e.g. the all-electric WC2B 4AW flats at SAP
26/38 vs a cohort median of 67).
Reaches the fuel via epc_main_heating_detail through the indexed property_id —
still never touches the recommendation table. Surfaced by the portfolio-814
recommendation audit (Work item B, #1388).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
