Add _apply_ventilation_mode: set the predicted mechanical_ventilation_kind to
the recency/geo-weighted cohort mode (mirrors _apply_glazing_mode — MEV/MVHR is
a new-build/retrofit feature clustering by era + street). Only the kind moves;
the template's sheltered_sides etc. stay. Natural cohorts mode to None and stay
natural (§2 default), so this only moves genuine MEV/MVHR neighbourhoods.
Display-only for the calc gate: component-accuracy (26) + corpus (6) + e2e (1)
all green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
_apply_heating_donor now also deepcopies the donor's main_heating and
main_heating_controls EnergyElements alongside its calc sap_heating cluster, so
the building-passport heating rows are coherent with the donated system and a
control row the size-template lacked (but the donor lodges) is populated. Fixes
'Heating Control: Unknown' on predicted properties (e.g. 721167). Display-only:
component-accuracy gate + corpus harness unchanged (26/6 green).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A predicted EPC is seeded by deep-copying one representative neighbour's
structure. _template chose the member whose floor area was closest to the
cohort median, ignoring building-part labels. When that member's only part
was lodged with a null identifier (mapped to OTHER), the prediction had no
MAIN part and the modelling_e2e handler rejected it as "not predictable" —
discarding an otherwise-rich same-type cohort.
Restrict the template to MAIN-bearing members (median still over the whole
cohort); fall back to closest-on-size only when none are MAIN-bearing, so an
all-unlabelled cohort is left for the handler's MAIN-part guard to reject
rather than silently relabelling real data.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A dwelling's heating is one conceptual system, but its fields are scattered
across EpcPropertyData (a gov-API schema mirror): the cluster on sap_heating, the
electricity tariff on sap_energy_source.meter_type, hot-water flags loose at top
level. Three places synthesise a heating system — Measure Options, Landlord
Overrides, EPC Prediction's donor — and each hand-copied a different ad-hoc
subset. The override and donor both dropped meter_type, so an electric-storage
system landed on the template's single-rate meter and billed overnight heat at
the peak rate: property 713406 scored SAP 13 (G) vs ~50 (E), inflating the HHRSH
measure to +45.8 and overshooting the plan to band A.
Establish a single Coherent Heating System boundary (CONTEXT.md) that every
synthesiser must cover, with a source-appropriate fill policy (ADR-0035):
- Override overlay *completes* the partial system the landlord named. Companion
fields are now DERIVED from the SAP code, not hand-attached per archetype: the
off-peak meter from the calculator's single off-peak classification (new
OFF_PEAK_IMPLYING_HEATING_CODES = SAP §12 Rules 1-2), and an unobserved storage
charge control defaults to the conservative manual control (Table 4e 2401). So
adding a heating archetype is just adding its code — companions can't be
forgotten. A contract test guards it (every off-peak code drags a Dual meter).
- Prediction's heating donor now *carries* the donor's meter_type alongside its
sap_heating cluster — the donor is already coherent.
Coherence is a synthesis-time obligation only; the calculator still scores a real
lodged cert exactly as lodged.
Verified on 713406: baseline 13 -> 47.8 (E), matching its recorded rating; the
phantom HHRSH recommendation is gone and the plan no longer overshoots to A.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two review points from @dancafc:
1) Rename the `Comparable` dataclass → `ComparableProperty` (it models one
comparable *property*; the collection stays `ComparableProperties`). Applied
across domain, repositories, orchestration, harness, scripts, and tests with a
word-boundary rename so `ComparableProperties` is untouched.
2) Move `PredictionTarget` out of comparable_properties.py into prediction_target.py
(where `PredictionTargetAttributes` + `build_prediction_target` already live).
comparable_properties.py now imports it; no import cycle (prediction_target no
longer depends on comparable_properties). Importers updated.
92 tests pass across the touched suites; pyright strict clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
README at domain/epc_prediction/README.md — the flow diagram, where each piece
lives, links to the ADRs/CONTEXT/handover/migration note, and a runnable test
command. The team's entry point.
tests/e2e/test_epc_prediction_e2e.py — the whole gap-fill flow against the REAL
Postgres Unit of Work + EPC/Property repositories + EpcComparablePropertiesRepository
+ EpcPrediction, with only the three external HTTP clients faked (EPC API,
geospatial S3, Solar). Proves: EPC-less Property → Ingestion predicts from its
postcode cohort → persists to the predicted slot → reloaded Property resolves
effective_epc via source_path == "predicted". The canonical "see it in action".
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
build_prediction_target assembles an EPC-less Property's PredictionTarget from
its identity (postcode), resolved coordinates, and Landlord-Override attributes
(property_type / built_form / wall_construction). The eligibility GATE: a Property
whose property_type is unknown returns None — never sized from a mixed-type
cohort (ADR-0031). property_type is the hard cohort filter.
The override attributes are read through a PredictionTargetAttributesReader port
(stub seam) — the real adapter (a read over property_overrides) is being built
separately by the team; ingestion wiring depends on the abstraction and tests
substitute a fake. 2 tests (assembly + gate); pyright strict clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Size the predicted dwelling from the geo-proximity-weighted median of the
cohort's floor areas rather than the plain median: homes built together share a
footprint, so a nearer neighbour's area should count for more (the same street
signal #1227 already wired into age / wall / glazing). Reuses `_geo_weights` and
adds `_weighted_median`, which reduces exactly to `statistics.median` under
uniform weights (geo off / no target coordinates) — including the even-count
midpoint average — so the MAD-minimising guarantee is preserved.
Measured over the 514-target SAP-10.2 corpus (leave-one-out):
floor_area MAE 10.48 -> 9.73 m² MAPE 13.2% -> 12.2%
Re-baselines the n=36 fixture floor_area ceiling 11.8983 -> 12.0378 (a method
change, not a loosening; the small fixture subset moved +0.14 the other way as
sample noise while the population improved decisively). The ceiling still pins
the new deterministic value exactly, so the tighten-only ratchet resumes.
Investigation ruling out the adjacent floor-area levers (kept in the follow-up):
lowering minimum_cohort (9.78-10.03, worse), hard same-form filter (10.19),
mean instead of median (10.68), constant bias correction (10.47),
extension-conditioning (oracle 9.50, not worth the misclassification cost) and
room-in-roof conditioning/additive (RiR is a confound for large multi-part
outliers — RiR area is only ~21% of total, and the increment breaks the homes
already predicted exactly). Remaining cohort lever is built-form soft-weighting,
gated on a denser corpus.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Folds a haversine distance kernel into the categorical-mode weighting so a
nearer neighbour counts for more — applied ONLY to the components that showed
a clear distance signal in the corpus pre-check (age band, wall + floor
construction, glazing: homes built/retrofitted together cluster). Roof
construction showed no decay and is excluded; heating keeps its coherent
donor. Predictor stays pure: weights come from target.coordinates vs each
Comparable.coordinates (resolved at the boundary); geo is OFF when the target
has no coords, neutral for a neighbour with none.
Scale chosen on the harness: _GEO_SCALE_KM=0.1 is the gate-safe optimum
(0.05 lifts the corpus more but regresses fixture floor_construction).
Corpus (150pc/514, geo off->on): age 0.564->0.572, age_pm1 0.841->0.847,
wall 0.902->0.912, floor_con 0.786->0.796, glazing 0.667->0.673; roof
unchanged. Fixture: glazing 0.5278->0.5833 (floor ratcheted), all else held.
Refactored recency into a reusable _recency_weights vector composed via
_combine, so similarity/recency/geo factors multiply uniformly. Fixture ships
a committed _coordinates.json (OGL OS OpenData; build script carries it from
the corpus sidecar on rebuild) so the gate exercises geo without S3.
This is the per-component method applied to geography ([[feedback_per_component_best_method]]).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds coordinates: Optional[Coordinates] to Comparable and PredictionTarget
(data carriers — the pure predictor stays IO-free), and wires load_corpus to
read an optional _coordinates.json sidecar ({uprn: [lon, lat]}) and populate
each Comparable from its cert's uprn; iter_predictions threads the held-out
target's coordinates through. Absent sidecar -> geo-weighting stays off (no
behaviour change yet — weighting lands next slice). fetch_corpus_coordinates
now writes the sidecar into the corpus dir. load_corpus populates 99% of
corpus comparables.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds roof_insulation_thickness_pm1 (mirrors construction_age_band_pm1, issue
#1222): adjacent RdSAP thickness buckets (0/NI,12mm..400mm+) carry near-
identical roof U-values, so an off-by-one bucket is a SAP-neutral hit. 'ND'
(no-data) is off the ordered scale, so only an exact match counts there.
Honest measurement of SAP-relevant roof-insulation quality.
Corpus (150pc/514): exact 49.3% -> +/-1 53.7% (the misses are often multiple
buckets or ND, so the band gain is smaller than age's). Fixture: exact ==
+/-1 (0.4118) — its misses are all >1 bucket; gate floor added at 0.4118.
Also fixes two pre-existing pyright errors in the touched test file
(_epc main_fuel_type/main_heating_control were Optional but the
MainHeatingDetail attributes are non-optional Union[int, str]).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Heating sub-fields can't be field-moded without breaking system coherence,
so the whole SapHeating cluster is now copied as a unit from a single
coherent donor rather than inherited from the structural template: the
neighbour matching the cohort's modal heating signature (main fuel +
category + cylinder presence), most recent among the matches (recent cert =
current system). Including cylinder presence in the signature is load-bearing
— it protects has_hot_water_cylinder + cylinder_insulation (a bare fuel+cat
signature regressed them).
Corpus (150pc/514): heating_main_control 66.3 -> 73.9% (+7.6, the target),
main_fuel 92.8 -> 96.9, category 90.7 -> 95.7, water_fuel 92.8 -> 96.3,
water_code 88.5 -> 95.3, has_cylinder 81.1 -> 89.7, secondary 36.2 -> 42.0.
SAP MAE vs lodged 7.08 -> 6.00 (calculator floor 1.57). cylinder_insulation
-13.6 corpus (tiny-n) but +33pp on the fixture; AC requires control up +
fuel/category hold + SAP not worsened, all met.
Gate (36-target fixture): zero regression; ratcheted main_category
0.8889->0.9444, main_control 0.7500->0.8056, water_fuel 0.9167->0.9722,
water_code 0.8889->0.9444, cylinder_insulation_type 0.1667->0.5000. This is
the per-component heating method ([[feedback_per_component_best_method]]):
coherent donor, never field-mode.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Per-component method: glazing type is now the recency-weighted cohort mode
applied to every predicted window, rather than copied from the template.
Glazing is retrofitted over a dwelling's life (single -> double), so a
recent neighbour reflects the current state — same family as roof-insulation
thickness. Recency is the CORRECT weighting here: plain moding regressed the
fixture (-5.6pp) and was previously reverted; similarity weighting also
regressed it; recency improves BOTH (window geometry stays on the template,
only the glazing categorical moves).
modal_glazing_type: corpus (150pc/514) 60.7 -> 66.7% (+6.0pp); fixture
0.5000 -> 0.5278 (floor ratcheted up). Heating, geometry residuals and all
other components unchanged. Refactored _recency_weighted_mode to a reusable
_recency_weighted_choice(value_of) shared by roof insulation + glazing.
Closes the #1223 per-component approach: floor-area (median estimate) +
glazing (recency) shipped as distinct best-fit methods rather than a global
recency template, which would have disturbed the coherence-coupled heating
cluster.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Per-component method, not a global template change: the predicted floor
area is now the cohort median (the MAD-minimising point estimate of the
target's size) rather than whichever structural template's own area. The
calculator derives heat loss from building-part geometry, not this scalar,
so decoupling them is safe and the scalar becomes a better size estimate.
floor_area mean|.|: corpus (150pc/514 targets) 10.62 -> 10.48; fixture
12.2175 -> 11.8983 (ceiling ratcheted down). No other component moves.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
ADR-0029 decision 5: survivors were treated equally; now each neighbour's
vote in the cohort mode decays with its distance from the cohort's physical
centre (floor area from the median, age band from the modal band), so the
mode leans on the most representative neighbours instead of being swayed by
size/era outliers. Scales (size 20 m^2, age weight 0.5) chosen on the
validation corpus; the tight size kernel is load-bearing (looser scales
regress floor_insulation on the fixture).
Corpus (181 SAP-10.2 targets): wall_insulation 83.4->86.2%,
roof_construction 86.2->87.3%, floor_construction 78.8->81.2%,
floor_insulation 92.9->94.1%; net +7.5pp gained vs -1.1pp (two 1-cert dips,
both held on the fixture). Geometry/residuals untouched (template unchanged).
Gate (36-target fixture): zero regression across all 24 floors/ceilings;
ratcheted wall_insulation_type 0.7778->0.8333, floor_construction
0.7500->0.8125, floor_insulation 0.9062->0.9375. Dead _mode/_int_mode
removed (superseded by the weighted variants).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds PredictionConfidence (cohort size + per-component agreement = the
modal value's share among neighbours that lodge one) and
EpcPrediction.confidence(), a compute-only signal so downstream can flag
low-confidence components (ADR-0029 open item: 'confidence signal').
Sanity check on the 40-postcode corpus (1068 component predictions):
agreement is strongly predictive of correctness — pooled hit-rate 21.9%
(<0.5) / 46.7% (0.5-0.7) / 73.6% (0.7-0.9) / 95.5% (>=0.9); point-biserial
corr(agreement, correct) = 0.582. Cohort size tracks too (<6 -> 68.4%,
>=20 -> 96.0%). Surfacing / persistence is a separate HITL follow-up.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Measurement honesty so we optimise SAP-relevant accuracy, not SAP-neutral
misses (ADR-0030 Component Accuracy):
- Add construction_age_band_pm1: an exact-or-adjacent-band hit. Adjacent
RdSAP age bands carry near-identical U-values, so an off-by-one is
~SAP-neutral. Full corpus: exact 78.5% but ±1-band 91.7% (fixture
63.9% -> 83.3%) — most age misses are adjacent.
- Drop window_count from the gate's residual ceilings (cosmetic): the
predicted picture clusters at a mapper-default 4 windows vs actuals 1-21,
but total_window_area (the SAP-relevant signal) stays tight at ~3.4 m2.
Gate: + construction_age_band_pm1 floor 0.8333; window_count no longer gated.
Closes#1222
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Investigated recency-weighting (weight cohort votes by an exponential decay
in cert age). Key finding: it must be SELECTIVE. On the validation corpus it
HURTS permanent categoricals (wall 91.2->89.5, age 78.5->75.7 — discards
still-valid data) but clearly HELPS time-varying ones, where a recent
neighbour reflects the current physical state:
roof_insulation_thickness 56.7 -> 60.7% corpus (+4pp)
29.4 -> 41.2% fixture (+12pp)
So apply a recency-weighted mode only to roof_insulation_thickness (loft
top-ups happen over time); keep the plain mode for permanent categoricals.
tau = 4yr (~2.8yr half-life); falls back to plain mode when no registration
dates are lodged. Gate floor ratcheted 0.2941 -> 0.4118.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
These independent fabric categoricals were template-copied; mode them like
the construction categoricals. Verified mode beats template before applying.
Big fixture win on roof insulation thickness (doubled), floor insulation
neutral-to-positive:
roof_insulation_thickness 14.7% -> 29.4% (gate floor ratcheted up)
floor_insulation 90.6% (unchanged on the fixture)
Glazing type was tried too (+1.6pp on the 40-postcode corpus) but REGRESSED
the 36-target fixture (0.50 -> 0.44) — the gate caught it. Glazing moding is
marginal/noisy, so it's left on the template; revisit with a larger corpus.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
"One scorer, two harnesses" (ADR-0030): the committed gate, the local script,
and the future battle-test must run the *same* scoring. Extract it:
- domain/epc_prediction/validation.py — `iter_predictions` (the single
leave-one-out orchestration: latest-per-address hold-out, SAP-10.2 target
filter, all-vintage source) + `evaluate_component_accuracy` (calculator-free
ComponentAccuracy aggregation, the primary signal). Unit-tested.
- harness/epc_prediction_corpus.py — `load_corpus(dir)` IO: corpus dir ->
Comparable cohorts (maps payloads, carries address + registration_date).
validate_epc_prediction.py now just loads + calls the scorer for the component
section and iterates iter_predictions for the calculator-floored end-to-end.
Identical numbers (181 targets, SAP MAE 6.34) — behaviour-preserving.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Heating is the dominant SAP lever (ablating it to actual cut the SAP error
~7 -> ~4.5) yet was entirely unscored. Add the heating group to
compare_prediction's categorical_hits: main fuel / category / control (off
the primary MainHeatingDetail), water-heating fuel / code, has-cylinder,
cylinder insulation, secondary heating (off SapHeating).
Template-copied baseline on the 40-postcode corpus (no predictor change
yet — this just makes the signal visible):
heating_main_fuel 93.4%
heating_main_category 92.7%
water_heating_fuel/code 91.7% / 92.4%
heating_main_control 62.1% <- weak
has_hot_water_cylinder 78.5%
cylinder_insulation_type 35.8% (n=120) <- weak
secondary_heating_type 16.8% (n=125) <- weak
Fuel/category predict well from the template; controls, cylinder, and
secondary heating are poor and now drive the next predictor slices.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
ADR-0030 commits Component Accuracy to ~19 categorical components (5 today
+ 8 heating + glazing/renewables). Flat *_correct dataclass fields don't
scale — each needs manual runner wiring. Collapse them into a single
`categorical_hits: dict[str, Optional[bool]]` keyed by component name, which
also matches the runner's name-keyed aggregation (now generic: it tallies
whatever components the comparison reports). No behaviour change; the
classification rates are identical (wall n 578->575 is the 3 certs whose
actual wall is None, now correctly counted as not-applicable via _classify).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The register lists every historical lodgement, so a postcode cohort
contains the same physical address many times (LS61AA: 15 certs / 11
addresses; NG71AA: 15 / 9 — "FLAT 3" appears 3x in each). Two
consequences:
- Production: a re-lodged neighbour was counting up to 3x towards the
cohort mode. select_comparables now dedupes candidates to the latest
cert per address (one comparable per real neighbour) — Comparable
gains address + registration_date (the register metadata its docstring
already anticipated, read straight off the cached payload).
- Validation: leave-one-out leaked — predicting a flat from a near-
identical re-lodgement of itself. The harness now holds out a whole
address (excludes every sibling cert) and evaluates on the latest cert
per address (the best ground truth).
Removing the leak gives the honest numbers (19 distinct addresses):
wall_construction 93.1% -> 89.5%
construction_age_band 65.5% -> 52.6%
roof_construction 79.3% -> 68.4%
floor_area mean|.| 37.9 -> 52.6 m2
The earlier figures were inflated by self-leakage; these are the real
accuracy to beat.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Only main wall_construction was set to the cohort mode; the other
homogeneous categoricals (wall insulation, construction age band, roof
construction, floor construction) were left as template-copied, so one
median-size template's quirks set them. Apply the same cohort-mode
mechanism to all of them per ADR-0029 decision 4 — the template still
supplies geometry, only the categorical codes move to the mode.
Verified mode beats (or ties) template-copy per categorical before
applying. Smoke corpus (29 leave-one-out) classification rates:
construction_age_band 55.2% -> 65.5%
roof_construction 72.4% -> 79.3%
floor_construction 46.2% -> 84.6%
wall_insulation_type 93.1% (tie — already template-strong)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The comparison only scored main wall_construction; everything else the
predictor produces (by template-copy) went unmeasured. Extend
compare_prediction to the rest of the ADR-0029 homogeneous categoricals —
wall insulation type, construction age band, roof construction, floor
construction — and aggregate per-categorical classification rates in the
runner. A categorical hit is "not applicable" (None, excluded from the
denominator) when the actual lodges no value, so absent-roof flats don't
score free wins.
Smoke corpus (29 leave-one-out, all but wall are template-copied today):
wall_construction 93.1%
wall_insulation_type 93.1%
construction_age_band 55.2% <- loud; candidate for cohort-mode
roof_construction 72.4%
floor_construction 46.2% (n=13)
These numbers drive the next slice (extend cohort-mode coverage).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Template (the comparable whose structure/geometry is copied wholesale)
was members[0] — an arbitrary draw from the API search order. With floor
area varying widely within a property_type cohort (NG71AA houses span
51-340 m2), this made the copied geometry noisy and systematically large.
Pick the member whose floor area is closest to the cohort median instead,
implementing ADR-0029 decision 4's unimplemented "closest on size"
criterion while keeping the structure coherent (it is still one real
property, so floor dims / windows / parts stay internally consistent for
the calculator).
Smoke corpus (29 leave-one-out predictions):
floor_area mean|.| 68.0 -> 37.9 m2 (bias +46.8 -> -3.9)
window_area mean|.| 11.1 -> 7.3 m2
parts mean|.| 1.00 -> 0.38
SAP |pred-calc - calc(actual)| MAE 7.19 -> 4.86
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pure compare_prediction (TDD): wall-construction classification hit + signed
residuals on floor area, window count, total window area, building-parts count.
Plus validate_epc_prediction.py (IO plumbing): drops each cert from its postcode
cohort, predicts from the rest on guaranteed inputs only, aggregates the metrics,
and reports SAP three ways (pred-calc vs lodged / vs calc-on-actual / vs the
neighbour-mean baseline). Smoke run: wall 90.9%, floor-area mean|·| 42.6 m2 (a
real signal — template-copied floor area is noisy), SAP pred-calc edges baseline.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
predict() copies a representative template comparable's structure (coherent for
the calculator), overrides the homogeneous categorical with the cohort mode
(robust to an atypical template), then applies known Landlord Overrides on top
(a known value wins over the estimate). Proven on wall construction; roof/floor/
insulation/age extend on the same mode+override mechanism, driven next by the
validation harness metrics.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pure-domain select_comparables: property type is an always-hard filter; built
form and known Landlord Overrides (e.g. solid brick) are conditioning filters on
the filter-then-relax ladder — applied while >= minimum_cohort survive, relaxed
otherwise (the mixed-street border case degrades gracefully). PredictionTarget
(known inputs) + Comparable (epc + register metadata) + ComparableProperties
(selected cohort). Weighting (recency x similarity) follows in the synthesis slice.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
