The run only showed the measures the Optimiser selected, so a candidate it
passed over (e.g. an ASHP it found too costly for the target band) and that
measure's cost were invisible.
Add `harness.console.candidate_recommendations` — every Generator Option
with its per-Option cost, before optimisation — and have run_modelling_e2e
print the full menu per property (flagging the selected Options), write a
"cost per measure" section into the markdown, and emit a per-Option
modelling_e2e_candidates.csv.
Co-Authored-By: Claude Opus 4.8 <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>
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>
The committed gate needs frozen, reproducible data without dumping real UK
addresses into the repo. Add:
- harness anonymise_payload + stable_hash: hash street address + cert number
into opaque, dedup-stable tokens; blank secondary address lines + post_town;
keep postcode + all component/lodged fields (gov data is OGL). Unit-tested.
- scripts/build_epc_prediction_fixture.py: curate qualifying postcodes (>=1
SAP 10.2 target + >=2 distinct addresses) from the local scratch corpus,
anonymise, freeze under tests/fixtures/epc_prediction/.
- The frozen fixture: 15 postcodes / 280 certs / 36 SAP-10.2 targets.
Verified no plaintext address_line_1 and post_town all blank.
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>
Orchestrator runs recommend_secondary_heating_removal; report._triggers_for
explains it via the lodged secondary_heating_type; harness catalogue + ARA seed
price it. Re-pins the golden/integration plans it shifts: it is a cheap (\£250)
SAP lever, so on gas-main certs lodging an electric secondary (691) it displaces
the \£12k ASHP (0330, 0036) or joins the all-beneficial-measures package (000490,
where its marginal SAP is 0 under the category-4 ASHP but the heater is still
physically removed). Consistent with the optimiser's existing kitchen-sink
package behaviour.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Flat per-dwelling decommission price (sample_catalogue \£250) + 0.25 contingency
(covers unknown heater count / hard-wired-vs-plugged / repaint extent). The JSON
repo joins the contingency from config, proven by the new repo test. No composite
Products machinery — a lodged secondary is one roughly-fixed job, not room-scaled.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add the system tune-up to the heating Recommendation: keep the existing wet
boiler but install better heating controls and fix the cylinder. Two competing
Options (the Optimiser picks <=1 across the whole heating rec) per the user's
two best control end-states:
- system_tune_up — standard controls (programmer + room thermostat +
TRVs, SAP 10.2 Table 4e code 2106)
- system_tune_up_zoned — time-and-temperature zone control (code 2110, type 3):
more SAP uplift for more cost
Both keep the boiler (no fuel / SAP code / flue change), set the control
ABSOLUTELY to their end-state, and apply the conditional cylinder fixes (an
80 mm jacket when under-insulated, a thermostat when absent — only when a
cylinder exists). Each control option is offered only when it genuinely improves
the existing control — standard is skipped when the control is already 2106 /
2110 / 2112, zone when already 2110 / 2112 — so neither is ever a downgrade or a
no-op.
Validated against the Elmhurst "system tune up" re-lodgements (cert 001431):
nine befores spanning controls 2101-2113 all converge to the two common afters,
proving the control overlay is absolute. The cascade pin is parametrised over
two starting controls (2101 "no control" + 2113 "room thermostat and TRVs") x
both afters, delta 0 (SAP/CO2/PE).
Wires the two MeasureTypes through contingencies (0.15), the offline catalogue
(500 / 900), the catalogue-coverage list, the report triggers, and the ARA
first-run seed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add the `gas_boiler_upgrade` branch to `report._triggers_for`, mirroring the
generator's eligibility guard so a cohort report explains why the boiler upgrade
fired: the wet-boiler SAP code, the mains-gas connection that makes the gas
end-state installable, and the cylinder presence that shapes the bundle (combi
vs regular + cylinder fixes).
No golden API cert selects the boiler upgrade (it competes with — and on houses
loses to — the ASHP bundle within the one heating Recommendation), so the branch
is covered by a direct `_triggers_for` unit test, following the repo pattern for
testing internal helpers (cert_to_inputs).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add the first boiler-upgrade option to the single "Heating & Hot Water"
Recommendation (ADR-0024 expansion): a dwelling whose existing wet boiler heats
a hot-water cylinder is offered a new gas condensing boiler, with the cylinder
jacketed when under-insulated and given a thermostat when absent. One competing
Option (the Optimiser picks <=1), folded into one composite Plan line.
The end-state is read from the Elmhurst before/after re-lodgements (cert 001431,
gas boiler upgrade - with cylinder), which REVISE ADR-0024:
- Target is always a gas condensing boiler, not fuel-preserving: every after
lodges fuel 26. Gas->gas always; a non-gas wet boiler ->gas only with a
mains-gas connection; electric boilers are left alone (electrification is the
upgrade path). Eligibility = wet-boiler SAP code (Table 4a/4b 101-141 /
151-161 / 191-196) + not an electric boiler + mains gas present.
- End-state is a Table 4b SAP code, not a PCDB index: code 102 (regular boiler
+ cylinder). The calculator derives the condensing seasonal efficiency from
the code, so no efficiency input exists or is needed.
- A modern condensing boiler has a fanned flue: the after flips
`fan_flue_present` False->True on every cert (SAP 10.2 Table 4f flue-fan +
the Table 4b condensing-efficiency basis). Added as a new HeatingOverlay
field, routed to main_heating_details[0].
- Cylinder thermostat is always added when absent (user-locked); the jacket is
the 80 mm `cylinder_insulation_type=2` end-state, applied only when the
cylinder is below 80 mm (never downgrading a better one). Both are conditional
per-dwelling components, not a frozen overlay.
Cascade-pinned delta-0 (SAP/CO2/PE) against the relodged after via
`_assert_overlay_reproduces_after`. NB the absolute SAP on this dwelling is
subject to a separate Summary-path mapper roof-fidelity gap (we read the roof
better-insulated than Elmhurst, scoring ~75 vs the printed 56); the gap is
identical on before+after (the boiler measure never touches the roof) so it
cancels and the pin still proves the exact heating field-delta. Tracked on the
calculator branch.
Wires the new `gas_boiler_upgrade` MeasureType through contingencies (0.26),
the offline sample catalogue, the catalogue-coverage list, and the ARA
first-run integration seed (the option fires on any mains-gas boiler+cylinder
dwelling).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 5 (local run sources the DB, read-only) + slice 6 (optional persist),
landing together as one script rewrite (the persist path is interleaved with
the compute path).
The same local computation now runs whether or not the result is stored:
- Both modes price against the live `material` catalogue (read-only
ProductPostgresRepository over one shared Session) and model against a real
Scenario read from the DB (--scenario-id; its goal_value drives the band,
rejected if null) — so the inspected recommendations are exactly what gets
stored. The JSON sample catalogue is no longer used by this script.
- --measures restricts the run to a comma-separated considered_measures
allowlist (e.g. high_heat_retention_storage_heaters,solar_pv).
- --persist writes the inputs (EPC + spatial + solar) and the *same* computed
Plan via the production repos in one PostgresUnitOfWork, then commits
(idempotent: PlanPostgresRepository replaces by (property_id, scenario_id)).
Gated: --persist requires --scenario-id and --portfolio-id. Default is
inspect-only — no DB writes.
harness.console.run_modelling gains `products` and `scenario` overrides (the
seam the script drives); defaults unchanged, so existing callers are
unaffected. Suite 257 pass + 3 xfail; pyright clean; --help/guard/measure
parsing verified. Not yet executed against the DB (awaiting property_ids +
write-confirm).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add domain/modelling/considered_measures.py::restrict_to_considered_measures —
the pure allowlist that limits a run to a chosen set of MeasureType (mirroring
the legacy engine's `inclusions`). It filters at the Option level, so a
multi-option Recommendation (e.g. Heating & Hot Water competing HHRSH against
an ASHP bundle) is kept with only its allowed Options; a Recommendation left
with none is dropped. None = consider everything (unrestricted default).
Thread `considered_measures: frozenset[MeasureType] | None` through
ModellingOrchestrator.run -> _plan_for -> _scored_candidate_groups /
_candidate_recommendations (applies the filter) and _measure_dependencies
(suppresses a forced dependency whose required measure is outside the
allowlist, so a restricted run forces nothing it is not considering). The
local-run seam (harness.console.run_modelling) gains the same param.
The Optimiser still freely chooses among survivors — including none. Tests:
the pure filter (3 cases) + an orchestrator-seam test proving a
{solar_pv}-restricted run yields only solar_pv options. 257 pass + 3 xfail;
pyright clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Revives the local recommendation-inspection flow for specific Properties.
`scripts/run_modelling_e2e.py` reads each Property's UPRN from the DB
(read-only), fetches the latest EPC live from the gov EPC API by UPRN, runs the
Modelling stage in memory (all Generators → Optimiser → costed, attributed
Plan), and prints a per-Property plan table + writes a Markdown/CSV summary.
Persists nothing — purely for inspection.
The local DB's Properties have no linked ingested EPC (epc_property.property_id
is NULL for all rows; Ingestion's source clients are stubbed, #1136), so the
EPC must be fetched inline rather than read back. Builds the connection from the
`DB_*` env vars in backend/.env and the EPC token from `EPC_AUTH_TOKEN`.
Threads optional solar insights through harness `run_modelling` (so Solar PV
Options can fire once coordinates are wired) and adds the `solar_pv` catalogue
row. Solar + planning restrictions + DB persistence are noted follow-ups.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
recommend_heating now receives planning_restrictions in the orchestrator (the
ASHP planning gate); the ASHP bundle joins the free candidate pool for every
house/bungalow. Catalogue + contingency (legacy 0.25) gain air_source_heat_pump;
report.py _triggers_for explains the ASHP trigger; the harness forcing test
covers it. Integration tests seed an air_source_heat_pump MaterialRow (ASHP
fires on every house, the broadest trigger yet). NB the optimiser correctly does
NOT select ASHP for an EPC-band goal — gas->electric does not improve the SAP
cost-rating; ASHP is a CO2/PE measure, selectable once non-EPC goals land. ASHP
bundle COMPLETE (S5-S7). ADR-0024.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
recommend_heating joins the free candidate pool in _candidate_recommendations;
the HHR storage bundle reaches the optimised package for an electric/off-gas
dwelling. Catalogue + contingency (legacy 0.10) gain
high_heat_retention_storage_heaters; report.py _triggers_for explains the
heating trigger (electric/off-gas main); the harness _GENERATOR_MEASURE_TYPES
forcing test covers it. ASHP + boiler bundles still to come. ADR-0024.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 4 of the lighting generator (ADR-0023): run recommend_lighting in
_candidate_recommendations (no planning gate). Price low_energy_lighting in the
offline catalogue + contingency table (0.26, the legacy rate); the
_GENERATOR_MEASURE_TYPES forcing test enforces both. A run_modelling test pins
the wiring end-to-end (an incandescent-lit dwelling gets the LED upgrade in the
optimised package).
Downstream updates, all because lighting now fires on any cert with non-LED
bulbs: report.py gains the low_energy_lighting trigger (the non-LED counts); the
two golden-cert report tests and the multi-measure integration test now expect
low_energy_lighting alongside the fabric measures (the sample/golden EPCs lodge
low-energy-unknown bulbs); first-run integration seeds a low_energy_lighting
MaterialRow.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 4 of the glazing generator (ADR-0022): run recommend_glazing in
_candidate_recommendations, threading the Property's PlanningRestrictions so a
protected dwelling is offered secondary glazing instead of double (mirrors
recommend_solid_wall). Price both Measure Types in the offline catalogue
(double £600/window, secondary £510 -- the legacy 0.85x scaling) and the
contingency table (0.15, the legacy windows_glazing rate); the
_GENERATOR_MEASURE_TYPES forcing test enforces both entries exist.
run_modelling tests pin the wiring end-to-end on an all-single-glazed dwelling:
double when unrestricted, secondary when listed. The first-run integration test
seeds a double_glazing Product because its lodged EPC has a single-glazed
window. _single_glazed_epc() deep-copies build_epc() (which shares its window
objects) so the mutation can't leak into other tests' baselines.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 4 (ADR-0021). The roof dispatcher can now emit sloping_ceiling_insulation
and flat_roof_insulation, so wire both into contingencies and the sample
catalogue; the forcing-function test now asserts every generator measure type
is both priced and has a contingency rate, so an offline/live run over a
sloping or flat roof never dies on a missing entry.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 3b+3d (ADR-0019/0020). Property gains a planning_restrictions attribute
(default unrestricted); the ModellingOrchestrator threads it from the Property
through _plan_for -> _scored_candidate_groups -> _candidate_recommendations into
recommend_solid_wall, replacing the unrestricted default. run_modelling exposes
a planning_restrictions param so the offline harness can inspect restricted
properties. Integration test: a listed solid-brick dwelling that gets IWI when
unrestricted now yields no wall insulation. 145 tests pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 2e. recommend_solid_wall joins the orchestrator's fabric generator pool
(restrictions default unrestricted until slice 3 sources them); the harness
catalogue + contingencies (26%) gain external_wall_insulation /
internal_wall_insulation. run_modelling on an uninsulated solid-brick dwelling
(baseline SAP 36.6) now selects internal wall insulation into the optimised
package; the catalogue-completeness guard covers both new measure types.
Golden cohort 57/57 still error-free; IWI now fires on a real cohort cert.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
fetch_epc_bulk_sample streams certificates-<year>.json out of the bulk ZIP via
range requests, keeps the first N SAP-version matches, and writes each cert's
inner document to <out>/<cert>.json for run_property_report. Stops after N, so
only the member prefix transfers, not the 15.7 GB archive (RangeFile.bytes_read
reports the true transfer vs the absolute ZIP offset). Verified on 2026: 100
SAP-10.2 certs -> report ran 81 scorable (MAE 2.03), 46 flagged, 19 raises
(11 full-SAP schema 19.1.0, 7 unmapped floor_construction 0/3, 1 missing
post_town) — real shadow-validation signal vs the curated golden 57.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The bulk endpoint 302-redirects to a 15.7 GB S3 ZIP with one NDJSON member per
year; each line wraps the per-cert payload in a stringified 'document' that
parses to the same RdSAP-Schema-21.0.1 shape from_api_response already handles.
parse_bulk_line unwraps a record; is_sap_version filters to SAP 10.2; RangeFile
exposes the S3 object as a seekable file so zipfile streams a single year's
member (and a sampler stops early) without downloading the whole archive.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
format_report_csv emits one comma-safe row per property: the calculator-error
fields (lodged/calculated/Δ/flag), the Plan headline figures (baseline+post
SAP/band, measures, cost+contingency, bill & CO2 savings, valuation %), the
flattened measure triggers, and any captured error — sortable in a spreadsheet
for a large dump.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
format_report_markdown emits: (1) cohort parity stats + a per-property
lodged-vs-calculated table flagging |Δ| > 0.5 (errors shown inline),
(2) Plans + costings (SAP/band jump, cost + contingency, bill & CO2 savings,
valuation uplift), (3) each fired measure with the EPC attributes that
triggered it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
build_property_reports models a dump in order (errors captured per-cert);
parity_report_for aggregates the lodged-vs-calculated SAP across the cohort
into the existing ParityReport (MAE/RMSE/bias/worst-N), excluding certs that
couldn't be mapped or scored. Residual convention is the calculator's own
(predicted - actual), the negative of PropertyReport.sap_error.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Section 3 of the report: build_property_report now runs the Modelling stage
and, for every Plan Measure, records the EPC attribute(s) that caused its
generator to fire (MeasureTrigger) — wall_construction/insulation for cavity
fill, roof thickness for loft, floor thickness/construction for floors, the
absent mechanical kind for ventilation. Modelling raises are captured as
plan_error, independent of the calculator-error capture.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Section 1 of the property inspection report: PropertyReport compares the
cert's lodged energy_rating_current to Sap10Calculator's un-rounded SAP and
flags |Δ| > 0.5 (the ADR-0010/0013 shadow-validation design target). A
mapping/scoring raise is captured per-cert as calculator_error, never
propagated, so one bad cert can't abort the sweep.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CertResult now carries its Plan (with flat baseline/post-SAP/measures
properties), and `format_cohort_csv` renders one browsable row per cert
(SAP transition, band, measures, cost, bill saving, valuation %, error).
`scripts/run_modelling_cohort.py` is turnkey: no args runs the committed
golden cohort, prints a sense-check table for the first measure-bearing
certs (a capped preview so a large dump doesn't flood the terminal), the
summary, and writes modelling_cohort.csv (gitignored). Point it at the
EPC dump when it lands.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
`harness.cohort.run_cohort(paths)` parses each API-shaped EPC JSON with
from_api_response and models it via run_modelling — no database, no
network — capturing per-cert errors instead of aborting the sweep, plus
`format_cohort_summary`. A thin `scripts/run_modelling_cohort.py` CLI
points it at a directory. Proven over the 57 golden API certs: 56 ran
offline, 15 produced measures, 1 errored (COAL has no Fuel Rates entry —
a BillDerivation coverage gap, not a harness one). Ready for the EPC dump.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two fixes that unblock offline, no-database inspection over an arbitrary
EPC dump:
- Complete the harness sample catalogue with loft_insulation and
solid_floor_insulation — the four fabric generators can emit five
Measure Types, but the catalogue priced only three, so an offline run
on a property with an uninsulated loft or solid floor raised mid-run.
A new test pins the catalogue to cover every generator Measure Type.
- Add `run_modelling(epc, ...)` — runs ONLY the Modelling stage (no
Ingestion / Baseline), so it needs no lodged recorded-performance / RHI
and inspects recommendations on any calculator-scorable EPC. `run_one`
(full pipeline) stays for when you want Baseline too.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The Plan derives its Valuation Uplift (ADR-0018) from its baseline -> post
band jump and works+contingency cost, given one external input — the
Property's current market value (a Property Valuation, mostly absent).
`Plan.valuation` / `Plan.baseline_epc_rating` are derived like the other
headline figures; `PlanModel.from_domain` maps the £ forms to the live
plan.valuation_* columns (NULL when no value — the percentage is not
persisted on those columns). `Property.current_market_value` is the new
optional source; the orchestrator threads it onto the Plan. `run_one`
takes a `current_market_value` so the harness can value the uplift, and
the sense-check table shows the average % (always) plus the £ forms when
known.
Sourcing the current market value (upload / default) remains deferred
(ADR-0018); it is None throughout until that lands, so the columns stay
NULL at scale.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 3. `harness.console.run_one(epc, goal_band=...)` wires the full
AraFirstRunPipeline against in-memory fakes — no Postgres, no network —
runs one property, prints the sense-check table, and returns the Plan
for interactive poking from a REPL at the worktree root. Defaults to the
committed harness sample catalogue.
Refactors the slice-1 integration test to delegate to run_one (dropping
~70 lines of duplicated wiring + the now-unused test catalogue fixture),
so it exercises the shipped entrypoint rather than a parallel copy. The
new console test covers run_one's print/return contract.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Slice 2. `harness.plan_table.format_plan_table(plan)` renders a Plan as a
plain-text table — one package summary line (baseline SAP/band -> post
SAP/band, CO2 saved, cost of works + contingency, bill saved) and one
line per Plan Measure (signed SAP points, cost, delivered kWh + £
savings). Pure presentation: reads the Plan, computes nothing. The
DB-less First Run test now prints it (visible under `pytest -s`) so the
modelled package can be eyeballed and debugged by hand.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
