One 0-to-100 number per UK area. Four scoring profiles tuned to four workflows: residential origination, commercial site selection, investment underwrite, and research. Each profile uses its own five-dimension set; override the weights per request or save your own recipe against your org. The engine is frozen, golden-tested, and AI never touches the scoring path. Every response stamps the engine version that produced it.
Pick a postcode and a scoring profile. See the composite number, the five dimensions that produced it, the weight applied to each, and the engine version that ran.
safety_crimetransport_linksamenities_servicesdemographics_economyenvironment_qualityReading this view
presetweights_sourceengine_versionSample shape and illustrative values. Real scores vary per release of the underlying sources and the engine version your org is pinned to.
Each profile selects a different five-dimension set plus default weights. The matrices are different on purpose: a residential origination flow does not care about commercial costs, and an investment underwrite does not care about schools. The engine should not pretend those are the same.
preset: movingHow liveable is this area for a household. Drives the area-quality lens on listing detail pages, valuation flows, and relocation tools.
5 Dimensions
preset: businessWhere to open. Foot traffic, competition density, transport access, spending power, occupancy cost. Drives shortlisting at portfolio scale.
5 Dimensions
preset: investingWhat this area looks like as an asset. Growth trajectory, yield, regeneration context, tenant demand, downside risk.
5 Dimensions
preset: researchAnalyst-friendly default. Balanced weights across safety, transport, amenities, demographics and environment. Survives FOI and procurement review.
5 Dimensions
Each call runs the same three steps in the same order. Caller weights enter the pure aggregation, not the engine. Same inputs always produce the same score.
Hybrid read across the persisted store and live upstream fetches. Deprivation, property and crime served from the store when LSOA coverage exists; amenities, transport, schools and environment live per request.
Frozen v2 deterministic engine produces the five per-dimension scores for the chosen preset, plus per-dimension confidence. Golden-tested. AI does not touch this step.
Pure aggregation outside the engine. Default weights from the preset, overridden by caller weights or a saved org preset_id. Aggregate confidence is weight-weighted across dimensions.
Plain JSON over HTTPS, Bearer-token auth with the oga_ prefix, all paths under /v1/. Score endpoint is free of the monthly report quota; the legacy /v1/report endpoint still consumes it.
The primary scoring endpoint. Send an area and a preset; get a 0-100 composite, the five dimensions that produced it, the weight applied to each, per-dimension confidence, the aggregate confidence, weights_source and the engine_version that ran. Not metered against the monthly report quota. No report is generated.
areastringRequiredUK postcode or place name. Geocoded server-side.
presetenummoving · business · investing · research. Defaults to research. Each preset has its OWN 5 dimensions.
weightsRecord<key, positive number>Override the preset's default weights. Keys must be in PRESET_DIMENSION_KEYS[preset]. Partial overrides keep defaults for missing keys.
preset_idstring (Lever)Reference a saved per-org preset (spr_…). Mutually exclusive with preset + weights. 422 preset_id_conflict otherwise.
X-Engine-VersionheaderPin response engine version per request. Must be in SUPPORTED_ENGINE_VERSIONS = ['2.0.0','2.0.1','2.0.2']. Beats org-level pin.
ScoreResult: { area, preset, score, area_type, dimensions[5]: { key, label, score, weight, confidence }, confidence, weights_source ('preset' | 'custom'), engine_version }. Response header X-Engine-Version carries the effective pin.
What the score is FOR changes per buyer. The deterministic pipeline underneath does not.
The problem
A mortgage lender's model risk register treats every API the underwriting model depends on as a model input. If the supplier silently changes a coefficient, that's an undisclosed model change, a regulated event. Auditors ask: 'what version of the area score produced this decision, and is it byte-equivalent to the score you'd compute today?'
Why Scores
Every /v1/score response carries engine_version in the body AND X-Engine-Version on the response header. Both can be locked at the org level (Levers methodology pin), owner-only, write-time validated against SUPPORTED_ENGINE_VERSIONS. The deterministic engine is frozen v2, golden-tested. AI never touches the scoring path.
Their value
Every decision produced from /v1/score is reproducible to a known methodology version. The model risk register has a 1:1 mapping between an API call and the engine that ran. Closes a compliance gap that's table-stakes for any production underwriting model.
Versioned, pinnable, deterministic area scoring with the audit trail your model risk register already asks for.
The problem
An underwriter's area-risk view is a weighted blend the actuary owns (safety, flood, demographics, property volatility) and the weights change with each pricing cycle. Off-the-shelf 'area scores' are black-box composites with frozen weights the vendor chose; useless inside an underwriting model because the actuary cannot see or tune the weighting.
Why Scores
POST /v1/score returns every dimension's raw score, weight, and per-dimension confidence. The actuary sees the components, not a black box. With Levers saved presets the carrier saves a preset once and references it as preset_id forever. Confidence is variance-aware on property-backed dimensions so wide swings cap at MEDIUM.
Their value
Configurable weights without per-call payload. Transparent components for actuary review. Honest confidence signal that flags volatility before adverse selection bites. One preset_id per underwriting model lifecycle.
Configurable composite scoring the actuary can audit, with a saved preset_id per model and honest confidence on every dimension.
The problem
PropTech tools need a single composite 'how good is this area' number for the UI, not a 12-signal forensics view. But buyers, renters and agents see different things in 'good': a family cares about schools and safety, an investor cares about growth and yield, a business cares about footfall. One score does not fit; building four bespoke scores in-house means owning the data pipeline.
Why Scores
The four scoring profiles are exactly those four audiences. One endpoint, one call, four UX flavours by parameter. The engine is deterministic, so the score is stable for a given postcode + profile across deploys, so cached UI states stay coherent. Components plus weights plus confidence come back in every response.
Their value
Four audience-matched scores from one endpoint, no in-house pipeline. Deterministic + stable for caching. Transparency on demand (drill into the five dimensions) without a second API roundtrip.
One endpoint, four audience-tuned composite scores, deterministic enough to cache and transparent enough to drill into.
The problem
A CRE site-selection workflow needs to compare hundreds of candidate locations on a consistent yardstick: footfall demand, competition density, transport access, local spending power, commercial costs. Most data vendors stop at single-signal lookups; rolling them into a defensible score is the analyst's problem.
Why Scores
POST /v1/score with the commercial site-selection profile returns exactly the five dimensions a site-selection analyst needs, with weights the team can override per portfolio class or save once via preset_id. Every response carries per-dimension confidence + the deterministic engine version, so the committee paper has a defensible methodology pointer.
Their value
Purpose-built commercial-workflow dimensions instead of a generic composite. Configurable weights per portfolio class. Defensible methodology cite. Free of the monthly report quota: score thousands of candidates at the per-key rate limit.
A purpose-built commercial site-selection score with per-portfolio weights and a methodology cite you can put in the committee paper.
The problem
A council planning team or regeneration body needs a consistent way to triage LSOAs across a borough. They do not need an AI narrative; they need a reproducible, methodology-documented number that survives FOI scrutiny and the next procurement review.
Why Scores
POST /v1/score with the research baseline profile is the general-purpose composite. The engine is deterministic: same postcode + profile equals the same score across deploys. The methodology changelog is published and engine_version is stamped on every response. An FOI request can be answered by pointing at the version.
Their value
FOI-defensible methodology trail. Reproducible across deploys and procurement cycles. Researcher-tuned default preset that does not bias toward commercial or investor lenses.
A reproducible, methodology-documented composite for triaging areas at the LSOA level, built for the procurement notice and the FOI request.
Four presets, each with its own five dimensions. Override weights per request or save them as a per-org preset. AI never touches the scoring path. The methodology version is on every response, both body and header.