Build on UK area intelligence.

The deterministic, addressable UK area-data layer. Every signal is a typed primitive with value, normalised value, national-within-country percentile, source period, and per-signal confidence. LSOA grain, monthly cadence on the dynamic series.

• Typed Signal primitive (Zod-validated in packages/contracts)
• Honest provenance via meta.fetch_mode (live, store, hybrid)
• Append-only monthly time-series, immutable per observed_period
• Derived signals are first-class store rows (YoY, momentum, slope, peer-relative z)

Deterministic composite scoring. Pick one of four presets (moving, business, investing, research), each with its own five dimensions. Override the weights per request, or save a preset against your org. No AI in the scoring path; the engine version is stamped on every response.

• Four presets, each with a different five-dimension set
• Custom per-request weights or saved preset_id (per org)
• Per-dimension confidence plus aggregate confidence
• engine_version on the body and X-Engine-Version response header

Portfolios plus change detection. Save a book of areas (postcodes or LSOAs), bulk-enrich through the scoring engine, and detect material moves across the time-series. Signed webhook alerts (HMAC-SHA256) when something material shifts.

• Portfolios as named collections of areas (CRUD, dedup)
• On-demand period-vs-period change detection
• Sample-size gated (default min 8 transactions on price moves)
• Three webhook events: report.created, score.changed, signal.changed

A typed query plus insight plane over the moat. Six plan ops (rank_areas, get_area, score_area, find_peers, find_insights, find_forecast) reachable as a Zod-strict programmatic plan, or as natural language that the planner translates into the same plan.

• POST /v1/query with {plan} (programmatic) or {question} (NL)
• k-NN peers with materialised ~840k-row peer graph
• Anomaly screening via pre-materialised peer-relative z-scores
• Linear-regression forecast — not a learned model, not ARIMA, not Prophet