One deterministic peril model. Fifty-three perils. A score for 282.9 million parcels — zero null cells — with full methodology trace and reason code preserved end-to-end. Patent-pending USPTO 64/070,786.
Each stage emits a verifiable artifact: a parcel hash, a model manifest, and a signed score response. Every score in production can be replayed offline and reproduced bit-for-bit.
Every address normalizes to a rooftop coordinate plus a stable parcel identifier. We don't model points — we model parcels, because that's what gets insured.
Atmospheric, surface, seismic, wildfire, convective, chronic-climate, historical, and forecast layers — every one tiled and cached at the parcel level so resolving an address is a single fan-out, not eleven sequential calls.
Fifty-three perils collapse into a single A–F grade plus full per-peril percentiles. The grade is rate-filable; the percentiles are model-feedable.
// POST https://api.atlasunited.io/v1/risks/fast { "address": "1242 Ocean Dr, Miami, FL 33139" } // 200 OK · 174ms { "parcel_id": "8f3a2b91d7", "grade": "F", "percentile": 98, "top_perils": ["hurricane", "coastal_flood", "storm_surge"], "model": { "version": "AEPM v2024.3", "calibration": "22-yr severe-event catalog", "layers": ["atmos", "surface", "convective"] } }
Most peril vendors give you a number. We give you a number plus the citation, version, and confidence interval — every single time. That difference is the entire reason actuaries can put us in a filing.
Drop in an API key. Stream a CSV. Get back graded scores with full provenance. No call required.