Why a model, and not seven dashboards
Hazards are bought one at a time and happen together. A wildfire burns the slope whose soil then moves; the ground already subsiding is the ground the flood finds first; the road closed for rockfall is the road the response convoy needed. An organisation that buys seven products gets seven pictures that do not reconcile, and a person whose job becomes reconciling them.
One model removes that job. Every OrbiVigil product reads from and writes back to the same territory model, so a question asked of one hazard is answered in the presence of the others. That is the whole architectural bet, and it is what the phrase Agentic World Engine names.
What the model holds
Six families of state, each measured independently and each carrying its own provenance:
- Geology and the ground beneath — mapped geology and catalogued cavities, the slow layer that decides what the fast layers do.
- Water — flood extent and hydrological forecast at territory scale, and street-level ponding rank inside a city.
- Ground motion — millimetre-per-year displacement of the surface, with multi-year history at a point rather than a single snapshot.
- Fire — thermal detection, confidence-scored events, spread forecasting, mapped burnt-area perimeters and burn severity.
- Weather — the wind, humidity and precipitation fields that drive the hazards above, from public model sets.
- Official risk — the warnings and criticality bands issued by national and regional authorities, relayed with the issuer's own attribution and update time. We carry them; we do not issue them.
Every layer has an honest cadence
The layers do not move at one speed, and pretending they do is how a decision-support tool becomes misleading. So the model reports the age of every figure it serves, and the cadence is stated in the terms the physics allows rather than the terms a brochure would prefer:
- Fire and weather move within the hour. Thermal detection is a minutes-to-hours signal; wind and precipitation fields refresh through the day.
- Official warnings move on the issuer's clock. Not ours — a relayed bulletin is exactly as fresh as the authority made it, and it says so.
- Water moves in hours to days for flood extent and forecast; a ponding rank is a property of the street, and changes when the street does.
- Ground motion is measured over seasons and read over years. A millimetre-per-year rate is a trend, not an event: its value is the multi-year history behind it, and it is never presented as a live reading.
- Geology barely moves at all. It is survey-era knowledge, and its honesty is in provenance and coverage, not in freshness.
A figure whose layer cannot answer at the speed of the question is refused rather than interpolated. Refusals are first-class values in this platform, for machines as for people.
What runs today, and what the architecture is built for
Today. Fire, flood, official-warning relay, ground motion and the urban layers run in production. Ground-deformation monitoring is live under every provincial road across a 91-municipality institutional pilot in Italy, with per-point five-year displacement histories and post-intervention verification — did the stabilisation work? The urban layers — mapped geology and catalogued cavities, street-level ponding rank, surface heat, millimetre movement beneath buildings and roads — are live across that pilot, municipality by municipality, with urban air quality in beta.
Built for, not yet serving. The model is designed so that identified landslide bodies, cross-hazard impact assessment and unified alert delivery attach to the same territory state without a second integration. Those are on the roadmap and are described as planned wherever they appear. The platform publishes its own capability register — live, beta or planned, per product and per market, computed at the moment you read it — so a claim on this page can be checked against the system rather than believed.
What it is worth to a buyer
To a public authority, the model replaces work that is currently done by hand: one regional bulletin, one colour over a region-sized zone, translated every morning into a decision for each municipality. To an insurer, it is the difference between a claim narrative and a measurement — exposure that was recorded before the event and verified after it. To an investor, it is why the marginal cost of the eighth hazard product is not the same as the cost of the first.
Related: agentic AI at OrbiVigil — how machines read this model; physical AI — how drones and ground robots consume it and feed it back.