Fragility.
How every asset actually responds.
Not one generic curve per occupancy class — a defensible β for each asset, built on 15 years of designing structures that cannot fail. Buy exactly the certainty the exposure justifies.
Four grades of β. Buy exactly the certainty you need.
Each grade adds a defined layer of building-specific evidence and earns a tighter, fully-sourced β. Mix grades across a portfolio — triage the long tail cheaply, spend engineering only where the exposure justifies it.
Catalogue
Location + occupancy. Regional archetype prior. No imagery, no visit.
Visual AI
Façade & overhead imagery. Structural system, height, irregularity, soft-storey flags.
Walk-down
On-site rapid survey. Drawings & retrofit evidence, captured and structured.
Analytical · PBSD
Full drawings + material data. Nonlinear performance-based model. Asset-specific, verified.
Visual AI matches the imagery-screening tier the market already knows — but per asset, and with every β fully sourced. Above it sit two engineered grades no class-based model reaches.
β ≥ 0.30 is physics, not a sales floor. The record-to-record variability of ground motion is irreducible — across the published fragility literature its dispersion floor sits near 0.30. We engineer β down to that floor and then tell you we've stopped — that honesty is the moat.
Every grade's β traces to a catalogue entry with documented provenance; the irreducible floor is anchored in record-to-record dispersion reported across the open fragility literature, not assumed.
Built on real design, not assumptions.
“Fragility curves today are built on simplified assumptions. Ours are built on 15 years of designing structures that must not fail. By pairing performance-based seismic design with AI-assisted evidence collection, we transform fragility from generic to engineering-grade.”
Vera assembles the evidence. The engine computes β.
Vera is the assistant that runs the upper grades: she knows what each structural family needs to climb a grade and drives the evidence collection. She never computes β — she hands a complete, sourced evidence file to the deterministic engine, which does.
- Visual AIFamily-specific checklist, classification, gaps flagged.
- Walk-downOn-site protocol issued, findings captured into the dossier.
- PBSDAnalytical fragility cross-checked, β sealed with its provenance.
The AI never touches your number.
Two separate layers. The AI-assisted elicitation layer (Vera) only assembles and structures the evidence. The deterministic engine computes β from that evidence — the same evidence in always yields the same β out.
Same evidence → same β, every run.
Re-run it tomorrow, or in five years, and get the identical curve.
Every β unfolds back to the sourced evidence that produced it.
Maps cleanly to Solvency II. Vera lives where the framework already asks for rigour — the data-quality layer, the completeness and appropriateness of each asset's evidence. The validated, deterministic engine is the model that computes the number. AI exactly where the regulation wants better data governance, never inside the regulated calculation.
Fragility is the vulnerability layer.
Combine it with Sismicus hazard, and Risco turns both into loss and pricing.