Outcome · Catastrophe readiness
The storm is over the book by Tuesday. The decisions it forced are still being argued about on Thursday.
A live event compresses a month of decisions into two days: moratorium or not, surge staffing or not, cede notice or not. Each one is defensible on the day and hard to defend a year later, because nothing recorded what was known at the moment it was taken.
What is solver-driven, and what is honestly replay
How we count this
The solver list is the shipped engine’s own, and the catalogue split is the demo-data corpus specification’s — specified and gate-carried, never a claim that the events were generated. The distinction the whole section rests on: gate G-63 makes any windstorm event that claims to come from a physics solver a hard build failure. Windstorm is replay of published footprints — real past events played back, not simulated ones — and it is labelled replay everywhere it renders. We would rather ship the label than imply the solver.SOV/services/nexus-disaster-engine/src/{lib.rs,sims/,feeds/}
The calibration band that refuses to round
G-69 is a specification gate: the referent — OPW CFRAM extents — is published and verified. The band and the threshold are what the corpus must satisfy against it.
How we count this
PERILS AG puts the Storm Éowyn Irish market loss at €316m; Insurance Ireland’s final bill came in at over €301m; 33,678 Irish claims, peak gusts of 185 km/h. The specification accepts the band €301–316m and the gate refuses to pick one figure, because neither source has that precision — a calibration target quoted to the million out of a disagreement between sources would be false precision wearing a gate.The flood referent is the OPW’s CFRAM programme: published extents at the 5%, 1% and 0.1% annual exceedance probabilities — the yearly chance of a flood at least that severe. Agreement is measured as Jaccard overlap (the shared area divided by the combined area, 1.0 being identical maps), floor at 0.82.The pilot metric
How we count this
We count from the signal arriving to the decision carrying a gate verdict, not to the first dashboard refresh. A screen that updates quickly is not a decision.The falsifier: minutes that fall because the decision skipped the gate are not readiness, they are exposure taken faster. The share of event-time decisions carrying a gate verdict is reported beside the timing, and rising speed with a falling gate rate is recorded as a failure.The mechanism: the actual event, against the actual book, quarantined
Peril simulators and live feeds re-run the unfolding event against the portfolio, per tick, rather than against a scenario chosen last quarter.
Every simulated value is quarantined: it is marked wherever it renders, it can never reach a certificate or an export without the marker travelling with it, and it is reversible in one transaction. Simulation runs inside the decision moment without contaminating live state.
What the event room proposes — a moratorium, surge staffing, a cede notice — is cross-model checked and gated before it takes effect, and the proposal, the check and the ruling land in the same thread.
Feeds are treated as observations, never as settlement indexes. That distinction is on the page because it is the one that decides whether a parametric product is a product or a dispute.
SOV/services/nexus-sim-schema/src/lib.rs
What this path shows during a live event
Every number this path renders carries a chain_stage flag naming which stage of the financial chain produced it — exposure objects, damage curves, deductible and limit financials, exceedance-probability curves — so a screen showing a net loss always says which stage it came from.
This path tells you where the event is going and what is exposed to it, on the actual book, per tick, as the event unfolds.
REFUSED
When a feed is dead, the run that depends on it declares its staleness and the dependent decisions are flagged. Yesterday’s value is not quietly served as today’s.
During an event, the difference between stale and fresh is the whole decision. A feed that fails silently produces a confident answer about a storm that has already moved, and it produces it at exactly the moment nobody has time to check.
What you can do next
Details for support
OFFLINE_NO_COMPUTE
What you can run before the season
You can rehearse the governed path itself — the event room, the gate and the record — on a historical event, and see exactly which decisions would have carried evidence and which would not. That is a rehearsal of the decision process, not a loss estimate, and on this page the difference matters.
A person replies with two or three times. Not a sequence.