Skip to content
Decision-Evidence Operating System

Line of business · General insurance, or P&C

Your volume lines already run on models. The routing that decided a claim is the part of the file you cannot produce.

A motor claim is scored, a lane is chosen and an outcome is issued before a person reads the file. Six months later a complaint lands, and the reasoning sits in a triage engine, a queue tool and an adjuster’s inbox — three systems that were never asked to agree on what happened.

What breaks in a P&C book today

Triage is a queue, not a decision

The score that sent a claim down the lane where nobody looked at it is a side effect of a job, not a recorded choice. When the complaint arrives, the routing is the part nobody can produce.

The fastest lane has the least oversight

Straight-through processing releases an effect on one model’s answer. Nothing re-derives it, and nothing holds it if the answer is wrong.

Saturation is silent

When the queue cannot meet the deadline, the work does not announce that it is late. It simply becomes late, and the breach surfaces in a complaint rather than on the day.

Explanations are written afterwards

The answer to “why was this denied?” is composed from the outcome, not derived from what the decision actually used. A paraphrase is not an explanation, and it does not survive being checked.

What holds today, and how you check it

2heterogeneous models that must agree before an effect is released
35 of 102cross-model gate elements proven live in our own deployment
1named human recorded on every adverse ruling, at a durable waitpoint
2 December 2027the date the Annex III obligations begin applying to insurance risk assessment and pricing — deferred from 2 August 2026

How we count this

“Proven live” counts the gate elements we have exercised against the deployed service, not the elements the schema defines. We print the ratio rather than rounding it up.NCT/services/nexus-workflows/src/services/tool-executors/sentinel-executors.tsThe named human is the approval waitpoint in the dispatch spine, not a notification sent after the fact.NX/services/nexus-orchestrator/src/routes/dispatch-routes.tsRegulation (EU) 2024/1689 names insurance risk assessment and pricing in Annex III; Regulation (EU) 2026/1744, the Digital Omnibus on AI, moved when the obligations bite — from 2 August 2026 to 2 December 2027. It is printed as a date rather than a countdown, so it does not decay on this page.

Which pillars answer it

P1 · for the Chief Underwriting Officer

Convening Triage + Graduated Escalation

A meta-decision agent scores each submission or claim and convenes the right room, the right chair and the right autonomy tier — and the routing decision is itself a logged event.

The score that chooses the lane, and the autonomy tier it grants, are written down as events — so the routing can be argued about a year later.

NX/services/nexus-orchestrator/src/routes/dispatch-routes.ts

P2 · for the Head of Claims

The Chaired Deliberation Chamber

Versioned expert personas argue the case from different angles — including one seat whose only job is to attack the conclusion — and a named human chair rules, at a durable waitpoint.

The contested minority get a room whose seats must argue from different evidence, and a named chair who rules at a durable waitpoint.

PC/src/routes/rooms.routes.ts

P3 · for the Chief Risk Officer

Cross-Model Computed Gate → Sealed Certificate

Two or more heterogeneous models re-derive the answer. A computed — not learned — agreement predicate must pass before anything is released, and the inputs and model versions are sealed into the record.

Nothing releases on one model’s word. Two heterogeneous models re-derive the answer and a computed predicate must pass first.

NCT/services/nexus-workflows/src/services/tool-executors/sentinel-executors.ts

P5 · for the Chief Compliance Officer

One-Log, Many-Regulator Evidence Fabric

One governed event stream compiles into every regulator’s artefact, instead of four teams reconstructing four different stories from the same week.

The stream that ran the claim is the stream a supervisor reads. You do not rebuild the story; you render the one you already have.

NX/services/nexus-workflows/src/services/governance.ts

See it in the console

What runs today, in a P&C book

Two dispatch tiers — carrying inline work and callback work — run in production. One governed event stream sits behind the record, so what happened is captured once, at decision time, and read by whoever needs it.

Start with one claim type, beside what you run today

You can point the decision seam at a single claim type and run it in parallel with your existing process — no rip-and-replace, and no change to the core system underneath. What you get back is the record of every decision that lane made, in a form you can hand to someone who was not there.

Get the free assessment

About four minutes. You keep the inventory either way.

Book the 20 minutes