G-Nosis is built on a schematised, bitemporal, append-only event log of decisions. No tribal knowledge. No "interpret the comment field". Twenty-seven edge types, versioned, addressable, immutable.
Proposals, rules, exceptions, approvals, citations, mandates, breaches, cures, executions, reversals. Each is a typed event with a schema-v5 contract: identifier, actor, source system, transaction time, valid time, schema version, and a structured payload validated at write.
The structured payload is not optional. Free-text notes are an audit liability and an extraction tax we refuse to pay later. Every claim about a decision is a field, not a sentence.
All twenty-seven, alphabetised. Each is a directed edge between two typed events. Schema-versioned. Index-backed.
Each event records when we wrote it and when it was true of the world. The two rarely match. A trade booked at 09:14 GMT with effective date T+2 has a transaction time of 09:14 and a valid time of T+2 settlement. A rule patched to RULES-V5 §sector-concentration in May 2026 retroactively re-evaluates a January 2024 decision — without rewriting history.
The log only grows. If a decision is wrong, the correction is a new event with a typed REVERSED_BY edge. If a fact is updated, the new fact is written with a fresh valid time. The prior fact stays where it was. There is no "edit history" tab to inspect, because there is no edit.
This is not a stylistic choice. It is the only design that makes the log regulator-grade, SOX-friendly, and forensic.
RULES-V5 is the rule layer above the event log. Each rule is a named, versioned, decidable predicate over the graph — written in a constrained DSL, reviewed like code, deployed like an API.
A decision passing the predicate gets a FIRED_BY_RULE edge, the rule version, and a re-pricing hook. A future schema bump re-runs the predicate retroactively against the old facts; the trace stays consistent because the version was bound at write.
Everything else is sugar. Connectors translate OMS / risk / compliance source systems into typed write calls. The trace API returns a graph slice at a chosen schema version.
Onboarding always starts by sitting with the desk and writing the v0 of your decision schema. That conversation is the first decision G-Nosis records.