Routing

Two labels, because they answer different questions: maturity says how far it is built, availability says who can reach it. The authoritative, dated record is build status.

Deterministic rules first, a model classifier only where the rules are inconclusive, its own eval set and telemetry, and every failure path — timeout, invalid output, telemetry error — failing open to the ordinary response path.

It supports three modes:

• Off — the source default. In this mode it changes nothing about a response.
• Shadow — routes and records a decision without applying any Skill.
• Active — may apply only the approved Skills.

Eight Skill descriptors are registered and enabled in the routing registry — construction invoice audit, document evidence analysis, electrical troubleshooting, HVAC diagnostics, financial analysis, investment research, operational exception analysis and operational financial impact — declaring which others they compose with. Three of them, electrical troubleshooting, HVAC diagnostics and construction invoice audit, are currently approved for application when Router mode is active. The other five are registered but not applied in any mode.

The sixteen work modes in the product are separate: you select them yourself. Skills and work modes →

A routing decision says what a task appears to need. It is not permission to do anything. Broader orchestration — several capabilities contributing to one piece of work — is still being built. Model selection and risk scoring are design, not code.

Telemetry keeps the routing decisions, for evaluating the router. The eval sets and harnesses that measure it are internal, run offline and in continuous integration. Agent evaluation →

Research · 1 Sep 2026 — A skill-routing layer, with its own evaluation set. A routing layer landed in the engine together with an eval set for measuring it. Its source default is off, and in that mode it changes nothing about a response. News →