Health & verification

A cognition graph is only useful if you can trust it. ThinkingRoot continuously verifies its own contents and surfaces three signals: contradictions, stale claims, and grounding failures.

Health

Health is the result of running the graph through verification probes. The GET /api/v1/ws/{ws}/health endpoint returns a snapshot aggregating pass/fail counts per probe; the Console's Overview tab renders it as a status strip and the System Doctor card. Health is a real signal — the Console never shows "all systems nominal" without one behind it.

Contradictions

Two claims that assert incompatible things about the same entity or fact are contradictions. They're detected during admission and stored so they can be surfaced (gated by a confidence threshold). Contradictions are also what guard branch merges: by default a merge that would introduce a contradiction is blocked.

Staleness

A claim is stale when the source it was extracted from has changed since the claim was last validated (its last_rooted_at is older than a configurable threshold). Re-validation re-checks stale claims against current source bytes, so the graph reflects reality rather than a snapshot from weeks ago.

Gaps

The engine can also report gaps — questions it can see are relevant but can't yet answer from the graph (GET /api/v1/ws/{ws}/gaps). Gaps are the raw material for the engine's just-in-time capability acquisition and a useful prompt for "what should I feed it next?"