Skip to main content

Traceability And Evidence

FST records why a route was returned.

That evidence is what makes agent work inspectable after the fact.

What Evidence Records​

Evidence can include:

  • requested action
  • payload hash
  • actor
  • run id
  • profile id, version, and hash
  • gate id and type
  • route
  • reason and instruction
  • missing artifacts
  • required approval
  • created artifacts
  • hook input and output hashes
  • materialization plan and result

Agent Claims Are Not Enough​

Invalid:

The agent says the user approved.

Valid:

An approval record exists from a trusted approval surface, scoped to the action.

Invalid:

The agent says it ran the checks.

Valid:

A test_run_artifact records the command, result, run id, and profile version.

Replay​

fst replay show --latest should explain:

  • which profile version controlled the run
  • which action was submitted
  • which gate fired
  • why the route was returned
  • which artifacts or approvals were missing
  • which evidence refs were written

Replay is not logging for curiosity. It is how FST proves the route decision was based on controlled state.