Skip to main content
An outcome is the governed claim a completed workflow run makes: a headline and summary, a claimed status, metrics and exceptions, each backed by evidence references. Outcome reads require runs:read, decided like every run read: the caller must be able to read the run that produced the outcome at the run’s home and at every company on it.

What an Outcome carries

The claim (headline, summary, claimedStatus, period), metrics and exceptions with their evidence references exactly as stored, humanReviewStatus, the Judge evaluation summary, the recorded token cost of the attempt and of every attempt up to it (with its coverage), and the latest person rating value and time. Evidence references (step_output, approval, file) are not content: read file bytes through Files and step outputs through Get run, each under its own permission. The origin (chat channel, session and message, the launching principal), human decisions, the trust receipt, project and objective names, who rated and the rating notes are never returned.