KindcapabilityValue3Unitconsecutive clean runs before expanding scope
NotesThe only quantified promotion gate found. Paired with a named metric set: completion rate, correction minutes, false-positive rate, source freshness, approval count, failed runs, and cost per accepted result.
Expand only when the Bot completes at least three consecutive runs with source-linked output, no unauthorized actions, predictable cost, recoverable failures, and less human correction time than the original workflow.