Dispatch SLO Glossary

This page owns the SLO vocabulary for issue-dispatch reports and runbooks. Use these words exactly in status rows, dry-run tables, close-arm notes, and operator summaries.

spawned

A worker process was actually launched for one issue. Dry-run rows that only say would_spawn are not spawned.

attempted

A dispatch or close arm took a real action for an issue: spawning a worker, auditing a candidate SHA, closing, reopening, or requeueing. A skipped candidate is not attempted.

witnessed

An external check the worker did not author confirms the claimed effect. Accepted witnesses include a passing parent-rerun test, dos commit-audit with a diff witness, dos verify, or an independent read-back.

closed

The GitHub issue is closed after the close arm reverified the resolving SHA. A worker saying it is done, or a local commit mentioning #N, is not closed.

retried

A later dispatch attempt intentionally reuses an issue after a failed, blocked, stale, or unwitnessed attempt. A retry must carry a new worker/run id and preserve the prior witness or blocker row.

stale

The remembered status is older than the source it summarizes: git, the issue state, a worker lease, a run ledger, or a commit-audit result. Stale rows are refreshed before they drive spawning, closing, or throughput math.

unsafe-to-dispatch

The issue must not be handed to an autonomous worker in its current form. Typical reasons are missing route/scope/witness fields, a private boundary, dependency hold, duplicate-risk hold, live-worker overlap, self-source guard hold, or an operator gate.