Mission Run Overview
Mission Run: CI Pipeline Honesty — Actionable Fixes
Make CI verdicts honest: close four settled cases where the pipeline reports green work as red, or leaves real failures unrun or unreported.
Spec Kitty's agent fleet relies on CI as the release authority — green must mean actually-proven and red must mean a real regression (ADR 2026-07-17-1). This mission fixes four independent, settled cases from epic #4437 where that contract breaks today: (1) a diff-scoped PR whose selected shards all pass is falsely failed by the aggregate expecting all registry shards (#4360-B); (2) a tests-only change runs none of its own tests yet reports green (#4454); (3) timeout-killed jobs are indistinguishable from external cancellations at the router gate (#4208); (4) the nightly suite swallows real failures and reports green (#4212). No fix may widen the green path. The coupled main-tip fan-out cluster (#4347/#4371/#4360-A/#4430) is out of scope, handled via a separate design ADR.