You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Repair current-main specialized CI baseline so unrelated PRs are not born red #391
Recent narrow PRs inherit many red specialized workflows even when their focused contracts pass. Example exact head #384 is green on candidate smoke, package graph, frozen fetch, copy-mode OCI, polyglot, standalone locked install, durable first-install, R2 E2E, agents, formal review and ores-lint, while portability, several Nix/offline/export lanes, repository hardening, global profiles, process locks and normal CI are red.
Task:
reproduce every red specialized lane on current main and classify source regression vs stale baseline vs infrastructure/admission failure;
repair true main regressions instead of waiving them on feature PRs;
where a baseline is intentional, encode a reviewed exact baseline with new-occurrence detection;
make status output identify baseline-vs-candidate deltas explicitly;
add a meta-check that a no-op/current-main candidate does not start with unexplained red required lanes.
Acceptance: a narrow PR can distinguish its own regressions from main debt without disabling specialized checks.
Activity