# Before → after, runs per PR push (measured 2026-09-30 ~16:30Z, fedechapa token, 236 calls)
- BEFORE (12 fede PRs merged Sep 22–26): median 32 workflow runs per push (28 excl. cancelled)
- AFTER (12 fede PRs opened today 14:44–15:56Z, pieces 1–15 mostly merged): median 15 runs per push (13 excl. cancelled) → −53%
- Machine-minutes: AFTER median 62 machine-min/push via /timing (52.6 excl. cancelled) ≈ $0.04/push at 2-cpu spot rate. BEFORE not measured with /timing (API budget); run-wall proxy not comparable. No valid before→after $ figure yet.
- Still firing on nearly every push (next fold targets): Auto-merge on green, Branch Preview Domain, CI Dispatch Guard, Nested Package CI, Claude Code Review, Claude Code, Review execution re-pin, Claude Review Check Override, Claude Review Recheck (9/12), NO-OP review debounce (3/12). ~10 satellites per push.
- Gone since baseline: PR body gate, file collisions, size tier, docs fence, self-review, Semgrep, Quality Gate POC, Dependabot gate, Lambda bundle, harness/eval workflows (4 of these are temporarily disabled today; re-enable tonight).
- Also seen today: the same head SHA got two pull_request CI runs 18 s apart (#14282), one cancelled the other → duplicate-run-per-push defect to fix after the freeze.