Bound Pages preview health checks to tolerate observed deployment propagation #396

Observed failures (not application/test failures)

• PR fix: preserve consumer Git rule precedence #392 Reports run39650547786, job119398066395: six HTTP404 checks from18:13:20 to18:13:45 UTC on2026-10-21, then the same preview became200; exact failed-job retry passed without a source change.
• PR fix(ci): distinguish pending status guard diagnostics #395 Reports run39654366180, job119411337272: six404s from18:38:48 to18:39:13. HEAD request at18:40:32 confirmedHTTP200 with Last-Modified18:39:15 (two seconds after failure). Exact failed-job retry passed, same HEAD6077972.

Single source / current contract

.github/actions/github-pages/verify-deployment/action.yml defaults attempts=6, delay=5; run.sh polls each configured URL. Legacy .github/workflows/reports.yml calls this action for main and PR checks. Native asset hash propagation was separately addressed by #364; do not duplicate or regress that implementation.

Proposed small correction

Use a bounded propagation budget for this health-check action (e.g. 12 checks,10s interval), with explicit curl connection/total timeouts, keeping final nonzero failure for missing/broken deployments. Preserve configurable attempts/delay and validate bounded positive inputs. No infinite retry, workflow reruns, publishing changes, credentials/permission changes, or status success without actual successful HTTP response.

Acceptance

Deterministic fake curl/sleep tests:404→200 succeeds; persistent404/500 and network timeout exhaust budget and fail; success exits immediately; multiple paths checked; input validation rejects zero/malformed/unbounded values; timeout options passed. Document timing budget, retain failure annotations and gate behavior.

This is a robustness improvement backed by two propagation incidents, not a diagnosed application defect. Related historical #70 and native-only #364; separate from guard diagnostics #394/#395.