Release History

E2E skip audit — re-qualified all 163 skips from a fresh full-suite baseline — Requested full-suite E2E refresh (334 passed / 0 failed / 163 skipped), then a review of whether each skip was a genuine state/time-of-day dependency or masking a defect. Grouped skips by spec+project; several files were skipping 100% of their own tests, which a truly state-conditional skip design should essentially never do.

Root cause: Locator.isVisible({timeout}) doesn't poll in Playwright — The timeout option is deprecated/ignored — the call returns an instant DOM snapshot, not a real wait. Confirmed via skip-duration data (resolved in 1-2.5s despite a stated 15-30s "timeout") and reproduced consistently. Traced via git blame to commit e3d4d85a (2026-06-07, "Sprint AF Unit 4"), which implemented the pattern exactly as specified in aidlc-docs/aidlc-state.md's own plan for that unit — the plan itself was wrong, not just the code, trying to fix real Keycloak-redirect flakiness with an API call that doesn't actually wait.

Fixed across 18 spec files, 5 commits — 49071bf0 — branch-ops.spec.ts, transactions-gl-columns.spec.ts, admin-local-rates.spec.ts (the 3 files false-skipping 100% of their tests). 7202bbaa — a strict-mode locator bug the first fix surfaced: branch-ops.spec.ts:180's date regex was page-wide and started colliding with ProofIn/ProofOut/TellerProofIn "Submitted counts (date)" text added in later sprints; scoped to the Session step card instead of using .first() (which would have silently matched the wrong card). cbb97d0b — branch-ops-proofin-icd.spec.ts + branch-ops-proofout-icd.spec.ts, confirmed via _proof-count-section.tsx that the section renders unconditionally so there was no real state gate being tested. 506529c7 — the remaining 73 call sites across 13 more files, mechanical swap to .waitFor(). Highest-risk find in that batch: admin-bank-admin-gl-pages.spec.ts's loadPage() gated ~26 tests behind the same broken pattern, right next to its own comment warning that GL pages can take 20-40s to cold-compile under load.

Final E2E baseline — 497 total / 362 passed / 0 net failed / 134 skipped — Verified across 5 full OCI-VM runs (one per fix, to isolate each change's effect). 27 tests recovered real, trustworthy coverage that had been silently masked as clean skips. The suite's remaining skips are now all genuinely state- or time-of-day-conditional.