cicd-testinglisted

# CI/CD Testing ## Test tiering (don't run everything on every PR) - **PR gate (minutes)**: unit tests + fast API contract checks + a thin smoke E2E suite covering the money path. This blocks merge — keep it fast and reliable or people start bypassing it. - **Pre-deploy / merge-to-main**: fuller regression suite, still bounded (10-20 min). - **Nightly / scheduled**: full regression, cross-browser matrix, performance baseline, security scan (SAST/DAST) — anything too slow to gate a PR. - **Post-deploy smoke**: a small suite that runs against the just-deployed environment before traffic fully shifts, to catch environment-specific breakage config/secrets differences that staging didn't have. ## Quality gates Define explicit, machine-checkable pass/fail criteria per stage, not "look at the dashboard": - Merge gate: 0 failing required checks, coverage delta not regressed below threshold (if coverage-gating is used — see the caution about coverage-as-a-metric in [[qa-fundamentals]]). - Deploy gate: regression suite green, no new Critical/Blocker security findings ([[security-testing]]), perf within budget ([[performance-testing]]). - Make gates fail loud and specific — a gate that just says "checks failed" without linking the failing job wastes the next person's time. ## Parallelization & sharding - Split large suites across N workers by file/test-duration balancing (not just alphabetical file split, which produces uneven shard times) — most modern runners (Playwright, pytest-xd