mejbaurbahar/fagun/regression-testing

Regression Testing

Scoping a regression pass

Don't blindly re-run everything every time — scope by change impact:

1. Diff-driven scope: what files/modules changed? Trace which features depend on them (shared components, shared API endpoints, shared DB tables/migrations).
2. Always-run core: checkout/payment, auth/login, and whatever the business calls its "money path" — regardless of whether the diff touches them, run these every release as a floor.
3. Full regression only for major releases, schema migrations, or dependency upgrades touching shared infra (auth libs, ORMs, framework versions).

Building a regression suite

• Promote flows from manual/exploratory testing into automation once they've been run manually 2-3 times and are stable — see [[test-automation]].
• Tag tests by area/feature so partial regression runs are possible (@checkout, @auth, @admin).
• Keep the suite fast enough to run every PR (smoke tier, minutes) with a separate nightly/pre-release full tier.

Common regression sources to specifically re-check

• Shared components/design system changes — check every consumer, not just the one that prompted the change.
• Database migrations — check both new-record behavior AND existing legacy-data behavior (old rows without new columns populated).
• Dependency/library upgrades — check changelogs for breaking changes, not just "tests still pass" (tests may not cover the changed behavior).
• Config/feature-flag changes — verify both flag-on and flag-off states still work, and that the flag actually gates what it claims to.

Running it via this harness

qa regression

Runs the harness's regression suite through the qa CLI dispatcher; results land in ~/qa-agent/reports/.

Triage discipline

When a regression test fails:

1. Confirm it's a real regression (re-run once, rule out flakiness — see [[test-automation]] flaky-test triage) before filing.
2. Bisect to the change that caused it if not obvious (git bisect, or compare against the last known-good deploy).
3. File via [[bug-reporting]] with the specific commit/PR that introduced it if identifiable — this is the single most useful piece of info for a fast fix.