VALUE-06 — Regression workflow

[daniele21]

Goal

Turn the already-implemented compatibility-first regression engine and versioned threshold policy into an end-to-end product workflow: choose an explicit baseline/candidate and configured policy, obtain a typed PASS/FAIL/NOT_COMPARABLE/NOT_EVALUATED decision, and inspect exactly which rules/evidence produced it.

Delivery / REAL_ENVIRONMENT policy

Deterministic software slices may integrate into dev once repository-owned gates pass. Representative baseline/candidate evidence (EVID-004) is deferred until final software convergence and a fresh exact-head PRE_REAL on the final candidate dev.

Existing owners on dev

• regression/engine.py: explicit immutable baseline binding + per-dimension compatibility-first comparison;
• regression/policy.py: versioned exact-match threshold rules and typed policy decisions;
• automation.py + CLI regress / regress-ci: stable machine-readable gate and exit-code semantics;
• UI already lists baselines/policies and exposes raw comparison deltas;
• Compare UX contract requires compatibility reasons before deltas and forbids regression verdicts for incompatible dimensions.

Software slices

VALUE-06A — Policy-backed regression projection

Add one application/UI API projection that reuses BaselineRegressionEngine + apply_regression_policy for an explicit baseline binding, candidate Run and configured policy. Return policy identity, overall decision, per-rule typed state/reason and compatibility-owned comparison evidence. Do not duplicate threshold logic.

VALUE-06B — Compare policy outcome

When configured policies/baselines are available, Compare lets the user evaluate the selected candidate against an explicit policy-backed baseline and renders compatibility -> policy identity -> PASS/FAIL/NOT_COMPARABLE/NOT_EVALUATED -> rule rationale -> existing deltas. Raw comparison remains available when no policy is configured.

VALUE-06C — Deterministic product acceptance

Fixture/browser coverage proves PASS, FAIL and NOT_COMPARABLE product states and backend/frontend agreement. No representative claim.

VALUE-06D — EVID-004 representative regression evidence

After final software convergence, retain compatible and incompatible real baseline/candidate pairs on representative hardware and at least one versioned policy outcome where applicable.

Acceptance

• baseline is explicit and immutable; never implicit latest;
• policy id/version is explicit and rules are not frontend-authored;
• compatibility precedes every threshold verdict;
• incompatible dimensions never expose false deltas or FAIL/PASS verdicts;
• PASS/FAIL/NOT_COMPARABLE/NOT_EVALUATED match the existing regression policy owner;
• Quality / Performance / Resources remain separate;
• deterministic CI proves software semantics only; EVID-004 is required for real claims.

[daniele21]

VALUE-06 software convergence update:

• VALUE-06A/B/C are integrated on dev via PR feat: add policy-backed regression workflow #164.
• merge commit: d92a0cdf6b425ab5afa9ab6aa2fc75283ee5c148.
• exact-head integration/strong validation: run 33962958565 PASS.
• Built Product PASS included Python 3.12/3.13, frontend check/test, product E2E, PRE_REAL browser J0-J9, media verification, packaged full-product E2E and operations/E2E contract verification.
• post-merge dev validation run 33964777209 PASS and correctly reused tree-equivalent validation evidence for tree d3e12c89c2f99ffd5f3fe7be62f0fc72cb89f7ae instead of rerunning deterministic product gates.

Remaining scope: VALUE-06D / EVID-004 representative regression evidence only. Keep the issue open until the final candidate dev completes the consolidated REAL_ENVIRONMENT phase.