Release gates and measurable results
These gates are proposed product-governance rules. Passing them does not automatically establish legal compliance or independently verify all WCAG conformance requirements.
Gate 0 — Scope integrity
Require the target version/level, build, page and complete-process inventory, relevant roles, dependencies, states, environments and owners. An empty inventory blocks evaluation. Any limited sample or untested environment must remain visible in the report. [S-CONFORMANCE]
Gate 1 — Requirement coverage
Review every one of the 55 A/AA registry entries for applicability. Expand applicable criteria into concrete assertions and instances. A registry row cannot receive a PASS merely because one example page passes. Record reviewed absence or exception decisions explicitly.
Gate 2 — Evidence adequacy
Require current-build evidence for all planned required assertions. Complete the necessary manual reviews and all required critical-process tests. Invalidated, missing or inaccessible evidence blocks readiness.
Gate 3 — Failure handling
IF any confirmed applicable A/AA assertion fails
THEN NOT_READY_FOR_AA_SIGNOFF.
ELSE IF any required assertion is NOT_TESTED, NEEDS_REVIEW or BLOCKED
THEN INCOMPLETE_ASSESSMENT.
ELSE IF any required product policy fails
THEN POLICY_GATE_FAILED, while keeping the WCAG verdict separate.
ELSE IF scope/coverage review has not been approved
THEN NEEDS_SCOPE_REVIEW.
ELSE READY_FOR_ACCESSIBILITY_SIGNOFF_FOR_DECLARED_SCOPE.A single minor A/AA failure cannot be canceled out by many passing checks. Triage severity determines work order, not whether a known WCAG failure disappears. A management risk acceptance must remain distinct from a conformance assessment.
Gate 4 — Independent review and publication
An accountable reviewer checks scope, coverage, exception logic, third-party steps and evidence limitations. A public claim requires its own correctly scoped conformance review; do not publish a claim directly from a scan score or the readiness flag. Full pages, full processes, accessibility-supported use and non-interference remain part of that review. [S-WCAG22]
Accessibility is also not the whole of production readiness: security, authorization, transaction correctness, performance, resilience and operational readiness require their own gates.
Metric definitions
Freeze a named assessment scope before comparing numbers. Let N be all planned required test instances; P verified passes; F confirmed failures; A reviewed not-applicable instances; U unresolved instances. Require N = P + F + A + U after deduplication.
| Metric | Formula or measure | Decision rule |
|---|---|---|
| Evaluation coverage | (P + F + A) / N | Report N and U; if N = 0, undefined and blocked, never 100%. |
| Tested applicable pass rate | P / (P + F) | If denominator = 0, undefined; always display U alongside it. This is not a conformance score. |
| Unresolved work | U, broken down by verdict | Required unresolved work must reach zero for signoff readiness. |
| Open conformance failures | F, with criterion and task impact | Must reach zero for an A/AA readiness claim in the declared scope. |
| Critical-journey completion | Passed required journey/environment instances divided by planned required instances | No skipped/blocked run counted as passed. Require full planned completion. |
| Evidence coverage | Required obligations with reviewed current evidence divided by required obligations | Missing observations block signoff. |
| Regression count | Previously repaired failures reproduced in the new build | Fix and identify propagation source. |
| Repair duration | Time from confirmed finding to independently verified fix | Set a product service target; not a WCAG threshold. |
| Barrier arrival rate | New confirmed unique defects per comparable release/week | Track alongside release size and tested scope. |
| Backlog accumulation | Open confirmed defects now minus previous comparable window | Do not improve the metric by removing scope or relabeling failures. |
Example arithmetic only: 90 passes, 2 failures, 3 reviewed N/A and 5 unresolved yields 95% evaluation coverage for 100 instances. It still fails the release gate. The example is not a measurement of any real website.
Priority is separate from conformance
Suggested work order: unsafe or task-blocking interactions; widespread shared defects; inaccessible error/recovery paths; remaining confirmed failures; then enhancements. Record impact, reach and reproducibility separately. A severity label or custom weighted priority is not part of WCAG A/AA/AAA levels.
Regression maintenance
For every merged repair, retain a test that fails on the old behavior and passes on the repaired behavior, where technically feasible. Add a manual regression protocol when automation cannot measure the outcome. Changes to the grammar itself require versioning and re-evaluation of dependent rules.