Implement CognitiveAccessibilityAudit rules engine
epic-cognitive-accessibility-wizard-ui-task-014 — Build the CognitiveAccessibilityAudit utility class that programmatically checks a set of cognitive accessibility rules against any screen widget tree: single-action-per-screen enforcement, presence of a labelled back/cancel action, absence of more than 5 choices on a single screen, confirm-before-submit gate presence on wizard-submission routes, and inline help availability on all text input fields. Each rule returns a typed AuditViolation with severity and location.
Acceptance Criteria
Technical Requirements
Execution Context
Tier 5 - 253 tasks
Can start after Tier 4 completes
Implementation Notes
Implement CognitiveAccessibilityAudit as a pure Dart utility class (no Flutter rendering dependency) that receives a WidgetTester and uses `tester.widgetList` and `tester.getSemantics` for tree traversal. Use Finder composition (`find.byType`, `find.bySemanticsLabel`) to locate widgets without coupling to internal widget keys. Model AuditSeverity as a Dart enum with values high and low. For Rule 3 (choice count), define a const `kMaxChoicesPerScreen = 5` in the class to make the threshold configurable in tests.
For Rule 4 (confirm gate), check via `find.byType(ConfirmBeforeSubmitScreen).evaluate().isNotEmpty` scoped to wizard routes. Annotate the class with `@visibleForTesting` to communicate intent and prevent accidental production use.
Testing Requirements
Unit tests for each of the 5 rules in isolation using flutter_test: for each rule, construct a minimal widget tree that (a) passes the rule and (b) violates the rule, assert correct AuditViolation output. Test that AuditViolation.severity is AuditSeverity.high for rules 1–4 and AuditSeverity.low for rule 5. Test the combined `audit()` method with a widget tree containing multiple simultaneous violations and verify all are reported. Test that a clean wizard screen with all rules satisfied returns an empty list.
The WizardStateManager BLoC must guarantee that step transitions only occur on explicit user action, never automatically. Subtle reactive patterns in Bloc (e.g. stream listeners triggering add() calls) could inadvertently auto-advance the wizard, violating the core cognitive accessibility rule and creating a regression that is difficult to detect without dedicated tests.
Mitigation & Contingency
Mitigation: Write a dedicated unit test that subscribes to the BLoC stream and asserts no StepChanged event is emitted for 5 seconds after a state update, without an explicit user-sourced event being dispatched. Make this test part of the CI gate for the WizardStateManager.
Contingency: If an auto-advance regression is discovered post-integration, introduce a mandatory UserActionToken parameter on all step-transition events so the BLoC can structurally refuse transitions that do not originate from a user gesture handler.
The ConfirmBeforeSubmitScreen requires deep-linking back to specific wizard steps for corrections. Implementing bidirectional navigation within a multi-step wizard while preserving all previously entered state is architecturally non-trivial and may conflict with the existing StatefulShellRoute navigation setup described in the app architecture.
Mitigation & Contingency
Mitigation: Design the ConfirmBeforeSubmitScreen's back-navigation links to dispatch a GoToStep event on the WizardStateManager rather than using GoRouter's pop() chain. This keeps navigation state entirely in the BLoC and avoids coupling to the router's stack semantics.
Contingency: If BLoC-driven step navigation proves incompatible with the router, implement the correction flow as a dedicated sub-route that pre-populates its form from the WizardStateManager's current draft, then merges the edited field back into the draft on completion before returning to the confirm screen.
The CognitiveAccessibilityAudit utility must inspect live widget trees for violations such as icon-only buttons. Flutter's widget tree inspection APIs are available in test mode but have limitations in identifying semantic intent (e.g. distinguishing a decorative icon from a navigation button). False negatives could give a false sense of compliance.
Mitigation & Contingency
Mitigation: Augment the audit with a convention-based approach: require all navigation buttons to use a named wrapper widget (e.g. LabelledNavigationButton) that the audit can detect by type, rather than relying solely on widget-tree semantics analysis.
Contingency: If widget-tree detection proves insufficiently reliable, scope the CognitiveAccessibilityAudit to route-configuration analysis (verifying back navigation availability per route) and static analysis of wizard step count definitions via the CognitiveLoadRuleEngine, which provides deterministic results.
The InlineContextualHelpWidget sources content from a bundled JSON asset via the HelpContentRegistry. If help texts are missing for newly added screens or fields (a likely scenario as the 61-feature app grows), the widget silently shows nothing, degrading the accessibility experience without any visible failure.
Mitigation & Contingency
Mitigation: Integrate a CognitiveAccessibilityAudit check that verifies every registered (screenId, fieldId) pair that requests help has a corresponding entry in the HelpContentRegistry bundle. Run this check in CI as part of the audit report.
Contingency: Add a debug-mode overlay that highlights fields with missing help entries using a visible warning indicator, making coverage gaps immediately obvious to developers during local development before they reach CI.