[OA-017] Measure four-source behavior and close accessibility gaps #17

Planned implementation slice OA-017. No application behavior is implemented by this issue.

Phase: P5 · Dependencies: OA-011, OA-012

Full roadmap detail · Contracts

Requirements

OA-SRC-002, OA-SRC-004, OA-AI-001, OA-AI-002, OA-AI-005, OA-CTL-003, OA-ACC-001

Target files

• tools/benchmarks/four-source.ts
• tests/e2e/monitoring.spec.ts
• tests/accessibility/dashboard.spec.ts
• docs/benchmarks/reference-host.md
• docs/testing-release.md

Acceptance criteria

• Measure four-source 720p target previews, 2-second eligibility, one local request, frame age, queue depth, and resource usage on a named reference class.
• Publish achieved results and delays; eligibility is not an analysis SLA and no claim of every-frame detection is made.
• Keyboard, focus, contrast, labels, screen-reader announcements, 200% zoom, and reduced-motion checks cover core paths.
• Stop/permission revocation, one-source outage, provider timeout, hot-unplug, suspend/resume, and a 30-minute soak have evidence. Stop blocks new work immediately, acknowledges/ceases acquisition within 1 second, and tears down app-owned resources within 2 seconds on the reference scenario.

Verification

• Synthetic deterministic scheduler/load suite on all CI OS targets.
• Windows native opt-in run records preview and analysis separately, including stop latency and released handles.
• Playwright/axe automation plus manual keyboard/screen-reader session; findings are tracked before release.

Gate and estimate

G5a: reference baseline, stop behavior, and core accessibility evidence accepted.

4-8 first-pass work-unit hours; NOT an alpha completion estimate; assumption: 8 bounded 30-60 minute work units, prepared environment and fixtures; integration waits/rework excluded

Handoff

Record changed files, actual commands/results, mock versus live evidence, residual limits, and review verdict. Do not connect personal feeds or accounts in CI.

Bounded work units

• Create benchmark recorder (30-60 minutes)
• Add four-source synthetic scenario (30-60 minutes)
• Test provider/source fault combinations (30-60 minutes)
• Measure stop and handle cleanup (30-60 minutes)
• Run Windows reference benchmark (30-60 minutes)
• Run accessible keyboard flows (30-60 minutes)
• Run screen-reader/zoom review (30-60 minutes)
• Publish results and residual gaps (30-60 minutes)

Linked dependencies

• OA-011
• OA-012