✅ Phase-5 – Automated regression, spectrogram diff & cross-browser stress tests

[ford442]

Parent: #972
Blocked by: #977 (Phase-4)

Why

We need objective, repeatable proof that the new high-fid engines are both more authentic and safe to use across browsers.

Scope

1. Spectrogram Diff / Audio Quality Tests

   • Load Phase-0 reference + renders from stock, highfid-cpu, and gpu-highfid.
   • Compute error metrics (MSE per band, peak, etc.) and assert against the thresholds defined in the authenticity-gaps doc.

2. Performance Benchmarks

   • Script that measures wall-clock time for representative offline renders at different oversample factors.
   • Record GPU dispatch + readback latency.
   • CI can later enforce soft budgets.

3. Cross-Browser Matrix (Playwright)

   • Chrome / Edge (WebGPU expected) – GPU path active.
   • Firefox / Safari (or equivalent) – fallback to CPU path, no crashes, correct UI badge / toast.

4. Stress / Memory

   • Concurrent offline renders via the worker pool; assert no OOM / crashes under reasonable load.

5. Manual A/B Checklist

   • Extend or create docs/audio-engine/303-A-B-checklist.md in the same style as the existing voice docs.

Acceptance Criteria

• Spectrogram / quality tests pass with the defined error thresholds.
• Benchmark numbers are captured (as CI artifacts or logged).
• Playwright (or equivalent) matrix covers the main browsers and shows correct engine selection / fallback.
• Stress test completes without OOM or unhandled rejections.
• Documentation and badges updated to reflect high-fid status.

Sub-tasks

• Add / reuse FFT utilities for spectrogram diff tests.
• Write the diff + quality test harness using Phase-0 reference material.
• Implement benchmark script.
• Extend Playwright config / add fallback-specific e2e tests.
• Stress test using the offline worker API.
• Update CI if needed to run the new jobs and upload artifacts.

Labels / Milestone

Intended milestone: M5 – Test Suite, Performance & Cross-Browser