Same-page WebAV and Diffusion benchmark

The competitor harness runs all engines in the same Chromium page, device profile, media set, warmup policy, sample count, and measurement clock. It records environment identity and raw observations so a summary cannot be detached from its source evidence.

Covered scenarios include cold initialization/first frame, repeated warm seeks, rapid seek supersession, steady preview/playback where comparable, memory/resource observations, and cleanup. An unsupported scenario is reported as unsupported, not converted into zero time or removed from the comparison.

Interpreting results

• Compare the same semantic work and output dimensions.
• Separate cold setup from warm steady state.
• Use median and tail latency together; inspect raw samples.
• Treat browser scheduling noise and codec/backend choice as part of the environment.
• Do not extrapolate a single clip to long timelines, effects, audio, export, or mobile.
• Re-run after dependency, browser, hardware, or implementation changes.

The repository benchmark report is evidence for that exact revision and environment, not a permanent universal ranking. Use Performance and resource budgets to define product-specific acceptance.