iter8 is the final iteration. Locks Track E (carried iter1..iter7)
as the empirical-anchor closing artifact: measure CPU%, drops,
frame timing, GPU freq, memory across four consumer configurations
(mpv DMA-BUF, mpv vaapi-copy, Firefox-fourier, SW baseline) on
bbb_1080p30_h264.mp4 against the iter7-end driver.
Why now: iter1-iter7 prioritized binary blockers; measuring a broken
decoder is useless. iter7-end driver is the first stable substrate
where numbers don't drift between consumer probes.
Why this matters even without upstreaming (D dropped 2026-05-06):
- Personal regression detection for any future fork change
- Realism check on the campaign's own qualitative claims
- Calibration for follow-on campaigns (fourier-fresnel will
compare RK3399 numbers against this anchor)
Phase 1 success criterion (5 parts):
1. Reproducible script in tests/
2. Anchored numbers in a campaign artifact
3. Honest qualitative interpretation (no spin)
4. Phase 5 sonnet review confirms script is fixture-agnostic
5. Campaign close doc states "campaign closes"
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
