FFmpeg release patch lifecycle evidence¶

The previous checker could succeed after fetch failure and reused a checkout with reset/clean. Patch 0018 duplicated percentile cases from patch 0005 and failed cumulative replay against n9.0.1. The required CI aggregator did not cover full-series replay. These are source and execution findings from 2026-09-08, not inferred release readiness.

The repaired series applies all 18 patches to the official released tag n9.0.1, commit bf1b838f2ab88b4f8fd83443325c782ea0e0f7fa. The extracted pool mapper compiles and passes checks with and without percentile support. The maintained patch automation has real Git fixtures for stable tag selection, rebase, idempotence, network and conflict failures, inherited hook variables, hostile global Git configuration and rollback after a replacement failure. Workflow contracts check impact routing, required registration, failure propagation through diagnostic tee and scheduled-only upstream discovery.

Reproduce the fixture suite with:

python3 -m unittest discover -s scripts/ci -p 'test_ffmpeg_patch*.py' -v
python3 scripts/ci/ffmpeg_patch_stack.py --check

This evidence validates patch maintenance; it does not replace the FFmpeg integration builds, numerical golden gate or complete release acceptance. See ADR-1240 for alternatives.