Question
Build ticket (decided by #296): retire Tier A — the per-merge CI trend — in one slice.
Delete:
.github/workflows/bench.yml
scripts/bench_points.py, scripts/bench_cold_reset.sh
test/test_bench_workflow.py, test/test_bench_points.py, test/test_bench_doc.py
dev/bench/ on the gh-pages branch — a frozen chart that looks live is a trap; the 92 published points stay reachable in that branch's git history.
Keep (Tier B and C, per #296): scripts/bench_launch.py, test/fixtures/bench_harness.py, rust/devlaunch-core/src/timing.rs, the frozen stage names in public-api.txt.
One check the tier table glossed: test/test_bench_record_schema.py was tallied as Tier A, but it guards the record bench_launch.py --record writes — Tier B output. If its assertions bind the kept harness's record format, it stays; delete it only if they are trend-specific. Default: keep.
Docs ride along in the same PR: docs/performance.md (loses its doc-guard when test_bench_doc.py goes — rewrite its trend/chart sections to describe on-demand measurement via the local harness), the README's mention of the trend/chart if any, and CLAUDE.md's guard list (it names test_bench_doc.py reading docs/performance.md). Any bench-related pixi tasks that only fed the trend go too; bench itself (Tier B) stays.
Blocked by #502 — the e2e foreign-uid case lands first, so the #240-class coverage never lapses.
Worked via wf-tdd, reviewed via wf-review, per the map's Notes.
Question
Build ticket (decided by #296): retire Tier A — the per-merge CI trend — in one slice.
Delete:
.github/workflows/bench.ymlscripts/bench_points.py,scripts/bench_cold_reset.shtest/test_bench_workflow.py,test/test_bench_points.py,test/test_bench_doc.pydev/bench/on thegh-pagesbranch — a frozen chart that looks live is a trap; the 92 published points stay reachable in that branch's git history.Keep (Tier B and C, per #296):
scripts/bench_launch.py,test/fixtures/bench_harness.py,rust/devlaunch-core/src/timing.rs, the frozen stage names inpublic-api.txt.One check the tier table glossed:
test/test_bench_record_schema.pywas tallied as Tier A, but it guards the recordbench_launch.py --recordwrites — Tier B output. If its assertions bind the kept harness's record format, it stays; delete it only if they are trend-specific. Default: keep.Docs ride along in the same PR:
docs/performance.md(loses its doc-guard whentest_bench_doc.pygoes — rewrite its trend/chart sections to describe on-demand measurement via the local harness), the README's mention of the trend/chart if any, and CLAUDE.md's guard list (it namestest_bench_doc.pyreadingdocs/performance.md). Anybench-related pixi tasks that only fed the trend go too;benchitself (Tier B) stays.Blocked by #502 — the e2e foreign-uid case lands first, so the #240-class coverage never lapses.
Worked via
wf-tdd, reviewed viawf-review, per the map's Notes.