root/fusion
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

FN-6617: add weekly test velocity baseline

Browse Source 
Adds a report-only weekly baseline for test feedback-loop velocity.

- Add a test:velocity script that measures gate, boot smoke, and changed-only test durations and regenerates a #leads-ready report.
- Track committed baseline history, slowest-file summaries, and quarantine deletion-clock counts.
- Document the weekly refresh workflow in testing guidance and agent instructions.

Files changed:
 AGENTS.md                                         |   4 +-
 docs/test-velocity-baseline.md                    |  92 ++++++
 docs/testing.md                                   |  18 ++
 package.json                                      |   1 +
 scripts/__tests__/test-velocity-baseline.test.mjs | 121 ++++++++
 scripts/test-velocity-baseline.mjs                | 339 ++++++++++++++++++++++
 scripts/test-velocity-history.json                | 231 +++++++++++++++
 7 files changed, 805 insertions(+), 1 deletion(-)

Fusion-Task-Id: FN-6617

Fusion-Task-Lineage: 50a07298-74ec-4481-b288-4fadb7d3d29b
This commit is contained in:
gsxdsm
2026-06-17 20:09:05 -07:00
parent a013bc0309
commit fbd9ad2f3b
7 changed files with 805 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
docs/testing.md
View File

@@ -225,6 +225,24 @@ shard artifacts into `.timings/` first (the default lookup directory), or pass
scheduled job can gate on freshness via `node scripts/ci-test-shard.mjs --check-timings-staleness`,
which exits non-zero when the snapshot is missing or older than the 30-day budget.
## Weekly test velocity baseline
FN-6612 tracks feedback-loop velocity as signal-per-second, not as a new blocking gate. Refresh the weekly baseline from a clean worktree with:
```bash
pnpm test:velocity -- --measure --write-report
```
The script runs `pnpm test:gate`, `pnpm smoke:boot`, and `pnpm test` with bounded async process supervision, then appends the measured row to `scripts/test-velocity-history.json` and rewrites the postable artifact at `docs/test-velocity-baseline.md`. It reads the slowest 20 files from the committed `scripts/test-timings.json` snapshot and the flake/quarantine count plus 14-day deletion-clock buckets directly from `scripts/lib/test-quarantine.json`; do not run the full suite just to populate the slowest-file table.
Use cheap report-only regeneration when measurements already exist:
```bash
pnpm test:velocity
```
Each week, copy the `Post to #leads` block from `docs/test-velocity-baseline.md`. If a measured command fails because the local environment is not ready, keep the failure recorded in the report instead of fabricating a time, then fix or rerun separately as appropriate. Do not wire `pnpm test:velocity`, `test:full`, or any slow-suite expansion into PR checks; the merge gate stays the thin Lint, Typecheck, Build, and Gate path.
## Targeted commands
```bash