SDLC metrics dashboard for the fullsend-ai organization. Tracks deployment frequency, PR lead time, PR volume, issue volume, community attribution, and PR delivery mix (conventional-commit type + fix-filer source) across repos. Published daily as a D3.js dashboard on GitHub Pages.
Dashboard: https://fullsend-ai.github.io/metrics/
Dedicated tab: Delivered PR Types (docs/delivered-pr-types.html).
Daily collector for merged PRs on fullsend + agents:
| File | Purpose |
|---|---|
docs/pr-type.csv |
Per day × repo counts by type (feat…other) and fix filer (fix_core / fix_external / fix_bot / fix_unlinked) |
docs/pr-type-details.csv |
Per-PR drill-down |
scripts/collect-pr-type.sh |
Daily (wired into .github/workflows/collect.yml) |
scripts/backfill-pr-type.sh |
Date-range backfill |
Core-team roster comes from docs/community-config.json (same list as community metrics).
# Daily
./scripts/collect-pr-type.sh # yesterday UTC
./scripts/collect-pr-type.sh 2026-08-11
# Backfill (rewrites CSVs)
./scripts/backfill-pr-type.sh 2026-05-14 2026-08-11See docs/design.md for the full design spec.
The Quality Signals tab
reports two GitHub PR-based proxies for fullsend and agents:
- Defect-labeled rate:
fixPRs linked to an issue with a defect label divided by merged PRs. - Revert event rate: PRs titled
Revert ...or typed subjects such asfix: revert ..., plus Git commits containingThis reverts commit; duplicate evidence and temporary changes reverted within the same merged PR are excluded before division by merged PRs.
The defect-labeled rate is an explicit-label proxy; it does not prove
preventability or distinguish a review-escaped bug from a missed requirement.
These are leading indicators, not DORA Change Failure Rate. DORA requires
production deployment and rollback/hotfix evidence. The daily workflow derives
docs/quality.csv from the PR-type datasets and GitHub issue/commit metadata.