[difficcd]

A first Measure → Analyze → Change cycle on the development process itself. It starts from a written review of how this project is built — incremental development run on XP practices, checked practice by practice, ending in a risk plan and five steps. The review ordered those steps with the measurement baseline first: it is the cheapest, and every other step is judged by it. So that is where this began, and the changes below were chosen from what the numbers showed, not from the list.

The full record is docs/process/2026-10-14.md.

Measure

scripts/process-metrics.mjs reads the DevOps measures straight out of the repository's own records — merged PRs, issues, Actions runs, git history. Nothing to install; rerunnable at the end of each cycle, with --json saved beside the record for diffing.

measure	all history	last 30 days
deployment frequency	32.8 / week	39.7 / week
automated test pass ratio	96.9%	99.7%
defect escape rate	43.2%	46.5%
fixes carrying a test	43.6%	60.4%

Three corrections were needed before the numbers could be trusted — 55 PR titles written as prose and counted as nothing, App.jsx/App.tsx split across two names, translation tables ranking as "hotspots" — and the escape rate moved from 48.5% to 43.2% because of them. The record lists each.

Analyze

• CI says yes; defects ship anyway. 97–99.7% pass while nearly one fix lands per two changes.
• They land in one file. App: 63 fix-touches. The next: 11.
• Test-after, measured. 56% of all fixes landed without touching a test.
• CI ran four of its ten gates. ci.yml claimed to run npm run check and instead listed four steps by hand. Every gate added locally since — helper index, unreachable, unused imports, stroke writes, native dialogs, i18n — bound only on a machine where someone ran it. CONTRIBUTING said "npm run check is what CI runs"; it was not.
• The map was 29% missing. ARCHITECTURE.md named 108 of 152 files under src/; export/ had no section.

Change

	change	expected to move
C1	scripts/fix-needs-test.mjs — a branch with a fix: commit must change test/, or carry a No-Test: trailer with a reason	fixes carrying a test: 60% → ~100%
C2	scripts/architecture-index.mjs — every src/ file must be named in the map; the 44 added	stays 152/152
C3	CI runs npm run check as one step — one list cannot drift from itself	pass ratio may dip at first; that is six more gates counting
C4	needs-verification label on #338 and #4, counted by the metrics	read 0 because nothing recorded the wait; now 2, should reach 0

The expected directions are written into the record now, so the next measurement is checked against a prediction rather than fitted to it afterwards.

Verification

• npm run check green locally: 1312 tests; both new gates run in it.
• The fix-needs-test gate checked end to end on a throwaway branch, not only through its unit tests: a fix: with no test fails; a No-Test: reason passes; an empty No-Test: fails; a fix with its test in a separate commit on the same branch passes.
• This PR is itself the first CI run of npm run check with full history, so the check job here is what proves C1 and C3 work in Actions.

Also corrected because they were the same drift: README's guard table with its last row stranded outside it, "six static guards" (there were seven; now nine), CONTRIBUTING's squash-merge where the repository rebase-merges, and "read it before editing App.jsx".

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
easy-mv-maker	Ready	Preview	Oct 14, 2026 8:41am UTC