[anth-volk]

Fixes #242

Summary

Follow-up to #237, from its review after merge. That review found no bugs; this PR fixes two small gaps in #237's record-and-continue change. Nothing that runs in production changes: one commit adds a test, the other changes documentation.

Since #237, staging reconciliation reports a run it can't publish or repair as failed, reconciles the remaining runs, and then exits with status 1.

1. Test for a reconciliation that writes. Keep same-named UK constituencies and local authorities apart in the calibration views #237's two tests of the failed outcome both run as dry runs. The new test in publish-calibration-tree-golden.test.ts runs a reconciliation that writes, using a store that records every operation. The newer of two UK staging runs has a manifest entry whose repair is refused. It checks:
   • the refused run is audited and then left alone: no upload and no manifest write;
   • the next run is uploaded, written to the manifest, listed again and audited, in that order, and reported published;
   • the manifest keeps the refused run's entry unchanged and gains exactly one entry.
2. Failure documentation. In docs/calibration-tree-artifacts.md, "Failure meanings and recovery" said a failed publisher workflow means publication didn't complete. It now separates the two kinds of run:
   • Release: reconciliation stops at its first error.
   • Staging: the workflow either stopped before reconciling any run, or printed a report in which every run without a failed outcome was reconciled. Every later staging workflow for that country retries the failed runs. Rerunning doesn't correct a refused repair of a published tree.

Commits

Commit	What	Red	Green
c5b7208	Test the written path of a failed staging reconciliation	No red step: the behaviour is already correct, so there's no bug for the test to fail on. As a check, I changed the per-run error handling to rethrow in a run that writes, without committing it: the test failed (0 passed, 1 failed, "Rebuilt staging index differs from immutable manifest entry aaaa…"). I then restored the file.	7 of 7 in publish-calibration-tree-golden.test.ts
0428dc0	Describe what a failed staging publisher workflow means since #237	Documentation only, so no red step	—

Tests run

Locally at the head (0428dc0), only the test file this branch changes: publish-calibration-tree-golden.test.ts, 7 passed and 0 failed. bun run lint (the dashboard and package typecheck) is clean. I didn't run the full suites; CI runs them.

Deploy order and #239

• Compare UK local-area Staging candidates with the local-area release #239 (uk-staging-comparison-baseline) is next in the deploy order. Every merge to main deploys production, so merge one PR at a time.
• No overlap with Compare UK local-area Staging candidates with the local-area release #239. This PR changes docs/calibration-tree-artifacts.md and frontend/packages/calibration-tree/src/publish-calibration-tree-golden.test.ts. Compare UK local-area Staging candidates with the local-area release #239 changes neither (gh pr diff 239 --name-only).
• This PR changes no runtime code, so it can merge before or after Compare UK local-area Staging candidates with the local-area release #239.

Environment variables

None. No workflow, deploy script, packaging or configuration changes, and the new test reads no environment variables. It sets nothing beyond what the neighbouring tests in the file already use.

🤖 Generated with Claude Code

[vercel]

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

Project	Deployment	Actions	Updated
calibration-diagnostics	Ready	Preview	Oct 20, 2026 4:52pm UTC