fix(migrations): show retry failure over stale precheck error in PrecheckPanel

[detail-app]

Detail bug report: View on Detail

Summary

Related Issue: polarsource/feedback#485

On the merchant-migration source_setup step, PrecheckPanel showed the stale prior-run error instead of a just-failed retry's own failure message. The fix reorders the error derivation so the newest failure wins.

What

• PrecheckPanel.tsx: the error ternary now checks precheck.isError first, then the cached operation.error (with a new generic fallback for a failed operation lacking an error string), then null.
• PrecheckPanel.test.tsx (new): the first component-test harness for this panel, covering the ordering invariant and the sticky-error-clearing behavior.

Why

A failed retry POST does not overwrite the cached migration.operation — the useRunMerchantMigrationPrecheck mutation only writes query data in onSuccess. So when a prior worker precheck left operation.status === 'failed' with a non-empty operation.error, the old derivation (failed ? operation.error : null) || (precheck.isError ? retryMsg : null) short-circuited on the cached value, and the precheck.isError branch was dead. After a failed retry the merchant saw the same stale "We couldn't verify the Stripe key right now…" text instead of "We couldn't start the pre-check. Please try again." — an attribution gap (the corrective action is identical either way). Introduced when PR #14253 rewrote the panel from sync to async polling.

How

One-line operand reorder so the newest failure takes precedence, mirroring the sibling ReviewTable's ordering. Added a generic fallback ("We couldn't run the pre-check. Please try again.") for the defensive case where a failed operation has no error string. No new imports, types, or behavioral contracts.

Testing

• New PrecheckPanel component suite (6 tests): mocks api.POST to drive the real useRunMerchantMigrationPrecheck mutation through a real QueryClient (shared with the mocked getQueryClient so onSuccess cache writes are observable) and stubs Orbit components. Covers: stale error shown when idle, retry failure wins over stale (the bug), retry failure with no prior run, failed-without-error fallback, in-flight spinner/disabled state, and no-sticky-error after a successful retry.
• Red/green confirmed: with the buggy derivation swapped back, 4 tests fail (stale text shadows the retry error); with the fix, all 6 pass.
• Routine checks pass: web unit suite (408 tests), tsc --noEmit, type-aware oxlint, and oxfmt --check on the changed files.
• End-to-end against the live local stack was not run: the dev docker stack couldn't initialize — the host filesystem (a 20GB loop device) was 100% full, so the per-instance Postgres database and MinIO buckets couldn't be created (No space left on device during CREATE DATABASE / MinIO setup). The behavior it would verify (the rendered error-text swap) is otherwise covered by the real-mutation component suite above.

Checklist

• This PR addresses a single concern (one bug fix)
• The diff is reasonably sized and easy to review
• New functionality is covered by tests
• Linting and type checking pass
• No unrelated changes or drive-by fixes are included

---

Automatic Fixes PRs can be configured here.

[vercel]

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

Project	Deployment	Actions	Updated
orbit	Ready	Preview	Sep 9, 2026 4:13pm UTC
polar-test	Ready	Preview	Sep 9, 2026 4:13pm UTC