PR-T2 — sync_versions round-trip test

What it locks

Lane 2 #1407 added a function to scripts/stamp-counts.sh that reads package.json and propagates the version to 6 sibling files. No test existed for it. If someone later refactors and breaks the propagation, the next release would silently drift across files (exactly the bug Lane 2 was designed to prevent).

Round-trip flow

1 — Pre-flight

Capture all 7 file versions
Refuse to run if any mismatch
Arm EXIT trap to restore

→

2 — Bump + sync

pkg.json → 999.888.777
Run stamp-counts.sh
Assert all 6 siblings match

→

3 — Restore

pkg.json → original
Run stamp-counts.sh again
Assert all restored
validate-counts agrees

Local run output

▶ Test 1: sync propagates a bump
  ✓ pyproject synced to 999.888.777
  ✓ manifest synced to 999.888.777
  ✓ market-top synced to 999.888.777
  ✓ market-plugin synced to 999.888.777
  ✓ release-please synced to 999.888.777
  ✓ CLAUDE synced to 999.888.777

▶ Test 2: sync propagates the restoration back
  ✓ pyproject restored to 7.56.0
  ✓ manifest restored to 7.56.0
  ...

▶ Test 3: validate-counts.sh accepts the final state
  ✓ validate-counts passes after round-trip

════════════════════════════════════════════════════════════════
  Total: 14  |  Passed: 14  |  Failed: 0
════════════════════════════════════════════════════════════════

Safety

Trap on EXIT + INT + TERM restores the original version regardless of how the test exits (including Ctrl-C). No wedged states.
Pre-flight guard aborts if any file is already mismatched — the test refuses to run on a dirty baseline.
Wired into tests/run-all-tests.sh under UNIT TESTS, so npm test runs it.