Coverage loop

1. Capture a coverage baseline with the repository-supported command.
2. Select uncovered behavior by risk, not by easiest lines.
3. Add behavior-focused tests that survive refactors.
4. Re-run targeted tests and the same coverage command.
5. Keep only tests that improve meaningful coverage or lock a real invariant.
6. Repeat until the selected risk area is covered or remaining gaps require a separate design change.
7. Run review.

Do not add assertions that merely execute code without checking behavior.