First refactor plan (maintainer)

A pragmatic first refactor should reduce risk, not expand scope.

Proposed first steps

1. Lock trace schema discipline
2. ensure every run writes a valid trace (normal run)

3. validate/upgrade trace payloads consistently

4. Align CLI and API

5. same minimal artifact semantics for equivalent inputs/config

6. consistent error mapping and termination reasons

7. Simplify configuration

8. make “what config keys matter” obvious

9. keep defaults in one place

10. Strengthen tests

11. add small fixtures that validate trace headers and failure taxonomy

If you cannot explain the refactor in one page, it’s probably too big.