Planned — no result published
Claude Code + Opus 5: with vs. without Ponytail
Does a minimalism add-on reduce implementation size without reducing task success?
Held constant
- Same repository snapshot
- Same task prompt
- claude-opus-5
- Same time and token budget
- Same tests and human rubric
Only changed variable
Ponytail instructions enabled for one arm only
Measures
- Acceptance-test success
- Regressions
- Human intervention
- Elapsed time
- Usage cost
- Lines and dependencies added
No winner or metric appears until both arms have reproducible run artifacts. This page is the public protocol, not a simulated result.