Summary
Once harness evolution results are real (#19), we need CI to ensure they stay verified and don't drift from committed state. This prevents accidental staleness of the README's key claims.
Related
Task
Extend .github/workflows/ci.yml harness-verification job to:
- Run
scripts/harness_evolution_6_epochs.py in CI
- Compare output JSON against committed versions
- Fail the job if results drift beyond a tolerance (e.g., >1% Sharpe change)
- Alternatively: add a
make verify-results / script for maintainers to run before updating README claims
Files to Modify
.github/workflows/ci.yml — Add verification job
- Optional:
Makefile or scripts/verify-results.sh — Maintenance script
Acceptance Criteria
Notes
This enforces the 'no lying in README' property going forward by catching accidental staleness.
Summary
Once harness evolution results are real (#19), we need CI to ensure they stay verified and don't drift from committed state. This prevents accidental staleness of the README's key claims.
Related
.github/workflows/ci.ymlTask
Extend
.github/workflows/ci.ymlharness-verificationjob to:scripts/harness_evolution_6_epochs.pyin CImake verify-results/ script for maintainers to run before updating README claimsFiles to Modify
.github/workflows/ci.yml— Add verification jobMakefileorscripts/verify-results.sh— Maintenance scriptAcceptance Criteria
Notes
This enforces the 'no lying in README' property going forward by catching accidental staleness.