Make 6-epoch harness evolution results real & reproducible

[OnePunchMonk]

Summary

The current 6-epoch results in results/harness_evolution_6epochs_results.json and benchmark numbers were generated as simulations, not from actual script execution. This needs to be real and reproducible.

Related

• Blocks: Add CI job to verify harness evolution results don't become stale #20, Verify test suite: confirm 63 tests passing #21, Implement Research Agent Phase 1: Literature Discovery Tool #22 (all other work depends on credibility)
• Depends on: Nothing (this is the critical path)
• Relevant docs: HARNESS_EVOLUTION_RESULTS.md

Task

1. Verify scripts/verify_tools.py loads the tool registry (with graceful fallback if APIs unavailable)
2. Execute scripts/harness_evolution_6_epochs.py end-to-end with the fallback paths
3. Regenerate all output files:
   • results/harness_evolution_6epochs_results.json
   • results/benchmark_report.json
   • .harness/v*.json evolved configs
4. Compare new numbers against committed versions - if they differ materially, update:
   • README.md results table (line ~24)
   • HARNESS_EVOLUTION_RESULTS.md metrics (all sections)
   • Mermaid diagram in README

Files to Check/Modify

• scripts/verify_tools.py — Run to verify tool registry
• scripts/harness_evolution_6_epochs.py — Execute this
• results/harness_evolution_6epochs_results.json — Output file
• results/benchmark_report.json — Output file
• .harness/*.json — Evolved configs

Acceptance Criteria

• verify_tools.py runs without errors
• Evolution scripts complete and produce JSON output
• Output files match committed versions OR all references updated to match new numbers
• No stale simulated numbers left alongside real pipeline
• Test the results by running the scripts locally

Notes

This is the critical blocker. The headline claim (+37.4% Sharpe) must be reproducible, not simulated.