Timmy
bf003cd944
feat: measurer.py — compounding intelligence metrics engine
Implements issue #14: 7 metrics that prove knowledge compounding.
Metrics:
- Knowledge velocity: new facts/day (from index.json)
- Knowledge coverage: % domains with 10+ facts (from YAML files)
- Hit rate: % sessions referencing bootstrap knowledge
- Error recurrence: same errors across sessions (should decrease)
- Task completion: % sessions with successful end_reason
- First-try success: actions without backtracking (tool/msg ratio)
- Knowledge age: staleness of facts (freshness score)
Data sources:
- knowledge/index.json + YAML files for fact metrics
- ~/.hermes/state.db sessions + messages tables
Features:
- JSON and markdown output formats
- --since, --repo, --format flags
- 7-day trend tracking via snapshot persistence
- Runs in 33ms on 11.9K sessions / 192K messages
- Dashboard auto-generation with --save-snapshot
Closes #14