backtest-report-gen

Problem: Execs remember charts, not your notebook. You already have eval JSON — you need a shareable, static report (email, PDF, artifact).

Solution: One command: metrics.json → single report.html (Chart.js over CDN): calibration vs observed, rolling accuracy, CLV KPIs, optional bet-ledger ROI.

JSON shape matches nba-clv-dashboard demo_metrics (model_name, overall_accuracy, brier, calibration[], optional rolling_accuracy, clv_summary, optional bets).

pip install -e .
backtest-report path/to/metrics.json -o report.html
# open report.html → Print → Save as PDF

Optional bets array

Each row: stake, won, odds_american (profit computed with standard American payout math), or precomputed ev_units times stake.

Non-goals

• No cloud upload — file output only.
• No bundled data — bring your own eval JSON.

License

MIT