backtest-diagnose

Fix failing backtests by reading artifacts like metrics.csv and equity.csv, inspecting signal_engine.py, classifying the issue via taxonomy, applying targeted code fixes, and re-running to verify improvements against hard-gate checks.

Diagnosing abnormal performance requires reading output artifacts such as trades.csv and equity.csv to identify anomalies like NaN values or zero trade counts, classifying the root cause using an error taxonomy, and applying targeted code fixes to the signal engine or configuration.

Verify backtest fixes by enforcing a hard-gate checklist that confirms metrics.csv and equity.csv are non-empty, trade_count is greater than zero, equity contains no NaN, and exit_code equals zero, followed by re-running the backtest to report new metrics for comparison.

Yes, backtest diagnosis works by reading standard pipeline artifacts including artifacts/metrics.csv, artifacts/equity.csv, and trades.csv, alongside code files like code/signal_engine.py and config.json, to classify issues and apply fixes using edit_file.

Fixing underperforming backtests requires complete artifact files and accessible code; if metrics.csv or equity.csv are empty, or if trade_count remains zero after applying fixes, the hard-gate verification will block approval and require further manual diagnosis.