I build and audit trading systems in Python.

Most of my work is the unglamorous half: checking whether a backtest can be believed before anyone risks money on it. Realistic costs per side, clustered t-statistics instead of trade-level ones, permutation controls against random entry, and a power analysis so a negative result can be told apart from an untestable one.

crypto-strategy-audit — the toolkit, plus what came out of running it on 52 hypotheses: the signals rank the cross-section with |t| from 5 to 9, and none of it converts into money, because the return distribution is severely skewed and the ranking does not predict the skew.

Also there: an execution skeleton with 83 reliability checks that need no network — exchange filters, rate limits, dropped connections, duplicate orders, and the unclosed-candle leak. Two of those checks found real bugs in my own code.

Python · pandas · numpy · exchange APIs · statistics that survives scrutiny

Open to work on backtesting, strategy implementation, and fixing bots that don't behave the way their backtest promised.