The Backtest API That Corrects Itself

Why this is the honest number, not just a different number

Run a parameter sweep across N strategy variants on the same historical data and the best-performing one will look great even if every single variant is worthless — pure selection bias. The Deflated Sharpe Ratio (Bailey & López de Prado methodology) corrects for exactly this: it uses the observed spread of Sharpe ratios across every configuration actually tried, plus the return series' own skew and kurtosis, to compute the real probability the true Sharpe exceeds zero. Below ~0.95, backtest-sweep's own response tells you not to treat the result as a discovery.

• Every sweep response ships both numbers — the naive/uncorrected Sharpe an ordinary backtester would show you, and the corrected one that should actually drive a decision.
• No arbitrary code execution. Strategies are a fixed, named set with numeric parameters — ma_cross, ema_cross, breakout, rsi_reversion, buy_hold. A backtest service that evals uploaded strategy code is a remote-code-execution endpoint with extra steps; this one isn't.
• Causal by construction. A signal fills at the next bar's open, never the signal bar's own close, and the currently-forming candle is excluded from every series — no lookahead bias baked into the engine.
• Reproducible receipts. Every result carries a SHA-256 receipt hashing the data, config, and result together — re-run with the same inputs and the digest must match, or something changed and the earlier claim no longer stands.

Endpoints (Base USDC via x402)

Every stat sheet: Sharpe, Sortino, CAGR, max drawdown (+ duration), win rate, profit factor, Calmar, volatility, num trades, Deflated Sharpe, and a bootstrap confidence interval on the Sharpe ratio.