One mechanical verdict for your backtests: PASS or KILL.
quant-harness screens a strategy's returns against four pre-registered gates — combinatorial purged cross-validation, the Deflated Sharpe Ratio, the probability of backtest overfitting and drawdown — and tells you whether the edge survives the number of things you tried.
Built for quants, systematic traders and the AI agents that generate strategies for them. The judge is calibrated on synthetic data with known ground truth: 0% of pure noise passes at six or more configs.
Web
Upload a returns CSV, get a report with CPCV path Sharpes, cost sensitivity and the four gates. Share it by link.
API
POST /api/v1/verdict with an API key from any pipeline; JSON verdict plus a report URL.
MCP
Give your AI agent a judge_returns tool so it cannot ship a curve-fit strategy without a verdict.