Backtests are only as good as their assumptions. We explicitly model (or document) the pieces that typically create a gap between "backtest" and "live".

• • Spread and bid/ask handling (not mid-price fantasy)
• • Slippage and latency assumptions where relevant
• • Commission/fees and contract specifics
• • Session/time filters and "no trade" windows
• • Sensitivity checks and parameter robustness

Note: No backtest can guarantee live performance. Our goal is to make assumptions explicit and falsifiable.

Traders should be able to answer: "Why did the strategy enter here?" We instrument strategies so every trade is explainable.

• • Rule tags (E1/E2/… for entries, X1/X2/… for exits)
• • Logged indicator values and filter states
• • "Blocked trade" logs (why a trade did not happen)
• • Chart overlays for validation

ML can help with regime detection, signal gating, and adaptive parameters — but it can also overfit. We treat ML as an optional module with strict validation.

• • Time-series aware splits (no leakage)
• • Walk-forward evaluation and out-of-sample testing
• • Baseline models first (interpretable where possible)
• • Monitoring plan: drift, confidence, and fallback behavior

Strategy code and data are sensitive. We can deploy backtesting infrastructure behind Cloudflare Zero Trust (Access + Tunnel) and apply principle-of-least-privilege access.

• • Private repos and versioned releases
• • Client-owned infrastructure option for high sensitivity
• • Documented handover and IP ownership terms

Want to understand the workflow? Read How it works.