How we know what we know.
Most trading products show you a backtest. We show you the gauntlet a strategy must survive before it touches capital — and the live verdicts, even when they say not proven.
The latest validation report is unavailable right now. When it loads, the verdict renders here — pass or fail.
Monte-Carlo resampling
We resample trade sequences thousands of times to ask: was this equity curve luck of ordering? If the strategy only works in one ordering of its own trades, it doesn't work.
Permutation tests
Signals are shuffled against the same prices. If shuffled signals perform about as well as real ones, the "edge" is the market's drift, not the strategy.
Probability of backtest overfitting (PBO)
Combinatorially split the data, pick the best configuration in-sample, measure it out-of-sample. PBO is the probability the in-sample winner underperforms out-of-sample — high PBO means we optimized noise.
Deflated Sharpe ratio
A Sharpe ratio earned after trying many configurations is inflated by selection. The deflated Sharpe corrects for how many things we tried; it is the number we trust.
In-sample / out-of-sample split
Parameters are chosen on one era and judged on another. The out-of-sample era is touched once.
Stress windows
The strategy replays through the worst regimes on record — crashes, chop, liquidity droughts. Surviving the average market is not the bar.
What this does not prove
No statistical test proves future returns. Validation reduces the probability that we are fooling ourselves; it cannot reduce it to zero, and it says nothing about regimes that haven't happened yet. That is why the engine trades behind hard risk rails — position caps, volatility-targeted leverage, daily loss halts and a kill switch — regardless of how good the evidence looks.
Read the full risk disclosure