Why we validate every strategy out-of-sample
A backtest that looks perfect is usually lying to you. Here's how walk-forward validation and our acceptance gate keep overfit strategies off your account.
Give anyone enough parameters and they'll produce a backtest with a beautiful equity curve. The question is never 'did it work on the past?' — it's 'will it work on data it has never seen?'
Walk-forward, not hindsight
We split history into in-sample (training) and out-of-sample (OOS) halves. We optimise on the first half and measure on the second. If performance collapses out-of-sample, the strategy was curve-fit to noise.
The acceptance gate
Before a strategy is fit to publish it must clear explicit criteria: IS→OOS Sharpe degradation under 40%, at least 30 OOS trades for significance, a positive OOS Sharpe, and an OOS win rate of at least 35%. Miss any one and the gate tells you exactly why.
Honest by default
If a strategy can't survive out-of-sample, we'd rather tell you than ship it. That's the whole point of the gate — it protects you from your own best-looking backtest.