ROC-12 Mean Reversion Killed by Insufficient Validation Data

We killed this mean reversion strategy at audit because its own validation data said it hadn't proven itself. The gene traded ROC-12 (rate of change over 12 periods) drops below -2 in low-volatility regimes (ADX under 30), aiming to catch snapback bounces on 5-minute bars across BTC, ETH, XRP, and BNB. Despite a 76.8% win rate in both training and validation, average P&L per trade sat at -0.003% in both windows. The deflated Sharpe ratio landed at -0.02, and after multi-test correction the probability this edge was real came back at 100%, meaning we had zero statistical confidence.

The core problem was sample size. The validation window produced only 151 trades. For a 5-minute intraday strategy running on four major coins with a 10-hour daily session (12:00 to 22:00 UTC), that count is anemic. The high win rate looks appealing until you realize the wins were smaller than the losses on average, and we had no statistical power to distinguish skill from noise. The ADX filter, designed to keep the strategy out of strong trends, also kept it out of most setups entirely. Low ADX means low volatility and fewer ROC threshold breaches, so the regime constraint strangled trade flow.

This is a soft kill, not a permanent one. We flagged it as reversible on May 2, 2026 because the strict-gate criteria that killed it are being re-evaluated lab-wide. If we adjust minimum trade count thresholds or loosen the deflated Sharpe floor, genes like this one get a second audit. That said, a strategy that can't generate 200 validation trades in a multi-coin, multi-month backtest has a structural problem, not just a statistical one.

The lesson here is that high win rate means nothing without volume and positive expectancy. A 76.8% win rate on 151 trades with negative average P&L is a recipe for slow bleed, not edge.

If you are testing mean reversion ideas, visit /prove and make sure your regime filters still leave enough setups to matter. For the full registry of live strategies that cleared these gates, see /survivors.