MFI+RSI double oversold: killed by negative validation Sharpe

We killed this strategy at audit after it failed strict validation gates despite showing a brief positive forward-test run. The strategy combined MFI(14) below 20 and RSI(7) below 25 to catch double-oversold bounces on BTC, ETH, XRP, and BNB on 5-minute bars, with an ADX filter to avoid strong trends. After 13 forward trades averaging +0.20% per trade, we ran the audit and found the underlying walk-forward validation was deeply negative: -0.172% per trade with a deflated Sharpe of -1.48 and a 100% probability of being a fluke after multi-test correction.

The problem was structural. The strategy generated 704 out-of-sample trades during validation, enough volume to be statistically meaningful, but the edge was upside down. The 57.8% win rate looked plausible at first glance, but the losses were larger than the wins, dragging average P&L negative. The regime filter, which tried to keep us out of strong trends by requiring ADX below 30, may have been counterproductive. Mean reversion in choppy, low-ADX conditions on 5-minute crypto often means getting whipsawed by noise rather than catching genuine bounces.

The forward test's brief positive run (13 trades, +0.20% average) was not enough to override the validation failure. This is exactly the scenario our audit gate exists to catch: a strategy that looks live-viable in a short sample but whose deeper backtest history shows no real edge. The deflated Sharpe of -1.48 told us that even after correcting for multiple testing, this was more likely a statistical artifact than a real phenomenon.

We soft-killed the strategy in May 2026 and flagged it for re-evaluation under revised gates. The takeaway: dual oscillator oversold signals are intuitive, but on fast timeframes in crypto, they need either tighter exits or stronger regime context to avoid bleeding on false signals. If you are working on mean reversion ideas, check out the full survivor registry at stratproof.com/survivors to see which oscillator combinations actually held up under audit, or head to stratproof.com/prove to test your own thresholds with proper walk-forward validation baked in.