Session filters killed every entry we tested

We tested whether restricting entries to specific trading sessions (Asia, EU, US, overlap periods, or US evening) would sharpen edge on known-decent patterns. We ran 283 candidates built from 4 base entries (RSI 14 oversold, RSI 7 oversold, StochRSI oversold, and a BB plus RSI compound signal) crossed with 5 forward-only session windows on 1-hour and 4-hour bars, using profit targets between 1.0 and 2.0 percent. Zero candidates graduated. Pass rate: 0.00 percent.

The hypothesis was straightforward: crypto volume and participant behavior shift by time of day, so an RSI oversold signal during Asian hours might behave differently than the same signal during US evening. If session timing mattered, we would expect at least a handful of the 283 variants to clear our multi-gate gauntlet of walk-forward validation, deflated Sharpe thresholds, and drawdown limits.

Instead, every candidate either failed to generate enough trades in its assigned window, failed walk-forward consistency checks, or produced a deflated Sharpe ratio that flagged it as likely overfit. The result does not prove that session filters are universally worthless. It proves that our specific encoding of session windows, combined with these four entry primitives and tight profit targets, did not survive these gates on this coin set and lookback period.

Two likely explanations. First, crypto trades around the clock with less pronounced session structure than FX or equities. Volume peaks exist, but they may not carve clean regime boundaries that a mechanical time-of-day mask can exploit. Second, our current system applies session filters as hard on-off gates. We cannot yet express more nuanced logic like "enter only if RSI is oversold AND volume in the past two hours exceeds the 24-hour median," which might be the real edge hiding underneath session timing. The primitive system that lets us test hundreds of ideas quickly also constrains the complexity of any one idea.

We are not pursuing session filters further until we add stateful primitives that let strategies condition on recent volume or volatility profiles. If you want to test your own hypothesis about time-based edge, head to stratproof.com/prove and encode it. For the full registry of strategies that did graduate, see stratproof.com/survivors.