SMA cascade killed at audit: insufficient validation trades

We killed this four-moving-average cascade strategy at audit after discovering it graduated on just 11 validation trades, far below the threshold needed to trust the numbers. Despite a perfect 5-trade forward run averaging 1.713% profit per trade, the backtest validation sample was too small to justify continued live exposure.

The strategy shorted when all four SMAs (21, 50, 100, 200) stacked bearish and price closed below the shortest average, filtered to low-ADX regimes (below 30) to avoid strong trends. It exited on a net profit target of 1.5%. Training showed 0.633% average profit per trade at 81.8% win rate, and validation matched exactly: same 0.633% and 81.8% win rate. That perfect match between training and validation should have been a red flag, not a green light.

The deflated Sharpe ratio came in at 3.36, which sounds excellent until you account for the multiple testing correction: probability of fluke hit 100.0%. With only 11 out-of-sample trades across BTC, ETH, SOL, and XRP on 60-minute bars, the backtest simply didn't generate enough evidence. The regime filter, designed to keep the strategy out of strong trends, likely choked off trade opportunities so severely that validation became a coin flip on a handful of setups.

Forward testing added irony to the failure. All five live trades won, averaging 1.713% each. But five trades in forward plus eleven in validation equals sixteen total pieces of out-of-sample evidence. No amount of perfection on sixteen trades justifies calling a strategy robust.

This is a soft kill, marked reversible pending a gate fix. The strategy might be sound, but the lab's original graduation logic failed to enforce minimum trade count requirements. The audit caught it. The lesson: a high win rate and tight Sharpe ratio mean nothing without enough at-bats to prove the edge survives variation. Insufficient sample size is the same as no test at all.

Check the full registry at /survivors to see which strategies passed audit with adequate trade counts, or visit /prove to design and backtest your own rules with proper validation gates baked in from the start.