Research

Popular trading beliefs, run through the real machinery.

Each piece below takes something the retail trading industry sells — a strategy, a statistic, a story — and runs it through the same validation gauntlet we apply to our own ideas: permutation tests, out-of-sample splits, honest sample sizes. Most things don't survive. Including, frequently, ours: most ideas die in research before ever becoming a play, and the full plays that stop working are killed in public — 24 so far.

The 97% win rate that loses money

We ran a strategy that won 97.5% of 244 live trades. Then we asked the only question that matters — and killed it.

n=244 forward · WR 97.5% · permutation p=0.9956 · 0 of 11 tickers pass BH-corrected p<0.05 · verdict KILLED · 2026-07-27

PF 14.6, on 16 trades

The most exciting number our machine ever produced, and why we refused to trade it.

OOS PF 14.60 · WR 81.25% · n=16 · one cell of a multi-split sweep · verdict KILLED (parent pattern failed permutation) · never traded · 2026-07-27

Your backtest is lying to you

A strategy validated at profit factor 4.05, went live at 0.48, got killed — and then the autopsy found the kill diagnosis was partly wrong. It still doesn't trade.

backtest PF 4.05 → live PF 0.48 (677 entries, 1-min bars) · re-audit residual alpha +2.7%/10d (t=3.09) · permutation p=0.056 FAIL · verdict stays KILLED · 2026-07-27