I spent months building trading strategies and then published the results that killed most of them

2026-07-29 · Yusuf Gadelrab

Here's what walk-forward testing did to my ideas.

The gap and opening range breakout proxy: negative expectancy, minus 0.28R. Dropped it.

VCP, the volatility contraction pattern, which is the setup I most wanted to work because the literature is beautiful: roughly breakeven. Not tradeable.

Anchored VWAP reclaim: one survivor. I first reported plus 0.23R over 101 trades. A later adversarial re-test killed that number — on a 129-symbol, 10-year universe it is plus 0.117R over 4,933 trades, 95% CI plus 0.057 to plus 0.174.

Then I tested two filters I was sure would improve things. A market regime filter and a new market leader filter, both standard advice, both intuitively obvious. Both made expectancy worse on my own data. I removed them.

The reason I'm writing this down is that almost nobody publishes the negative results, and the absence of negative results is exactly why retail traders keep rediscovering the same dead ideas. If every writeup you read is a winner, you will conclude that finding winners is easy, and you will interpret your own losses as a personal failing rather than as the base rate.

Plus 0.117R is a modest edge, and a risk-matched random entry captures plus 0.086R of it, so the part actually attributable to the signal is plus 0.030R with a confidence interval that crosses zero. That is why the engine has a hard gate: no real broker connection until 60 closed paper trades with positive realized expectancy. The gate is in the code, not in my willpower, because willpower is not a risk control.

I have not turned it on. The paper account isn't even funded with keys yet.

If you build systems, put your promotion gates in the code. The version of you that wants to skip them is not a hypothetical.