Most trading ideas that land on my desk come wrapped in a great story. Someone found an indicator, ran it on a chart, and the screenshots are gorgeous — winner after winner. That's exactly when I get suspicious. A lot of those charts are quietly hiding their losers. The MACD engine we looked at last week literally erases every trade that doesn't hit full target, so all you see is a wall of green. It's not lying about any single trade. It's just not showing you the other half.
So the first thing we do with any new idea is ignore the sales pitch and ask one plain question: does this hold up on our own data, after fees? Everything downstream is just answering that carefully.
Building it is the easy part. Our strategies are just JSON files that snap together small, reusable pieces — an entry trigger plus a few filters. To test a new idea I port its entry logic into one of those pieces, exactly as written, warts and all, and drop it in alongside the others but switched off. Nothing in the live system changes. That isolation is the whole point: I want to see this one entry naked, not tangled up with the strategies that already work.
Then we replay months of one-minute data through it, using the same stops and trailing logic we'd actually trade — and, importantly, we subtract commission. Gross profit means nothing to me. Plenty of “edges” look great until you take out four dollars a round-turn, and then they're just a slow way to feed the broker. We also split it by session, because a setup that prints money at the New York open can bleed all night, and one blended number would hide that completely.
The part everyone gets wrong is the tuning, because that's where you start fitting noise without noticing. We don't sit there grinding through parameter combinations hunting for the prettiest curve — that's just memorizing the past. Instead we pull up the trades that actually lost, and look for a real reason. Bad hour. Bad weekday. Wrong direction. Then we block that specific thing. If I can't explain why a filter exists in one sentence, it doesn't go in.
And a good profit factor still isn't enough to trade. A strategy can look profitable and still blow up a prop account, so we run it through a simulator that slides a thirty-day window across the whole history and asks how often the drawdown breaches the two-thousand-dollar limit. If it busts in a chunk of those windows, it's dead, no matter how nice the equity curve looks. Marginal edge is never a license to size up.
Every change we keep gets a note — what we tested, what came back, why we kept it. Boring, but it stops you from slowly overfitting your way into a fantasy.
Honestly, none of this is clever. It's mostly just refusing to believe the good news until we've tried hard to break it ourselves — and most ideas don't survive that.