How we test a trading idea

Almost any idea looks good on some chart. The hard part is telling an idea that works from one that was lucky in the months you happened to test. These are the rules we use so a result means something.

• Markets: NQ, ES, gold (GC), oil (CL) and Bitcoin (BTC) futures, on 1, 5 and 15-minute bars, from September 2025.
• Every result is after costs: a tick of slippage on entry and on stop or time exits, plus the exchange and broker fee.
• Results are often in R, where 1 R is the trade's stop. That lets us compare markets of very different sizes.

1. No look-ahead. A reading is used only from the minute it was known. A bar's close is known at its close, not at its open. Daily values from yesterday are used today, not today's.

2. Two halves. We split the period into a first and a second half and judge each. An idea that made money in one half and lost it in the other is not an idea, it is a period.

3. A luck check that keeps the day together. We shuffle the idea's results between whole sessions, 200 times, and see how often chance does as well. Trades on the same day are linked: a good day gives several good trades. Shuffling single trades hides that and makes streaky ideas look better than they are, so we shuffle sessions.

4. The reason comes first. Before a test we write down why the idea should work, on which markets, and which way. A result we find after looking is called a lead and must hold on new months before it counts.

5. We publish failures. Most ideas we test do not beat chance after costs. Those notes matter as much as the rest.

With five markets, three bar sizes and dozens of filters, hundreds of tests are run. About one in twenty will pass any single check by chance alone. Two halves, the luck check and a reason written in advance together cut most of those false wins.

When you test your own idea, ask three questions: does it hold in both halves, does it beat shuffled histories, and did you have a reason before you looked? Analyze shows all three on every idea.

Our history starts in September 2025, so every note covers about 14 months. That is enough to reject many ideas and too short to prove any. We re-run the notes as new months arrive.

Simulated results on past market data from 2025 and 2026. Research only, not trading signals or advice. Past results do not promise future ones; see the hypothetical performance disclosure at the bottom of the page.