The 95% Win Rate That Wasn’t

Every trader has seen the screenshot. A stretch of trades so clean it looks unbeatable: 17 wins, 1 loss, a win rate north of 90%. It gets posted, it gets admired, and somewhere in the replies someone asks the only question that matters: over how many trades?

That question is the whole post.

An illustration, not a result

Here’s a made-up example to walk through the mechanism. Synthetic numbers, not a real account, not a real strategy. Call it Strategy X.

Strategy X has been logged for 41 trades. Across the full 41, it wins 23 and loses 18, a 56% win rate. Fine. Unremarkable. The kind of number that doesn’t get screenshotted.

Now slice it. Look only at the last 22 trades: 15 wins, 7 losses, 68%. Better, but still slice-able. Narrow again to the last 13: 11 wins, 2 losses, 85%. One more cut, down to the last 9: 8 wins, 1 loss, 89%.

Nothing about the strategy changed between those four numbers. The same trades, the same rules, the same trader. All that changed was where the window started. And the smaller the window got, the better the number looked, right up until you’re one trade away from “9-for-9, undefeated.”

That’s the trick, and it isn’t even a trick anyone has to do on purpose.

Why the slice misleads

Three things are happening at once, and they compound.

Small samples are noisy by nature. A coin has a 50% chance of heads, but flip it nine times and a run of 7-heads-2-tails is not a shocking outcome. It’s a normal one. The fewer trades you look at, the more the result is dominated by short-run luck rather than the strategy’s actual edge. Win rate on 9 trades tells you almost nothing about win rate on 900.

Your own history gives you infinite chances to find a good stretch. With 41 trades logged, there are dozens of possible 9-trade windows sitting inside that history: sequential ones, overlapping ones. If you go looking for the best-looking window after the fact, you will find one, because with enough windows to choose from, one of them is going to look great by chance alone. This isn’t dishonesty. It’s just what happens when a real number gets reported without the sample size sitting next to it.

Fewer trades means fewer chances for the strategy’s real weaknesses to show up. A strategy that struggles in choppy conditions, or on a particular day of the week, or after two losses in a row, needs enough repetitions for those weaknesses to actually appear in the data. Cut the window down to 9 trades and you may simply not have sampled the conditions where it breaks.

Put together: a small, cherry-pickable window is not a smaller version of the truth. It’s a different kind of number entirely: one that answers “what happened in this stretch” instead of “what does this strategy actually do.”

None of this requires bad faith. A trader genuinely proud of a hot streak, screenshotting the last 9 trades because that’s what’s on screen, is doing nothing wrong by their own lights. The problem is structural, not personal: a thin number and a real number can look identical, and only the sample size tells them apart.

The design principle

This is why we think the right question for any trading tool isn’t just “what’s the number,” but “does the tool tell you when the number is too thin to trust.”

A win rate on 9 trades and a win rate on 900 trades are not the same kind of fact, even when they’re both technically true. A system worth trusting shouldn’t present them the same way. It should distinguish a stretch from a track record (surface the sample size next to the number, not bury it) and let a trader see the difference between “this looks great” and “this looks great, and here’s how much weight that should actually carry.”

We’re not describing a specific mechanism here. That’s deliberate. What we’re describing is a standard: the number and its confidence should travel together, always. A flattering number with no sample size attached is a headline. A flattering number with its sample size attached is data you can act on.

The house rule

We started this post by looking at someone else’s screenshot, so it’s only fair to hold ourselves to the same test. Every specific claim in this post that isn’t clearly marked “illustration” is one we could show our work for. Strategy X doesn’t exist. We said so, twice. That’s not an accident of style. It’s the rule this piece is written under: if we can’t prove it, we don’t say it.

That rule is going to show up again in things we publish about our own product, our own process, and our own results. Consider this the first time you’ve seen it named. It won’t be the last.