The usual answer is a hundred. The usual answer is wrong in a way that matters most to the people who trade most: fifty trades on five days is five pieces of evidence, not fifty.
Search for how many trades you need and every result says roughly the same thing: a hundred is usable, two hundred is convincing, five hundred is significant. The arithmetic behind it is right as far as it goes. Eight wins out of ten is a 95% interval of about 44% to 97%. A 70% win rate over twenty trades is anything between 48% and 85%. Even a hundred trades only narrows a 70% to something like 60% to 78%.
Every one of those numbers assumes your trades are independent draws. Trades inside one session are not. They share the day's direction, the day's volatility, and usually the same underlying: if the tape ran against you at 10:15, it ran against the condor you put on at 10:20 as well. Fifty trades spread over five days is closer to five pieces of evidence than to fifty, and the standard formula, which counts fifty, hands back an interval that is too narrow. It is narrowest, and therefore most misleading, for exactly the trader who trades most: the 0DTE seller with forty fills in an afternoon.
50 trades, 30 of them winners, on 10 sessions. The per-session win rates were 60%, 60%, 0%, 80%, 40%, 20%, 100%, 100%, 40%, 100%, which is what real trading looks like: whole days that go right and whole days that do not. Both intervals below are computed on this page by the same resampling code that computes what a subscriber sees.
| Counted as | Win rate | 95% interval | Width |
|---|---|---|---|
| 50 independent trades | 60% | 46% to 74% | 28 points |
| 10 trading sessions | 60% | 40% to 80% | 40 points |
12 points of width appear the moment you stop pretending a session's trades are separate evidence. And the honest reading of the wider one is the thing worth saying out loud: a 60% win rate over 50 trades on 10 days does not distinguish you from a coin flip. Not because you are not good, but because 50 trades on 10 days cannot.
Every estimate on every screen is bootstrapped by drawing whole trading days with replacement, which is the correction above, applied everywhere rather than in a footnote.
Under five days there is no interval, and the page says which number is missing instead of printing a range that would be theatre.
A strategy whose interval crosses zero has not been shown to work, however many of its trades won. That sentence is on the screen next to the bar.
One detail decides whether any of this is right: which day a trade belongs to. A broker writes local time with an offset, and a real export in this repository is from a New Zealand account where a fill stamped 2025-09-12 at 01:53 belongs to the session of 2025-09-11, exactly as its own OCC symbol says. Slicing the date off the timestamp, the obvious thing to do, moves that trade to the wrong day, and since the intervals are clustered by day, a day boundary in the wrong place quietly narrows every interval the product prints.
Intervals are on the paid plans, from $24 a month, because the code that computes them is server side and is not shipped to the browser. The free tool rebuilds the full lifecycle of every contract in your own browser and uploads nothing. The other page is about what your fill actually paid.