How many trades should you backtest?
Enough for the edge to clear its own noise. At 45% wins, +1.8R / −1R and 0.05R of costs (+0.21R a trade), that is 170 trades at 95% confidence.
Why there's no single number
The count depends on two things: how big the average trade is (+0.21R here) and how much single trades swing around it (1.39R here). The interval around the average narrows with the square root of the trades, so the count grows with the square of swing ÷ edge. Raise the costs to 0.155R and the edge halves to +0.11R: the count goes from 170 to 677. That's why rules of thumb (30, 100, 500 trades) disagree: each fits some strategy. Our review plan turns the same arithmetic into check-points and kill rules you write down before you start.
Is 100 trades enough?
Often not. On the example above, 100 trades put expectancy anywhere from −0.06R to +0.48R at 95%: zero is still inside.
What 100 trades does tell you
It is plenty to find operational faults: a rule you can't apply the same way twice, a stop that slips, costs far above what you assumed. It tells a large edge from zero (a 55% system at 2R / 1R clears zero well before 100 trades); it rarely tells a small one. Paste your own R results into the calculator: it says whether your interval already clears zero, and if not, about how many more trades the same edge and spread would need.
How many trades to know your win rate?
To pin a 45% win rate within ±5 points at 95% confidence: 377 trades. A win rate on its own also says nothing about the edge.
The interval this uses
The Wilson score interval, the one our review plan uses: after 30 trades an observed 50% fits anything from 33.2% to 66.8%; after 100, 40.4% to 59.6%. A win rate only means something next to the average win and loss, which is why the calculator counts trades for expectancy first. The expectancy calculator gives the break-even win rate for your payoffs.
Count trades that are really separate
The arithmetic assumes each trade is independent. Ten trades taken in the same move are closer to one.
Spread the sample across conditions
Trades taken in the same session, on correlated pairs or in one trend move tend to win or lose together, so they count for less than their number. Spread the backtest across months and market conditions (a trend, a range, a news-heavy week): 300 trades from one quiet month say less than 300 from three different years. And a backtest you tuned on the same data it's scored on will look better than it trades; keep the last stretch of history unseen until the rules are fixed. The Monte Carlo simulator shows how far equity paths drift apart on the same edge.
Where manual backtest trades come from
Bar replay: the chart hides the future and you trade it one bar at a time. TradingView and FX Replay both do it, with different limits.
FX Replay's limits per plan, read
| FX Replay plan | Sessions | Trades per session | History kept |
|---|---|---|---|
| Beginner (free) | 2 | 50 | 1 week |
| Intermediate | 10 | 200 | 6 months |
| Pro | Unlimited | Unlimited | forever |
From FX Replay's Plans & Pricing Compared (updated 3 October 2026), which also says the session limit “counts every session on your account. It doesn't reset when your subscription renews”. So the free plan holds 100 trades at a time, Intermediate 2,000, and Pro has no cap. TradingView's Bar Replay works on every plan for daily bars and needs a paid plan for minute bars: how to backtest on TradingView has its limits and steps. The FX Replay plan picker works out which FX Replay plan the rest of your setup needs.