Why one backtest is one path
This tool generates independent win/loss outcomes using the constant win probability, payoff and risk percentage you enter. Both the number of wins and their order can differ between paths. Equity multiplies by 1 + risk% × reward R on a win and by 1 − risk% on a loss. Rearranging an identical set of those fixed multipliers leaves final equity unchanged, apart from numerical rounding, but can change drawdown. This is a synthetic betting model, not a shuffle of your historical trades. For resampling imported MT5 postings instead, use the local report analyzer.
What the percentile bands mean, honestly
Load the worked example, then press Run: 50% win rate, 1.5R reward, 2% risk, 100 trades, seed 7 and 2,000 paths. Final equity at the 5th, 50th and 95th percentiles is 1.072×, 1.596× and 2.377× starting equity. Median maximum drawdown is 13.6%; its 95th percentile is 23.9%. Drawdown is measured from each path's own running peak. These are sample order statistics, so tied outcomes can put more than 5% of paths at a percentile boundary. The displayed loss count identifies paths ending below their starting equity. The bands describe this model and sample; they do not bound future market outcomes. No changing regimes, separate trading costs, funding flows or broker fills are simulated.
What to do with the numbers
Save the example's JSON, then change only risk per trade while keeping seed 7 and 2,000 paths. At 1% risk, the final 5th–95th percentile band is 1.043×–1.555× and 95th-percentile maximum drawdown is 12.7%. At 5% risk, those values are 1.064×–7.693× and 51.6%. Compare the full set of results rather than choosing a risk percentage from the median alone. A new sample seed changes the simulated draws; retaining the same inputs, seed and engine version reproduces the sample. Downloads record those settings, full-precision results and definitions locally. A zero observed loss count is not proof of zero probability. For a fixed threshold below starting equity, use the separate risk-of-ruin model; its threshold is not this tool's running-peak drawdown.