Was that equity curve skill, or one lucky ordering?
A backtest is one draw from a distribution. Shuffle the same win rate and reward-to-risk a few thousand times and you get the range of curves the strategy could have produced — including the ones that would have made you switch it off.
Diagnosis
A distribution does not tell you what to do. These checks compare it against a ruin limit, against how deep the tail goes, and against the one result you actually observed.
The range of outcomes
Read the outer columns as «this happened in one run out of twenty» and the middle as «this was the typical run». The gap between them is the part a single backtest cannot show you.
| Measure | Low 5% | Median | High 5% | Worst run |
|---|---|---|---|---|
| Final balance multiple of the starting balance | — | — | — | — |
| Maximum drawdown deepest peak-to-trough loss within a run | — | — | — | — |
| Worst losing streak consecutive losses within a run | — | — | — | — |
How the runs spread out over time
The middle column is the typical path. The outer two are the one-in-twenty cases on each side. They separate as the run goes on, which is why a short backtest can look like almost anything.
| After trade | Low 5% | Median | High 5% |
|---|---|---|---|
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
| — | — | — | — |
How this is calculated
Each simulated trade either wins the reward/risk multiple or loses one unit of risk, drawn independently. That independence is the model's main simplification: real strategies cluster their losses, because the conditions that break them persist for weeks. Treat the tail here as optimistic, not conservative.
The point is not the median. Every strategy looks acceptable at the median. What decides whether you keep running it is the one-in-twenty path — the drawdown you did not plan for, arriving in the month you had just increased size.
Win rate and reward/risk are point estimates from a finite sample, and the simulation treats them as exact. Moving the win rate by two points shifts this whole distribution more than changing the seed does, which is worth trying before trusting any of it.
Nothing is uploaded. Every number on this page is computed in your browser, and the link you copy carries only the values you typed.
Related tools
Nothing is uploaded and no trade history is read. The distribution comes only from the numbers you typed, so it inherits every error in them.