Correction: our exit-ladder hit rates were overstated
Published 2026-08-11. Until today this site showed a panel on the track-record page reporting how often a take-profit ladder reached each of its three targets. Those numbers were measured with a model of how trades fill that was too generous to the strategy. Corrected, the same backtest does not make money.
What was wrong
The backtest walked daily bars and decided whether a target had been reached using an intraday-low proxy that let a trade take profit on a bar it would also have been stopped out of. On any bar wide enough to touch both levels, the simulated trade collected the win. Real trading does not offer that choice, and over a long sample the difference is not a rounding error: it is the difference between a strategy that earns and one that loses.
Both sets of numbers
| Measure | What we published | Corrected |
|---|---|---|
| Trades in the sample | 186 | 151 |
| Reached the first target | 48.4% | 33.8% |
| Stopped out | 51.6% | 64.2% |
| Average result per trade | +0.087R | -0.222R |
| Return per unit of risk | +0.36 | -0.99 |
Same universe, same window, same ladder. The only change is the fill model. “R” is the result of a trade measured in multiples of what it risked, so an average of -0.222R means the average trade lost about 22% of what it put at risk.
What we removed
- The exit-ladder evidence panel on the track-record page.
- The endpoint that served it,
GET /api/exit-plan/backtest.
We did not replace them with a better-looking version of the same claim. A feature whose own evidence says it loses money does not belong on the site in any wrapper, and the corrected artifact stays in the repository so the failure remains checkable rather than deleted.
Why you are reading this
A number that goes quietly missing is worth less than a number that was never published. We would rather be the site that shows you its own broken result than the one whose record only ever improves. If you acted on the withdrawn figures, the corrected ones above are what the same test says now.
What this does not affect
This correction concerns the take-profit ladder’s measured hit rates and nothing else. The band-calibration backtest published at /calibration and the dated forecast record at /track-record use different artifacts and are unchanged by it.