Keep unresolved chart-pattern outcomes in the evaluation

Rule deadline vs data cutoff: right-censored outcomes

The deadline belongs to the predefined rule: for example, target before stop within ninety days of entry. The actual observation cutoff belongs to the dataset. If data ends earlier and neither event has occurred, the outcome is not fully known. This is right censoring. The missing later price path does not become a loss merely because it is absent from the file.

Also specify what happens to cases that reach the regular ninety-day deadline without hitting either level. An event statistic can classify them as no target success within the deadline. A trading strategy additionally needs an executable time-based exit to establish its actual gain or loss. Failure to reach a target does not automatically mean a losing trade.

Why a success rate from completed cases can mislead

Fast target hits and early stops enter the completed subset sooner than slow-moving positions. Its rate need not equal the eventual rate for all original cases. Show known successes, known non-successes and unresolved observations together. If you are not yet using a suitable method for incomplete follow-up, transparently derived outcome bounds can be more useful than a seemingly precise rate from a selected subset.

Distinguish a planned administrative cutoff from a failed feed. If a data gap hides prices, even the assumed order of target and stop may be unknown. Record why observation ended. Do not extend the original success deadline afterward only for cases still waiting to reach their targets.

Example: 6 successes, 1 stop, 3 unresolved, bounds 60–90%

Ten invented entries follow the same predefined rule: the target must occur before the stop and within ninety days. Six entries have an observed target success and one has an earlier stop. At the dataset cutoff, the other three have only thirty, twenty and ten days of follow-up, respectively; neither target nor stop has occurred.

Among the seven resolved cases, the success rate is 6 / 7 ≈ 85.71%. That describes this subset, not the final ninety-day outcome of the original ten entries. If none of the three unresolved cases later succeeds within its deadline, the cohort ends at 6 / 10 = 60%. If all three succeed, it ends at 9 / 10 = 90%. These are logical bounds for these ten cases, not a confidence interval for an unknown market probability.

Gap labels based on later closure used as entry signals

A related selection error occurs when a gap receives a label because it subsequently closes quickly, and that category’s closure rate is then used as an entry forecast. The later outcome is already embedded in the category. First record features known at the decision time. Keep subsequent closure, its timing and any retrospectively assigned category as outcome data.