A conditional edge asks what followed when a clearly defined combination was true—not what usually happened somewhere nearby on the chart.
Useful conditions describe information available now: trend distance, volatility, breadth, cross-asset leadership, momentum transitions, gaps, or time since an event. A discovery engine can combine those dimensions, but every surfaced rule must remain readable and reproducible.
A broad state may remain true for weeks. Counting every day as a new instance exaggerates the sample and makes markers blanket the chart. Occurrences should be settled into independent episodes with a cooldown suited to the condition. Day-specific gaps and crosses can use tighter spacing than persistent trend states.
A 70 percent positive rate is not meaningful if the ticker rose 72 percent of the time over the same horizon. Report the difference from the unconditional rate, the median forward return, downside, sample size, and the exact horizon together.
Searching thousands of combinations will surface lucky results. Adjusted significance, minimum practical effect, stability across neighboring thresholds, and out-of-sample checks reduce that risk. Freshness should describe whether the condition is active now—not favor historical examples merely because they occurred recently.