Hypothesis. Large cumulative moves over a lookback window mean-revert or continue depending on regime and magnitude. Edge is a winsorized (5/95) mean against a matched baseline; significance by Welch t and Mann–Whitney U; p-values Benjamini–Hochberg corrected across each study's parameter grid; overlapping events de-duplicated before statistics. Every signal carries per-regime success counts and its own fitted forward-return distribution.