HOR is the forward horizon the edge is measured over. EFFECTIVE is the window the most recent trigger’s claim covers — trigger date to trigger plus that horizon in sessions — green with the sessions remaining while it is open, measured against the last closed session. WINDOW · LIVE hides every row whose window has run out. Weekends are skipped and holidays are not, so a window closes a day early rather than late. Score is a continuous, monotone function of the robust t-statistic, the winsorized edge and the effective sample size, with every threshold a study parameter rather than a global — so a score means the same thing whichever study produced it. It is deliberately not colour-banded: read it beside the t and the adjusted p in the same row, which are the quantities it is derived from. A candidate whose measured edge points against its own stated direction carries no score at all — the test is two-sided and can return one, and it is evidence against the rule rather than a weaker version of it. Those rows sort last rather than being hidden. 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.

n EFF is the independent-equivalent sample size. The raw trigger count in brackets is larger because forward windows overlap; every statistic in this table is computed against the effective count, and event-study OLS, Newey–West lag 84 is what the t is built on.

AS OF is when that signal was last genuinely evaluated. A run skips a study/ticker pair whose inputs hash identical to the previous run, so a date behind the newest one means the result still stands on the same data — not that it was recomputed and agreed.

ADJ p is Benjamini–Hochberg across every test the study performed for that ticker — the family a trade is decided within, and the bar the filter above applies. BATT BH and BATT BY correct across the whole battery instead: eighteen unrelated studies over a shared universe have an arbitrary dependence structure, which is the case BH’s guarantee is not built for and Benjamini–Yekutieli is. Both sit on the row rather than behind a toggle, because a signal that clears BH and fails BY is telling you how much of its significance rests on that assumption.