Scope Discovery — two-sided cohort scanner with rule emitter
What this is. Scans settled positions for compound cohorts where ROI diverges meaningfully from baseline in EITHER direction. Slices the cuboid (book × league × market × flow × mode × sharp pool × trajectory tag × game state × stake bucket × EV band × QK band × sharp-pool size × sharp-side aligned × day-of-week, 15 dims total) up to depth-3, applies separate-direction BH-FDR at q=0.10, and threshold-sweeps
minEv / minQk on the top-K most-actionable cohorts to emit scoped-rule patches with held-out PnL evidence. Winner candidate = n≥100, posterior ROI ≥ +5%, P(>+2%) ≥ 0.85, persistence pass. Winner confirmed = n_train≥200, ROI ≥ +7%, P(>+2%) ≥ 0.95, BH-FDR pass, both halves ≥ +3%, holdout ≥ +3% (looser than bleeder side: false-positive winners cost capital). Bleeder symmetric on the negative side with confirmed holdout ≤ −5%. Pairs with the Leak Detector (one-sided bleed scanner) and Segment Explorer (no-shrinkage bootstrap CIs).
Filters
Click Run Scan to begin.
Positions
—
awaiting scan
Slices Evaluated
—
cuboid 1-D / 2-D / 3-D
Surviving Prune
—
n ≥ 100 floor
Winners
—
▲ profitable cohorts
Bleeders
—
▼ losing cohorts