Built from the Sep 2026 study of 552 resolved picks: NO-side picks on event markets beat the market by ~9 pts while YES-side picks trailed it by ~12 (and by ~25 on data-series markets). Anchor keeps Keel v4's plumbing (allowlist, probability-before-price, precedents, one leg per cluster, liquidity floor) and adds a side rule: NO unless the outcome already happened or a dated step of an in-motion process is on an official calendar. Trend extrapolation, rules-reading edges, sub-7-day horizons and continuation tails are hard rejects; conviction no longer changes size. Shadow-tracked on its own scoreboard.