We reran the boards two ways — ranked on raw win rate, then on edge over the base rate. Nearly half of what “passed” the old way was drift or worse.
Switching the board from raw win rate to edge over base is not cosmetic. On the weekly board, 14 of 30 ideas — 47% — carried zero or negative edge: the setup did no better, or worse, than simply owning the name. Five of them sat in the old top 16. The new ranking drops them and pulls up names with real, buried edge.
The companion note showed why the engine leans long. This one shows what that costs in practice — the gap between what looked tradeable and what actually carried an edge, name by name.
Weekly ideas that cleared the old filter with a headline win rate but a base-adjusted edge of zero or less. On the daily board it was 4 of 19 — and three of those were sitting in the top sixteen.
Three names that read as high-probability — until you put the win rate next to its own base.
A “61% winner” that does five points worse than just holding DELL. Old rank #4 → new #19.
Top-4 on the old board on the strength of 60% — a setup that quietly underperforms the drift. Old #4 → new #27.
Same story: a respectable-looking number that is pure drift, and a touch less. Old #8 → new #28.
The weekly tradeable top-16, before vs after — the drift-riders make room for buried edge.
Roughly eleven of the sixteen slots reshuffle. Nothing about the underlying signals changed — only the yardstick. Names on the left were never edges; the market was doing the work and the win rate took the credit.
COIN short ranks #1 on both boards. A short is scored against a 56% base, so its 64% win rate is a genuine +8 — the largest true edge anywhere, even though it looks middling next to a 75% long. The filter that demotes the drift-riders is the same one that lets the best short keep its crown. That’s the tell that it’s measuring skill, not direction.
Every Radar card now prints its edge over base alongside the win rate, and each board is ranked on it. A 60% idea sitting on a 60% base reads as a zero, in plain sight, so it can’t pass itself off as a setup. The bar to make a board is no longer “wins more than half the time” — it’s “wins more than doing nothing would have.”
AlphaQuery Intelligence · internal methodology note · companion to AQI-MN-2026-01, “Why the Book Leans Long.” Figures from one representative session; base = the direction-adjusted unconditional forward win rate on each name’s own history. Descriptive — not investment advice.