The idea engine surfaces buys far more often than sells. It isn’t the bull market doing it — it’s the way edge is measured.
The long tilt is structural, not regime. The same scanner that returns ~18 : 1 long on equities returns short-leaning ideas on commodities. A rising tape adds to it, but the engine would lean long in a flat market too — because every setup is graded against a benchmark that already drifts up.
Across a normal session the boards are lopsided. The daily equity scan, the weekly scan, and the STORM board all come back almost entirely long. The natural read is “the bull market makes buys work.” The data says the cause sits one level deeper.
Long vs short ideas surviving each scanner’s edge filter, one representative session.
Same pattern family, same forward-study filter, run across four universes.
That is the whole tell. If a bull market were the cause, the identical method on commodities and ratios would lean long as well. It doesn’t — commodities come back net short. So the bias lives in the equity universe specifically, which points at how the scoring works rather than at the weather.
The unconditional forward win rate of a long equity in our sample — buy any name, any week, hold to the horizon. A short starts from the mirror image: a 39% base it has to overcome before it looks like a coin flip.
Every setup is kept only if its forward win rate clears ~50% and its median is positive. But the long benchmark isn’t 50% — it’s 61%. Longs clear the bar with room to spare; shorts have to beat an eleven-point headwind just to qualify, so most are culled before they ever reach a board.
We scan today’s large caps — the names that drifted up to earn a place in the index. The companies that would have generated the short edge have already been delisted out of the sample, so the forward-return history itself is tilted upward by construction.
This is the part the bull market actually owns. In a rising tape, buy-the-dip hammers print far more often than topping stars, so the raw signal count skews long before any filtering. Real, but the smaller effect — it rides on top of the two structural ones above.
A long idea’s headline win rate is part drift, part skill, and the two are easy to confuse. A name that wins 63% against a 61% base is barely saying anything. A short that beats its 39% base by twenty points is a genuinely rare, high-information signal — even though its raw win rate looks unimpressive next to the longs.
The honest metric is already computed: edge over base (favw − base_favw). Rank on that, not on the raw number, and the drift stops masquerading as edge.
Sort boards by edge-over-base rather than raw win rate, so a strong short isn’t buried under mediocre longs.
Put the base rate on every Radar card, so the drift is visible and a “63%” long is read against its 61% floor.
Commodities and ratio spreads carry no drift — they’re where balanced and short ideas actually live.
The weekly regime lens already flips a nominal long to short when the edge inverts in the live regime — the direct antidote to a stale drift-driven read.
AlphaQuery Intelligence · internal methodology note · figures from a single representative session’s scanner output; the 61% base is the direction-adjusted unconditional weekly-forward win rate on the equity universe. Descriptive — not investment advice.