Does turnover luck predict CFB covers? We ran 10 tests. Zero survived.

The cleanest way to fool yourself in this business is to test one angle, find a positive number, and publish it. So we don't test one angle. We test the whole family at once and report the family's answer, cherry-picks included.

This week the family was "native" college-football signals — angles built from what the box score itself says a team is, chiefly turnover luck and raw efficiency. Ten tests, 9,319 graded bets combined. The verdict: a weighted -2.79% ROI, and 0 of 10 tests survived our significance screen.

What we found

The headline hypothesis is intuitive and popular: teams whose recent margins were inflated by turnovers should be overpriced, and teams whose margins were deflated by them should be underpriced. Neither leg cashed. Fading teams whose recent margin was turnover-fuelled (+2.2) lost at -3.5% ROI over n=1,324 (p_raw 0.0886). Buying teams whose recent margin was turnover-cursed (-1.6) lost at -3.2% ROI over n=1,316 (p_raw 0.1088). Both legs of a symmetric hypothesis losing money is a strong hint the market has already priced the regression.

The closest thing to a positive read was fading extreme turnover luck, at +1.7% ROI — but only over n=582, with a p_raw of 0.3347. That is a coin flip wearing a lab coat, and we treat it as one.

Efficiency fared no better. Backing a yards-per-play edge of 2.0+ — the purest "this team is simply better" signal in the family — returned -4.7% ROI over n=487 (p_raw 0.1362). If a two-yard-per-play gap were free alpha, someone would have collected it before us. Someone did: the line-maker.

The bias hypothesis fails at n=9,319. That's not a shrug, that's a result — it tells you where the edge isn't, which narrows where it is.

The trends

While the corpus work ran, the live board kept moving, and it shows what a real, priced-in-doubt market looks like.

Concentrated money is landing pregame: the top campaign on today's board is $174,828 on 920SDSF SD ML at 96% one-sided, part of $270,704 in total tracked whale money across 2 campaigns on today's games. The same game drew $95,876 on Under 14 at 81% one-sided, and the biggest single ticket on the board today was $15,120. That kind of concentration is a directional read in itself — sharp-sized money leaning hard one way rather than spraying.

On the MLB side, the trends are pitcher-shaped. Bryce Miller (SEA, in SEA @ ATH) is trending the wrong way — an 8.71 ERA over his last 5 starts, 8.61 over his last 3, against a season mark of 4.22. That's not one bad night; that's a slope. On the other end, Cam Schlittler (NYY, in NYM @ NYY) is the hottest arm on the board at a 1.15 ERA over his last 5 with 10.9 K/9.

Totals skew over across today's slate: 5 of 5 flagged total edges point over, led by CIN @ MIL, where the model's fair total of 13.43 sits +5.93 runs above the 7.5 line — the biggest total edge on the board.

And the proven-signal book keeps holding out-of-sample: TRAILING_LATE is at 10.5% ROI over 289 graded bets, and THIRD_TIME_THROUGH sits at 9.9% ROI over 64 graded bets. That contrast is the point of this whole exercise.

What it means

Signals that survive family-wide testing — like TRAILING_LATE at 10.5% over 289 — get bet. Families that die at n=9,319, like native CFB turnover angles at -2.79% weighted, get retired, publicly, so nobody sells them back to you in October. Next in the queue: whether the concentration pattern we see in whale money — the 96% one-sided kind — carries information the closing line hasn't absorbed.

What would change our mind: a native-family cut that clears significance out-of-sample at comparable n, not a sub-600-bet coin flip. 21+ where legal.