Primetime public-team fade grades out flat: 32-32, 0.0% ROI at n=64

What we found

We tested the oldest idea in sports betting: primetime public teams are overpriced, so fade them. It needed a schedule/broadcast join to grade properly, and now that the join exists, the answer is unambiguous.

The hypothesis is NOT_SUPPORTED. Record 32-32 across n=64 bets over 32 events. ROI 0.0%. Raw p-value 0.5 — indistinguishable from a coin flip. Market-implied win rate 50.0%; actual win rate 50.0%.

That is the cleanest possible null result. The market is not systematically mispricing the popular side in primetime windows at any size we can detect. A n=64 sample cannot rule out a small effect, but it does rule out the large, easy one the folklore promises.

So we published the null and moved the search to where the money actually concentrates. The 2026-08-27 board, 7 games, is the working example.

The trends

Concentrated money is landing pregame, not in primetime. The top campaign on the board is $111,185 on 145AZSF Over 8 at 57% one-sided. That is the shape worth studying — a size-weighted pregame lean, not a broadcast slot.

The same game is one-sided on both moneylines. $85,970 on AZ ML at 86% one-sided sits alongside $79,301 on SF ML at 78% one-sided. Total tracked whale money across today's games is $276,456 across 3 campaigns. Notably, the biggest single ticket on the board today is only $5,891 — this is many hands agreeing, not one hand pushing.

The model leans underdog in 1 spot today and it is on the heavier-backed game. Top model moneyline read is SF ML +125, model 46.7%, EV +5.0% — worth reading directly against where the money sits. This is a pregame price, not a live in-game number.

Totals skew over. 6 of 7 flagged total edges point over, led by LAD @ ATL: model fair 8.41 versus a line of 6.5, a +1.91-run gap. That is the biggest total edge on the board and the kind of spread that survives a wide error bar.

Starting-pitcher form is trending hard in opposite directions in KC @ TOR. Spencer Arrighetti (TOR) is the coldest arm on the board — 9.95 ERA over his last 5, worsening to 10.8 over his last 3, against a season mark of 4.60. Across the same matchup, the hottest arm is Noah Cameron (KC) at 1.28 ERA over his last 5 with 7.2 K/9.

The proven-signal book is holding out-of-sample. VELO_DROP is running 17.3% ROI over 132 graded bets. TRAILING_LATE is at 14.5% ROI over 188 graded bets. Both have more than double the sample of the primetime test and both are still positive.

What it means

A broadcast slot is not information. Velocity decay and in-game game-state are. The two signals carrying real graded records — 132 and 188 bets respectively — are both built on measured physical or situational change, not on a guess about who the crowd likes.

That is the reason we are pointing the next round of tests at the size-weighted money itself rather than at the schedule. The 86%-one-sided AZ ML at $85,970 against a model that prefers SF +125 at 46.7% is a testable disagreement with a number on both sides. A primetime flag was never that.

Next up: grading a size-weighted-money-versus-model-disagreement signal on the same join, and a starter-form-delta test using gaps like Arrighetti's 4.60 season mark against his 9.95 last-5.

21+ where legal. We publish probabilities and their track records, not outcomes.

What would change our mind: a primetime split that clears 55% at n over 300 with the broadcast join held fixed — until then, 32-32 stands.