The Signal Book Holds: VELO_DROP at 25.6% ROI Over 101 Graded Bets

What we found

The number worth publishing today is the one that survived grading: VELO_DROP is running 25.6% ROI over 101 graded bets. That is not a backtest curve and it is not a sample of a dozen — it is 101 settled positions. The second name on the proven list, TRAILING_LATE, sits at 10.4% ROI over 106 graded bets. Between them, 207 graded bets across two independent triggers, both positive.

We publish those with the caveat that belongs on every out-of-sample record: 101 bets is enough to be interesting and not enough to be certain. A 25.6% ROI over that count carries a confidence interval wide enough to drive a truck through. What makes it worth reading is that it is *graded* — settled, not projected — and that the second signal, built on a completely different trigger, points the same direction.

On execution, the honest read is that we are close to a coin flip against the closing number: average CLV of 0.0018 and a 47.7% beat-the-close rate. Below half. That is the diagnostic we care most about, because CLV is the leading indicator and grading is the lagging one. A signal book at 25.6% with a sub-50% close rate is telling you the edge is coming from selection, not from timing — and timing is the cheaper thing to fix.

The trends

Today's board (8 games) gives five directional reads worth stacking:

One. Concentrated money is landing pregame. The top campaign is $216,028 on 215BALTB Over 8 at 66% one-sided — big, but not unanimous. Compare that to $101,041 on BOS ML at 99% one-sided and $81,168 on TB ML at 98% one-sided. Those two are the near-consensus tickets; the Over 8 is the contested one.

Two. Lines are drifting toward the money. 335BOSPIT-BOS moved 49.5c → 52.5c (+3.0) over the last 20 hours, 337NYYTOR-TOR moved 50.5c → 53.5c (+3.0), and 515STLCHC-CHC moved 60.5c → 62.5c (+2.0). Meanwhile 335AZATL-ATL drifted the other way, 54.5c → 53.5c (-1.0). When the money and the line agree, there is nothing left to buy.

Three. A pitcher trend the market may be underpricing: Freddy Peralta (TB) is going the wrong way — 8.44 ERA over his last 5, 8.18 over his last 3, against a season mark of 5.33. The recent form is materially worse than the season line, which is exactly the shape that a total-based whale campaign like the $216,028 Over 8 would be pressing.

Four. Totals skew over, hard: 8 of 8 flagged total edges point over, led by SEA @ HOU where the model prices fair at 10.15 against a 7.5 line — a +2.65 run gap. Eight-for-eight in one direction is not a set of independent reads. It is one read, repeated, and it means our run-scoring environment assumption is the whole bet.

Five. The model leans underdog in 5 spots — BAL ML +130, WSH ML +153, STL ML +167 — and the top moneyline read is STL ML +167 at a model 41.6%, EV +11.1%. Read that against where the money sits: $906,280 across 12 campaigns, with the biggest single ticket at $24,703.

What it means

The useful finding is the split. The graded signals — VELO_DROP at 25.6% over 101, TRAILING_LATE at 10.4% over 106 — are the part of the process with a settled track record. The 8-of-8 over lean is the part that isn't: a directional cluster that size is a hypothesis about scoring environment, not eight findings. We are treating it as one position, not eight.

Next test, running now: split the total edges by park and by starter-form delta (the Peralta-shaped cases, where last-5 diverges sharply from season) and grade them separately. If the overs only pay in the starter-form subset, the environment assumption is wrong and the pitcher read is doing all the work. That is a falsifiable question with a number attached.

What would change our mind: if the 8-of-8 over cluster grades out negative while VELO_DROP holds, the totals model is miscalibrated on run environment and we rebuild it from the park layer up. 21+ where legal.