VELO_DROP: +19.4% ROI across 110 graded bets

Diamond Labs grades every signal the engine fires. Not the ones that looked good — all of them, win or lose, priced at the number we actually got. Fourteen signals have accumulated enough settled bets to say something. VELO_DROP is the strongest of them.

What we found

VELO_DROP has graded out at +19.4% ROI over 110 bets, hitting 60.6% of the time.

One hundred and ten is a real sample for a baseball signal, and it is small enough to be honest about. At n=110, a 60.6% hit rate carries a confidence band wide enough that the true edge could plausibly sit well below the observed one. What n=110 does establish is that this is not a handful of lucky Tuesdays. It is a signal that has been firing, and settling, long enough to have survived variance we did not choose.

The number that matters more than the ROI is the relationship between the two figures. A 60.6% hit rate on its own is meaningless without knowing the prices — you can hit 60.6% and lose money if you are laying heavy chalk to do it. The +19.4% ROI is the statement that the hit rate is arriving at prices worth taking. That is the whole claim: the market is not charging enough for what this signal identifies.

VELO_DROP is not alone. Two other signals have cleared the grading threshold with positive settled records:

- RELIEVER_COLD_FIRST — +17.8% ROI, n=42 - TRAILING_LATE — +14.7% ROI, n=137

TRAILING_LATE carries the largest sample of the three at 137 bets. RELIEVER_COLD_FIRST, at 42, is the one we would caution against reading too hard — 42 settled bets is early, and we are publishing it because it is in the ledger, not because it is proven.

What it means

The honest read is that three of fourteen graded signals have separated themselves, and the separation is visible in ROI rather than in hit rate alone. That is the shape you want. A signal that wins often at bad prices is a signal that has found a favorite, not an edge. A signal that wins 60.6% of the time and returns +19.4% has found a mispricing.

The open question on VELO_DROP is decay. Signals built on observable pitcher characteristics tend to erode as the information diffuses — velocity readings are public, and public inputs get priced. We are running the record forward without intervention specifically to watch for that. If the ROI compresses as n grows past 110, that is the diffusion showing up, and it is the most likely way this ends.

Next up: we are segmenting the 110 by price bucket to find out whether the edge is concentrated in a subset of the prices we have been taking, or spread evenly across them. If it is concentrated, the signal gets sharper and the bet count gets smaller.

What would change our mind: a sustained ROI decline as the sample extends past 110 would tell us the edge is being priced out, and we would cut the signal rather than defend it.

21+ where legal.