VELO_DROP holds at +22.1% ROI across 113 graded bets

VELO_DROP is one of 14 signals in our library that have reached a settled, graded record. It fires on in-game fastball velocity decay — the pitcher's velocity band falling off its own established baseline — and it prices the market's slowness to re-rate a starter who is visibly losing stuff.

The archive is now 113 graded bets: 61.9% hit rate, +22.1% ROI.

What we found

The first thing worth stating is that the number went *down*. At 97 graded bets the same signal was running +26.5% ROI. Sixteen bets later it sits at +22.1%. That is not a break in the signal — it is the shape you should expect from any edge measured on a small sample: the early estimate is the noisiest one, and additional data pulls it toward whatever is actually there. If the ROI had gone *up* over that stretch we would trust it less, not more, because the most common way a betting record improves with sample is that the sample is being selected.

We do not claim 22.1% is the true edge. We claim 113 bets is the sample, 61.9% is the hit rate over that sample, and the direction of travel between the n=97 and n=113 readings is downward. A 113-bet record still carries wide error bars. It is enough to keep a signal in production. It is not enough to size aggressively off, and we do not.

Context from the rest of the graded library, all settled records:

- VELO_DROP — +22.1% ROI, n=113 - SPIN_VS_AVG — +9.4% ROI, n=157 - TRAILING_LATE — +11.8% ROI, n=91 - THIRD_TIME_THROUGH — +7.8% ROI, n=45

SPIN_VS_AVG has the largest sample at 157 and the more modest return. That ordering is informative: the signals with the biggest recorded edges are generally the ones with the fewest bets behind them, which is the reason we publish n next to every number rather than the ROI alone.

What it means

All four of these signals are pitcher-condition reads, and THIRD_TIME_THROUGH at n=45 is a hypothesis, not a result — 45 bets tells you almost nothing and we treat it that way. The interesting question for the next block of data is whether these are four signals or one. If VELO_DROP, SPIN_VS_AVG and THIRD_TIME_THROUGH are all detecting the same underlying thing — a starter past his effective workload — then their edges are correlated and stacking them is not diversification, it is concentration wearing a disguise.

That is what we are testing next: a co-occurrence study across the graded archive, measuring how often these signals fire on the same pitcher in the same outing and whether the joint cases outperform, underperform, or simply duplicate the single-signal cases. We will publish the result whichever way it lands.

What would change our mind: if VELO_DROP's ROI continues to decay at the rate it did between n=97 and n=113, it converges on the SPIN_VS_AVG range within another hundred bets — at which point the honest description is a modest edge, not a large one, and we would say so. 21+ where legal.