Built an AI model called Ol Softeee.
Saw @jakecrumpler breaking down how Lodolo's stuff was way better than his results showed. Got me thinking: how many pitchers are like that? Nasty whiff rates, elite chase rates, but the K prop market is still sleeping on them.
So I built a model to find that gap. 24 features: whiff rate, chase rate, spin, edge zone command, contact suppression, fastball velocity trends, all crossed against the market's posted K lines.
Backtest: 93.7 score, 31-47% ROI range across 2.4M data points. Those numbers are absurd and won't hold in real life. Backtests never do. But the signal was strong enough to publish it live on https://t.co/vovFep4B3i today.
8 games on the board. Let's see.
Great example of a signal I think you can model. Adding it to the list. I think I can build something that flags recent positive moves in a pitcher's stuff and test whether it sharpens strikeout predictions and edge before the line adjusts.
Here's the splitter adjustment that #Orioles Brandon Young made on May 24th.
5 mph slower, 3" more drop, 4" less arm-side movement
Old: 87.7 mph, 5.4" iVB, 11.8" HB
New: 82.8 mph, 2.1" iVB, 7.8" HB
Not much to go on, model produced exactly 1 pick yesterday. This happens when the inputs kinda narrow down the pool of pitchers available for predicting on game day. Or the model just isn't finding edge. Will keep watching, may take a while to see if this is going to go anywhere.
Built an AI model called Ol Softeee.
Saw @jakecrumpler breaking down how Lodolo's stuff was way better than his results showed. Got me thinking: how many pitchers are like that? Nasty whiff rates, elite chase rates, but the K prop market is still sleeping on them.
So I built a model to find that gap. 24 features: whiff rate, chase rate, spin, edge zone command, contact suppression, fastball velocity trends, all crossed against the market's posted K lines.
Backtest: 93.7 score, 31-47% ROI range across 2.4M data points. Those numbers are absurd and won't hold in real life. Backtests never do. But the signal was strong enough to publish it live on https://t.co/vovFep4B3i today.
8 games on the board. Let's see.
@LouisAnalysis What's cool is you can see he's getting the fewest strikes of his career (41% zone vs league's 48.7%) and at the same time dropped his MLB chase rate to 27.6%, down from 35.4% in '24. Pitchers are ducking him but he's also refusing to expand the zone. Boom there's your walks.
Day 3: Harder Barrel is live.
34 picks. 24-10. 70.6% win rate, +17.5% ROI.
The thesis: contact quality predicts hits better than batting average. 34 picks later, it's holding.
What's inside: barrel rate, hard-hit%, exit velo, line drive rate, zone contact rate, crossed against opposing SP contact allowed and platoon splits. No batting average. No BABIP.
You can build a model like this yourself at https://t.co/vovFep4B3i
Andy Pages went 4-for-51 in the postseason. Right now he's slashing .333 with a 58.3% hard-hit rate (95th percentile).
Is this real? I'm building a model to find out. First backtest just came back. More soon.
Update: named it Harder Barrel.
16-4 record so far. +35.7% ROI across 20 settled picks.
Barrel rate and hard-hit% predict batter hits better than batting average. Early results say the thesis holds. More tuning ahead.