Divot Lab is a golf analytics site I build and run on my own.
I'm Jake. I play golf, I read a lot of confident claims about it, and I wanted to see the numbers those claims were based on. Most of the time the numbers weren't there. So I started working them out myself.
The site does two things. It analyses PGA Tour play using strokes gained, and it applies the same approach to amateur golf so you can see where your own shots are going.
The main tool here weights each part of the game by how much a given course rewards it, then ranks the field on how well each player matches.
I tested it properly in August 2026, using 322 tournaments of round-level data. Trained on seasons through 2023, tested on 2024 and 2025. The weights turned out to have no predictive value. Once you controlled for how good the players were, the ones the model liked best finished top-20 slightly less often than the ones it liked least.
The weights are now derived by regression and kept only if they hold up on seasons the model never saw. Twenty-one courses didn't clear that bar and are back to neutral. At those tournaments the site tells you there's no fit signal rather than showing you a number.
If something wasn't measured, you'll see a dash instead of a zero. If a course has too little history to support a claim, the page says so. If a player rating was built on partial data, it's marked, and hovering it tells you what was missing.
The betting record works the same way. Every pick is logged when it's made and graded either way. Five weeks this season, all five picks lost.
Tour strokes gained and betting lines come from DataGolf. Historical rounds come from an archive I built covering 126,270 rounds since 2019, which is what makes out-of-sample testing possible at all. Amateur benchmarks come from USGA research. The models and the weights are mine.
If something here looks wrong, tell me. hello@divotlab.com, or Instagram.
Tournament recaps, key stats, and what to watch next — free forever.
No credit card. No spam. Unsubscribe anytime.