Home / Method
Method
How the model learns
Making picks is half the job. The other half is checking them. Every pick we publish gets graded the next morning, and what we find changes how future slips are built.
The loop
Grade. Tag. Wait for a real sample. Change the rule.
1. Graded against the box score
The morning after every game, each published pick is checked against the final box score. Every pick, not just the ones that make a good post.
2. Misses get tagged
A miss isn't just a miss. We tag it by market (points, rebounds, assists, threes, yards and so on) and by what happened in the game: a blowout that sat the starters, a minutes drop, foul trouble, or a live pick versus a pregame one.
Rule changes
3. Patterns need a real sample
A few nights can be luck. A pattern only becomes a rule change once it holds up across dozens of picks. Then it changes how slips are built: a market gets a tighter filter, a game script gets avoided, or a leg type gets capped.
Calibration
4. Our chances get checked
The model puts a chance on every leg. We compare those chances with how often picks at that level actually hit. If legs we rate around 35% land noticeably less often than that, the model is overconfident there, and we pull it back toward the market.
For how each pick is built in the first place (fresh FanDuel lines, a second source, game logs, injuries and defense vs position), see About.
What we don't publish
No records, no locks
We don't post a win-loss record, hit rates, ROI or profit claims. Grading is how we check our own work, not a sales pitch. These are model picks, not locks: a starting point you can water down, swap or skip.
21+ only. Bet only what you're fine losing. If gambling stops being fun, call 1-800-GAMBLER or see our responsible gaming page.
21+ | 1-800-GAMBLER