Methodology
A model posts the card, the card never changes, and every play is graded at the line we published. This page is the full rulebook.
Moneylines only: the team or side to win, at a posted price. No spreads, props, parlays or “locks.” Each sport runs its own card and its own record.
| Sport | Card | Free pick | Posts |
|---|---|---|---|
| NFL | One official card per week, Thursday through Monday, never edited after it posts. | Pick of the Week | Wednesdays, 8 AM ET |
| MLB | One card per game day, posted once morning moneylines are up. | Pick of the Day | Mornings, once lines are up |
| NBA | One card per game day from opening night, with a back-to-back rest adjustment. | Pick of the Day | Around 11 AM ET on game days |
| CS2 In testing | One card per day of series moneylines from named events and known orgs. | Pick of the Day | Around 5:30 AM ET |
Every model outputs one number per game: the probability each side wins. Edge is that probability minus the probability implied by the posted moneyline. A +4% edge means the book is pricing the side about four points cheaper than the model does (more on edge).
Team strength from season win percentage, run differential per game and pitching, plus a small home-field bump, passed through a logistic curve. Early in the season the gaps are shrunk toward 50/50 while samples are small. A calibration step fitted on our own graded picks pulls overconfident probabilities back toward 50% (why we show Brier score).
Elo ratings built from every game back to 2015, with a 35-point home-field adjustment, a margin-of-victory multiplier capped at 28 points, and a 40% pull toward average each preseason. It was tuned on seasons through 2024 and tested on the full 2025 season it never saw before it could post a pick (Brier 0.222 against a 0.250 coin flip).
Elo ratings from every game since 2014-15, with a 50-point home-court adjustment and a 50-point penalty for a team on the second night of a back-to-back. Tested on the untouched 2025-26 season: Brier 0.208 against 0.247 for “the home team wins at the league rate” (how the NBA model works).
Elo ratings from settled series with no home advantage, weighted by map score. Both teams need five or more settled series before either can make the card, and only named events or known orgs qualify for the free pick. CS2 is in testing; see the CS2 record for why.
A model’s accuracy is not a profit claim. The records are the only thing that says whether any of this works.
One play per card is public. It has to clear a floor on model win probability and sit inside an edge band. The band has a ceiling because our own graded history showed the biggest edges were the worst plays. If nothing qualifies, there is no free pick that day. We never force one.
Only the free pick is public before games start. Every other side on a card shows its matchup and start time, but the side, line, edge and write-up stay out of the page until that game is graded. Then it opens in full and joins the record. Pro will open the whole card when it posts.
Live cards show the model’s win probability and edge on the free pick. Archive pages and ledgers leave those numbers off.
No guarantees, no “locks,” no promise of profit. Losing stretches happen and stay posted (why variance matters). This is a model’s opinion on games, for information only. If you bet, bet what you can afford to lose. Gambling problem? Call 1-800-GAMBLER.
Daily Picks Free is independent. No sportsbook owns or pays for our picks. Kalshi sign-up links carry a referral code, disclosed on that page.