Six steps, no black box. Every number on the site comes from this.
Pick'em apps pay fixed multipliers, which fixes the win rate each pick needs for the entry to break even. Below it, the entry loses money on average, no matter how good the picks feel.
| App | Entry | Win rate needed per pick |
|---|---|---|
| PrizePicks | Power 2 | 57.7% |
| PrizePicks | Power 3 | 58.5% |
| PrizePicks | Flex 5 | 54.2% |
| PrizePicks | Flex 6 | 54.2% |
| Underdog | Standard 2 | 57.7% |
| Underdog | Standard 3 | 55.0% |
| Underdog | Flex 5 | 54.8% |
| DraftKings Pick6 | 2 picks | 57.7% |
| DraftKings Pick6 | 3 picks | 55.0% |
Standard payouts. Apps change multipliers and run promos, so confirm in the app before entering.
DraftKings, FanDuel, BetMGM and other regulated books price the same player props with real money behind them; for the NFL we add Pinnacle, the book professional bettors use. Each book charges a margin, so we remove it to get the fair probability. Books are weighted by how sharp they are, and a line needs two books (or Pinnacle) before we price it.
The pick'em line is often a different number from the books', so we need the whole range of outcomes, not just one line. Instead of assuming a textbook shape, the model learns it: for every graded prop it records how the actual result compared with the market's expectation, separately for each stat and line size. A 20-yard receiver has a very different downside from a 70-yard one, about 1 in 8 receiving props finishes at zero.
Thin data leans on a standard statistical shape; as results pile up, the learned shape takes over. Stats without enough history are priced on the card but never published as edges. The model retrains after every day of games.
Tested out of sample, the model is honest across the range but slightly optimistic at the book's own line. So a pick is published only at a fair win chance of 56% or more, above the 54.2% flex break-even. Most days, most lines don't qualify. That's the point. See the test →
Picks are frozen before each game and graded from official box scores. A player who doesn't play is void, as in the apps. Every result stays on the record.
You don't have to take our word for any of this. Every published pick sits on the record with the win chance we gave it, frozen before the game, wins and losses alike, each pick counted once however many apps listed it. The proof page shows whether picks we called 60% actually hit about 60%. A pick published in error stays visible and labeled; it just leaves the headline number. New picks also go out on X @SlipSharpPicks the moment they publish, so the timestamps are public too.
On game day the card re-prices each pick as the game runs: what the player still needs against what the rest of the game should produce, based on the time left. At kickoff it matches the pregame number; at the final whistle it's a win or a loss.