StackLab

Why the number isn't a multiply

A multiply is wrong in basketball, and it's wrong in both directions at once. Which one you get depends on whether your legs are one player or two.

Stack one man's night and the legs ride together. We measured it on 79,638 real player-games across three seasons: his points and his assists both going over hits 28.7% more often than the multiply says, and his rebounds and made field goals 36.2% more often. Pair his points with his points-rebounds-assists and it's 92.4% — the same night counted twice. A multiply prices that slip as a much longer shot than it is.

Two teammates fight instead. One ball, about a hundred possessions, so a teammate's big night comes partly out of yours: two players both going over on points lands 4.7% less often than the product, and on made field goals 4.2% less. Here the multiply flatters the slip, which is the trap, because that's the stack that looks diversified.

Opponents barely move at all — within half a percent of the plain product on points and rebounds. The pace both teams play lifts both benches a little, and that's the whole of it.

So StackLab doesn't multiply. Every game here is simulated thousands of times off the players' own game logs, and each posted line carries a record of which runs it hit. The chance a whole slip lands is the share of runs where every leg hit at once.

What the EV number means

EV is our win probability times what the slip pays, minus one. Positive means our model makes the parlay worth more than it costs. Negative means the book is ahead of us, and that's the common answer — it should be. Books hold a margin on every leg, and a parlay multiplies it.

Which is why the leg-count strip matters. Inside a fixed payout band, each extra leg multiplies the same hold again, so EV falls as the count rises. The strip shows that shape on the pool in front of you.

Fill the rest won't pick a leg we can't stand behind. 5 of the 14 legs we can compare sit more than 10 points from the book's own number, and our pricer calls a gap that size a stale line or an injury we don't know about rather than an opportunity. The model is unbiased on average here and wide game to game, so those are the rows where it's likeliest to be the one that's wrong — and ranking by EV would otherwise reach for them first.

They're still on the board, and you can still lock one. That's your call to make with the number in front of you; what the model won't do is choose it for you and call the result an edge.

Every leg here is one a book is quoting

A leg only appears when a sportsbook is posting that exact line and we hold simulated runs behind it. The percentage beside a player is the same one his prop board shows — one model, one number, two pages.

The runs come from his own games. A player's line is priced off the shape of his last 40-plus box scores rather than a bell curve, because a counting stat isn't shaped like one, and the simulation draws whole box scores rather than one stat at a time. That's why his points and his rebounds move together by exactly as much as they really do.

We also publish how the model does against the market, and on NBA player props the market wins. Nothing here is sold as a pick. It's our number next to the book's, and a payout that's honest about what the legs do to each other.