Most models copy the market.
Ours never sees it.
The whole edge rests on one thing. Our number is formed without ever seeing the price it has to beat.
Two ways to price a bet
Derived top‑down, “+EV”
- Start from posted odds.
- Strip the vig.
- Blend books into a consensus.
- Bet the outlier.
Finds slow books. Cannot find a mispriced game.
Originated bottom‑up
- Start from raw event data. No odds in the input.
- Model the full distribution of outcomes.
- Convert it straight to a fair price.
- Bet only a large enough gap.
The number exists before any line is seen.
Raw data to a bet.
Ingest
Play-by-play and lineups. No odds.
Rate
Signal separated from noise.
Simulate
The game run forward, many times.
Price
Distribution becomes a fair price.
Compare
Under threshold, passed.
One distribution.
Every line on the board.
Fit a separate model per market and the prices stop agreeing. The book then looks wrong exactly where you are wrong, which manufactures edges that were never there.
Legs are not
independent.
A 55% leg and a 52% leg. Multiplied, they look like 28.6%. But when one lands the other is likelier to, so the true number is 36.0%. Price the legs separately and every parlay, alternate line and player prop built from them inherits the error. One simulation gets the dependence for free, because the players are splitting the same game.
Illustration. The dependence is measured from play‑by‑play, never fitted to the market.
The bet is the gap.
That gap is the bet. Most games never get here.
The things you are right to ask.
Does it lose?
Yes. Losing seasons are on the results page and stay there.
Why CLV over win rate?
A few hundred plays is mostly variance. Beating the close is not.
Do you bet these?
Every one of them, my own money, at the same prices you get. It has gone well enough that I have not needed a backup plan. The record is public either way, losing seasons and all.
How many a day?
The threshold decides, not a quota. Some days, none.
See where the book has it wrong.
Every play with the model price, the book price and the edge, in Discord.