About
Books anchor prop lines too heavily on season-long priors and react too slowly to recent form, playstyle matchups, and star absences. Every pick is a projection built from recency-weighted rolling windows, adjusted for opponent defensive context and playstyle signal, compared to the posted line. No black box - pick any player below and watch the maths run step by step.
How a pick is made
Blend recent form, weighted by playstyle
The base projection blends season, last-5, last-10 and last-20 averages. The weights differ by archetype: a high-usage scorer's recent games carry more signal than a role player's, so her recent windows are weighted heavier. This is the same blend structure the NBA model runs.
Adjust for opponent context and pace
The base number is scaled for the opponent's defensive context and the expected pace of the game. A projection into a fast, leaky defence is worth more than the same projection into a slow, elite one.
Convert the gap to a z-score
The difference between our projection and the bookmaker line is divided by the stat's natural game-to-game standard deviation. A 2-point gap means something very different on a 24.5-point line than on a 6.5-assist line; the z-score puts every stat on one scale of "how big is this gap relative to normal noise".
Only large gaps become picks
STRONG requires a gap of at least 1.0 standard deviations; Bet requires at least 0.7. Anything smaller is a Pass and never reaches the board. Most player-lines on most nights are priced about right - the model's job is to say nothing until the gap is bigger than the noise.
No black box, but no interactive toy either: the previous simulator on this page described a retired model and has been removed rather than left up as decoration.
We grade the model against the market itself
The sharpest single benchmark in sports betting is the closing line: the last price before tip-off, after every injury report, lineup note and sharp dollar has been priced in. If a model consistently sits on the side the market ends up moving toward, it is finding real information before the market does. That test does not depend on short-term win-loss variance, which is why it is the first thing we measure.
When lines move between our publish time and tip-off, they move toward our picks 83.6% of the time versus 60.3% on a control slate (p = 1.5×10⁻¹²) - measured on real captured closing lines across two seasons.
| Lines moving toward model picks | 83.6% |
| Control slate (same games, non-pick lines) | 60.3% |
| Probability the gap is chance | p = 1.5×10⁻¹² |
Captured closing lines, not simulated ones: every line in the sample was recorded from live bookmaker feeds at publish and again at close.
How picks are graded
- Every pick is logged before tip-off. No retroactive edits.
- Results graded against official WNBA box scores.
- WIN = actual stat in the predicted direction. LOSS = wrong direction. VOID = player DNP or game postponed. PUSH = actual exactly on the line.
- P&L assumes a flat stake at standard -110 vig (WIN +1.00u, LOSS -1.10u).
- Win rate counts only actionable outcomes (W+L). VOIDs and PUSHes are excluded from the rate.
Maturity and the forward gate
What's proven:the closing market consistently agrees with the model's direction.
What isn't yet: a sustained units edge at real prices - the backtest shows +2.5% ROI at best-available-price execution, and the model must accrue 300+ live picks with positive closing-line value before we make any performance claim. Until then this page sells you the method, not a win rate.
Every live pick is graded in public on the slate as it settles. When the forward sample clears the gate, the numbers will appear here with the same sourcing as everything above. If it doesn't clear, that will be said here too.
The price you take is the edge you keep
Every pick is priced at the best available bookmaker at the moment it publishes, and that exact price is what the track record grades against. The spread between the best and worst book on the same line is routinely larger than any model edge, so execution is not a detail - taking a worse number than the one we graded quietly erases the value.
The Terminal shows the full per-book ladder for every instrument - current best over and under, line movement since open, and which books are off-market - so you can take the number the record is built on, not a worse one.