Live sportsbook odds

The board works without this. To see live sportsbook prices beside the model's numbers:

  1. Sign up free at the-odds-api.com . It only asks for an email, no card.
  2. Copy the API key they email you.
  3. Paste it here and tap Save, then Load odds below.

Your key stays in this browser only. The free plan gives 500 credits a month: every game's lines cost 3, and each player stat adds about 1 per game. Odds only sit beside the model's numbers. They never change the projections.

Bet slip

Tap any price on the board to add it here. The slip adds up the chance the whole parlay hits, allowing for legs in the same game that move together.

Odds sit beside the model's numbers. They never change the projections.

What to build

Edges come from the model's chances against each price. They compound in a parlay, and so do the model's misses: if each leg's chance is a few points too high, a +20% parlay can be a losing one. The "if the model is 3 points high" line shows that case. Track slips here to find out which it is.

A parlay pays only at one book, so pick the book you bet with. "Best price at any book" mixes books and is for comparing, not for placing. Odds never change the projections.

Download (CSV)

Record

Slips hit

Units won

Return per unit

Units won vs what the model expected
Every tracked slip

Each slip is one unit at the payout when it was tracked. A pushed or voided leg drops out and the slip pays on the rest. Parlays hit rarely, so the record says little until there are dozens of slips; the Expected column is what the model said they'd win on average.

Projected margin

Projected total

both teams' points

Spread
Total
Matchup: each offense against the other defense (current ratings, league rank in brackets)
Key players in this game

Click a player to open them in the Prop Lab.

All games this week

Projection (average)

Median

Projected distribution

Recent games against this line
Every stat for this player in this matchup

Click a row to load that stat.

Chance the whole parlay hits

Fair odds (model)

If the legs were unrelated

Your legs
Model vs sportsbook

Leave it empty to use the legs' odds multiplied, which is how a parlay across different games pays. Books price same-game parlays their own way, so for those type the odds your book shows.

How the chance builds, leg by leg
One leg against every book
Legs that move together
Lines ranked by the model's edge

Edge is the model's chance times the payout, minus the stake: +10% means a $10 bet returns $11 on average if the model is right. Needs is the chance the price needs to break even. "+ Parlay" adds the bet to the Parlay tab with this price as its odds, and "Save" adds it to My picks.

Track record is how well the model's over/under chances for that stat held up on past games it never saw (hover for the numbers). Edges on Strong stats are the ones to trust most; on Weak stats the model has been no better than guessing the usual over rate, so their edges are mostly noise. Game lines are Untested: the game model is tested on winners and scores, not against spreads or totals.

A big edge often means a stale price or news the model hasn't seen, like a late injury or the weather, so treat the list as bets to check, not bets to make.

Record

Hit rate

Units won

Return per unit

Units won vs what the model expected
Record by group

Model said is the average chance the model gave those picks. If the hit rate keeps landing near it, the chances are honest; if it keeps landing well below, those edges were noise.

Every pick
Trend
Trend
Team model: accuracy on seasons it never saw (2021 to now)

Win probability calibration
Player models: typical error (RMSE) by method, 2025 to now
Over/under at each regular's obvious line (their recent average)
Do the injury inputs help? Player error (RMSE), 2025 to now

Gain = share of error removed. The Friday report is what's known before the weekend; known inactives come out 90 minutes before kickoff. Each stat uses whichever version was more accurate on the Friday report.

How often regulars missed the game, by injury-report status
Parlays: combined chances on seasons the correlations never saw (2025 to now)

Same-game parlays at regulars' obvious lines, each leg randomly an over or an under. "Unrelated" multiplies the legs' chances; "linked" is the Parlay tab's method. Log loss: lower is better. The last two rows are the pairs with the strongest links, by whether the picks lean with the link or against it.

About the data