Tested on 3,959 NBA games
Understand every basketball game before the jump ball.
Calibrated win probabilities, projected scores and the reasons behind them, updated 24 h, 6 h and 1 h before every game.
The model’s favorite won 66.3% of 3,959 tested games. See the performance
All 30 NBA teams covered
Miami Heat
Next games
MIN Minnesota Timberwolves
—
MIA Miami Heat
—
Probabilities appear Tue, Oct 20, 7:30 PM ET, 24 h before tip-off.
DET Detroit Pistons
—
MIA Miami Heat
—
Probabilities appear Thu, Oct 22, 8:00 PM ET, 24 h before tip-off.
MIA Miami Heat
—
CHA Charlotte Hornets
—
Probabilities appear Fri, Oct 23, 6:00 PM ET, 24 h before tip-off.
DAL Dallas Mavericks
—
MIA Miami Heat
—
Probabilities appear Sun, Oct 25, 7:30 PM ET, 24 h before tip-off.
MIA Miami Heat
—
ATL Atlanta Hawks
—
Probabilities appear Tue, Oct 27, 7:00 PM ET, 24 h before tip-off.
IND Indiana Pacers
—
MIA Miami Heat
—
Probabilities appear Thu, Oct 29, 8:00 PM ET, 24 h before tip-off.
Tonight
Friday, October 9, 2026
No NBA games on Friday, October 9, 2026.
Tip-off times are shown in your time zone.
How it works
- 01
The model computes
Calibrated probabilities, learned from ten NBA seasons and tested on games it had never seen.
- 02
Updated before every game
24 h, 6 h and 1 h before tip-off: form, rest, schedule and matchup, as they stand at prediction time.
- 03
A clear summary, nothing invented
A short text written automatically from the model’s numbers only; every number is checked before publication.
Matchup analysis
Enter the two teams you want to analyze.
Next meetings, last meetings and the model’s probabilities.
The whole analysis in your pocket
On your phone or your computer, with nothing to install.
- Full analysis of every game: form, last results, team comparison
- Offense, defense and pace, rest and schedule, matchup
- Official injury report, shown for context
- Every player’s averages: points, rebounds, assists and minutes
- The game in brief: a summary written from the model’s numbers
- What changed between the 24 h, 6 h and 1 h updates
- Every game’s prediction history
Coming soon: the Pro offer opens shortly. Everything else on the site is free in the meantime.
Matchup analysis
Enter the two teams you want to analyze.
The model draws on ten NBA seasons for every probability.
Matchup to analyze
First team (e.g. Celtics)
VS
Opponent (e.g. Knicks)
Analyse the matchup
Total transparency
Performance figures are computed automatically from graded predictions, good and bad. Nothing is edited by hand.
- of the model’s favorites won
- 66.3%
- of the model’s favorites won
- games tested out of sample
- 3,959
- games tested out of sample
- log loss against Elo (lower is better)
- −0.0059
- log loss against Elo (lower is better)
Frequently asked questions
1. Is DunkScope free?
Yes: win probabilities, projected scores and the model’s performance are free, with no account. A Pro offer with the full analysis of every game opens soon.
2. How do the analyses work?
A statistical model computes the probabilities from past games: form, offensive and defensive efficiency, pace, rest, travel. It uses no data from after the prediction time. A short summary is then written automatically, by an AI, from the model’s numbers only; it never sets a probability.
3. Can I bet with DunkScope?
No. DunkScope is not a betting site: no odds, no picks sold as sure things, no betting advice. Probabilities, never guarantees. Betting is for adults only; if you bet, set limits.
4. Are the predictions reliable?
They are probabilities: a team given 60% loses about 4 times in 10. The Performance page automatically publishes the model’s results on every graded game and compares them with an Elo baseline.
5. When are probabilities published?
Published 24 h before every tip-off, then updated 6 h and 1 h before. Times are shown in your time zone.
Ready for tonight?
Probabilities, not guarantees. A team given 60% loses about 4 times in 10 when the model is well calibrated. DunkScope explains model outputs and gives no betting advice; betting is for adults only (18+ or your local legal age).