The NBA has introduced Handle Score, a statistic designed to rate the effectiveness and distinctiveness of a ball-handling sequence. It uses positional tracking and machine-learning models to assess how a player moves through space and penetrates a defense.
Rather than counting dribbles alone, the system combines movement details into a relative assessment. Its training choices and expert preferences mean the result should not be treated as entirely free of subjective judgement.
Tracking and human preferences
Sony Hawk-Eye Innovations tracks 29 points on each player's body and the ball 60 times per second. The model also considers the positions of the other nine players on the court.
Experts compared pairs of clips and selected the better display of handling. More than 10,000 comparisons were collected across over 1,000 plays from the 2023-24 through 2025-26 seasons, with NBA highlight selections providing reference points.
The model learns to anticipate those preferences rather than relying solely on a fixed rulebook.
Two complementary assessments
One approach examines the complete movement sequence, including flow, rhythm and creativity. The other analyses countable elements such as dribbles, behind-the-back moves, spins and separation from a defender.
Their combination produces a score reflecting how strongly the model expects a handle to be preferred over others. Standout sequences can approach or exceed 95.
That number is not shooting accuracy, a scoring probability or a match prediction. It offers an additional analytical lens for discussing ball handling, rather than a substitute for every other form of basketball evaluation.