Handle Score offers a quantifiable way to measure ball-handling through AWS AI and Machine Learning.
Handle Score is a new way to quantify the art of ball-handling, measuring how effectively a player navigates the court and penetrates the defense off the dribble.
Handle Score tells us whose handles are truly elite, assigning a quantifiable way to measure what has traditionally been a mostly subjective skill.
Historically, evaluating handles has been purely a matter of the “eye test.” Fans and analysts have debated the best ball handlers for decades, but there has never been an objective way to measure it. A between-the-legs crossover that freezes a defender and a basic dribble that doesn’t lead to points have both just been tracked as “dribbles” Now, with the power of AWS AI and Machine Learning, the eye test just got smarter with the introduction of a new official NBA stat: Handle Score.
Handle Score is the NBA’s first analytic category designed to quantify how effective and impressive a player’s ball handling is on any given play. Leveraging positional tracking data and powered by advanced Machine Learning models, Handle Score captures the full picture of a handle, from the complexity of the dribble moves to the defender separation created and distills it into a single score that encompasses the effectiveness of that handle.
The NBA’s optical tracking partner, Sony Hawk-Eye Innovations, uses 3D pose detection to track 29 points on every player’s body, and the ball, 60 times per second. Those detailed body-position coordinates, along with the spatial context of the surrounding nine players on the court, fuel the Machine Learning model, capturing the spatial relationship during a ball handling sequence.
The key innovation behind Handle Score was borrowing a technique from the world of large language models: human preference alignment. Instead of telling the model what a good handle looks like through a set of rules, the team asked a fundamental question repeatedly: “Which handle is better, A or B?”
Basketball experts reviewed thousands of handle clips side by side, selecting which one they believed was the better display of ball handling. Over 10,000 of these A/B comparisons were collected across more than 1,000 plays spanning the 2023-24 through 2025-26 NBA seasons, using the NBA’s own Top 50 Handles highlight videos as anchors. The model learned directly from these human judgments, training itself to predict and score handles the way a knowledgeable fan would.
Handle Score combines two complementary approaches to evaluate a handle from different angles.
The first approach analyzes the ball handler’s full movement pattern, consuming the 60-frames-per-second tracking data to understand the flow, rhythm, and creativity of the handle as a complete sequence. It captures the things that are hard to put into words but easy to recognize when you see them.
The second approach breaks the handle down into its building blocks: the number of dribbles, behind-the-back moves, spin moves, and the defender separation created. This gives the model a concrete, countable perspective on what happened during the play.
By combining these two views, Handle Score captures both the art and the science of ball handling. A fluid, creative sequence that leaves a defender in the dust will score high. The score reflects the degree to which the model believes a given handle would be preferred over others. A truly standout handle earns a score approaching 95 or above.
Handle Score captures something that has never been measurable before: the quality and artistry of ball handling. It goes beyond counting dribbles or tracking outcomes. It answers the question fans have always debated but never had data for: whose handles are actually the best?
Handle Score is designed to spark debate, expose fans to the world of elite ball handling, and bring a new lens to a part of the game that has always been appreciated but never quantified.
Handle Score is part of the next generation of stats and insights that the NBA is bringing to fans in real time with the power of AWS AI.
