Real-time Win Probability and Latent Player Ability via STATS X in Team Sports

arXiv:2602.19513 · stat.AP, stat.ML · Submitted 2026-02-23 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: We're looking at the paper: "Real-time Win Probability and Latent Player Ability via STATS X in Team Sports".

Jane: Here's the gist of the paper. This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data.

Tom: The framework introduces a continuous dominance indicator (T-score) and a latent contribution index (STATS X) to decompose competitive advantage into interpretable statistical components.

Jane: Framework Foundation The core of the study is the introduction of a continuous dominance indicator, the T-score, which maps final scores to real values consistent with win/lose outcomes.

Tom: This T-score is formulated as a time-evolving stochastic representation called a T-process driven by standardized cumulative statistics.

Paper summary: Tom: So wrapping up this paper, "Real-time Win Probability and Latent Player Ability via STATS X in Team Sports" by Yasutaka Shimizu and Atsushi Yamanobe proposes a statistically grounded framework for evaluating win probability and player ability using minute-by-minute cumulative box-score data.

Jane: The authors introduce the continuous dominance indicator, the T-score, which maps final scores to real values consistent with win or loss outcomes, modeling it as a time-evolving stochastic representation called a T-process driven by standardized cumulative statistics.

Lu: This structure captures temporal game dynamics and allows for sequential updates of in-game win probability through this process. They also define the latent contribution index, STATS X, which quantifies a player’s involvement in favorable dominance intervals identified by the T-process.

Meng: The paper suggests that total statistical output doesn't equal dynamic contribution to win probability at all. The PTS metric integrates relative performance and momentum contribution into a dynamic evaluation measure.

Lalam: The main implication is that contributions to momentum formation might not be sufficiently reflected in static indicators like EFF or PER, which are formed as linear combinations of aggregated quantities like points and rebounds >

Tom: So for someone who just listens to the show, this means we can start looking at player ability not just as a final tally of stats but as a measure of their impact on the game's actual trajectory.

Jane: It’s an interesting look at how we can use cumulative data to build a picture of dynamic competitive advantage that isn't just about the final scoreboard.

Conclusion: Tom: So we’re wrapping up this look at the paper "Real-time Win Probability and Latent Player Ability via STATS X in Team Sports."

Jane: It basically says they built a way to look at game data minute by minute to figure out how likely the team is to win right then and there.

Lu: They use something called a T-score, which is like a continuous number that tries to predict the final score outcome based on what's happened up until that moment.

Meng: So instead of just looking at the final score, you get this evolving measure of dominance in real time? That sounds like it could help coaching in games where you need to make big decisions quickly.

Lalam: Yeah, and they also have this index called STATS X which tells you how much a specific player actually contributed to that good flow the T-score was tracking.

Tom: It’s not just looking at total points or rebounds anymore, it’s about who was on the court when things were going right. That distinction is what really stood out to me from a modeling standpoint.

Jane: Right, because they showed that traditional stats like PER don't capture that moment-to-moment momentum contribution as well as this new PTS metric does. It shifts the focus from static output to dynamic influence.

Lu: The authors found that for B.League data, things like points and defensive rebounds were the main drivers in their T-process model, which is a pretty concrete finding about what matters statistically in those games.

Meng: From a production standpoint, it’s interesting that they pinpoint specific stats as the primary drivers; if you’re building an evaluation system, knowing which inputs are most predictive helps you design the right pipeline.

Tom: It shows that even in complex sports, there are core statistical components that drive the overall trajectory of success. It moves us away from just counting things and toward understanding their actual effect on the game flow.

Jane: And what this means for us as a listener is that when you watch a game, you might start seeing not just what happened, but *why* it felt like the team was gaining or losing momentum at that exact second.

Lu: The paper also points toward future work, specifically using deep time-series models to learn how the T-process actually behaves over long periods of play.

Meng: That sounds like a big computational lift; learning those complex dynamics on real traffic is always the hardest part for me when I’m thinking about deployment costs.

Tom: It’s a roadmap, then, showing that this kind of mathematical representation can be built on top of raw data to give us richer insights than just looking at simple averages.

Yasutaka Shimizu, Atsushi Yamanobe

Department of Applied Mathematics, Waseda University

stat.AP, stat.ML

Submitted: 2026-02-23

Updated: 2026-10-29

Project page: https://www.bleague.jp

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 75/100

The gist: This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data.

Key concepts

T-score
A continuous indicator that maps final scores to real values, consistent with win/loss outcomes. It evolves over time based on cumulative statistics, capturing the temporal dynamics and momentum of the game.
STATS X
A latent contribution index that quantifies a player's involvement in 'good flow' intervals identified by the T-process. It measures what fraction of time a player was on court when the game was exhibiting favorable dominance.
T-process
A stochastic process modeling game dynamics, combining deterministic contributions from statistics with random fluctuations like Brownian motion. It evolves over time to capture how competitive advantage changes throughout the match.
Player Total Score (PTS)
The aggregated measure of a player's contribution, calculated by summing their STATS X across multiple games. This metric is proposed as a dynamic evaluation tool that integrates relative performance and momentum.

Terminology

Summary

This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. p. 0

Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

The framework introduces a continuous dominance indicator (T-score) and a latent contribution index (STATS X) to decompose competitive advantage into interpretable statistical components. p. 0
Where the paper says it

Through this stochastic formulation, competitive advantage is decomposed into interpretable statistical components. Furthermore, we define a latent contribution index, STATS X, which quantifies a player’s involvement in favorable dominance intervals identified by the T-process.

Framework Foundation

The core of the study is the introduction of a continuous dominance indicator, the T-score, which maps final scores to real values consistent with win/lose outcomes. This T-score is formulated as a time-evolving stochastic representation called a T-process driven by standardized cumulative statistics. p. 0

Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

The framework captures temporal game dynamics and enables sequential, analytically tractable updates of in-game win probability. p. 0
Where the paper says it

We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics. This structure captures temporal game dynamics and enables sequential, analytically tractable updates of in-game win probability.

Modeling the Dynamic Process

The T-process is extended to a stochastic process that evolves over time, called the “T-process,” which includes both deterministic and random components. The deterministic component is specified by the function representing the contributions of each STATS, Fα: R d → R. p. 7

Where the paper says it

” That is, we regard performance as fluctuating over the course of a game due to both deterministic and random components. The deterministic component is specified by the function representing the contributions of each STATS, Fα : R d → R.

The random component is modeled as a continuous-time random fluctuation, specifically a standard Brownian motion. p. 7
Where the paper says it

The deterministic component is specified by the function representing the contributions of each STATS, Fα : R d → R. The random component is modeled as a continuous-time random fluctuation.

A modified T-process, mT(t), is proposed by incorporating the opponent's fundamental strength into the initial value.

Player Contribution Indexing

The latent contribution index, STATS X, quantifies a player’s involvement in favorable dominance intervals identified by the T-process. This is achieved through an X-index that measures the fraction of time during which a player was on the court when the game was in a “good flow”. p. 1

Where the paper says it

Because their structure is dominated by scoring contributions, they are unable to adequately evaluate contributions that form or sustain the real-time “flow” of a game, particularly in crucial moments, as well as indirect effects that are not explicitly recorded in the statistics. Metrics such as plus-minus measure point differentials while a player is on the court, but they cannot disentangle the influence of simultaneously participating teammates and lack a theoretical framework that decomposes structural ability and within-game fluctuations.

The overall contribution aggregated over n games is defined as the Player Total Score (PTS). p. 2
Where the paper says it

It is a natural choice for sports with relatively low scores and limited variation in total points across games, such as soccer. On the other hand, in high-scoring sports such as basketball, the score difference tends to depend on the overall scale of the game.

Real-Time Evaluation and Application

The win probability at time t is defined as a conditional probability, PWt:= P pT(1t) > c Ft. This dynamic evaluation index is continuously updated as the game progresses, playing a role analogous to the “AI evaluation value” displayed in shogi broadcasts. p. 7

Where the paper says it

Since PCS weights each PSS by its distance from the center (mean) of the PSS values, player evaluation should essentially be based on PCS. 4 Dynamic Modeling of the T-score The T-score defined above is a single index at the end of the game.

The framework suggests that “winning can be viewed as optimizing the duration of IoF”. p. 6
Where the paper says it

1 and condition (3. 1), we have E[T k ] = α0, and α0 can be viewed as the basic (average) T-score that the team possesses from the outset.

Player Evaluation Comparison

The study compares Player Contribution Score (PCS) and STATS X with existing metrics like EFF and PER. It is observed that total statistical output does not equal dynamic contribution to win probability. p. 0

Where the paper says it

This work provides a basis for advancing AI-driven sports analytics. Keywords: sports statistics, statistical STATS, latent contribution, dynamic win probability, stochastic process model, MSC2010: 62P25; 60G99, 68T05.

The PTS metric integrates relative performance and momentum contribution into a dynamic evaluation measure. p. 7
Where the paper says it

In this sense, P CSk,j is obtained by adding, to the “equal allocation of the team’s fundamental strength,” a quantity in which the within-game statistical deviations are reallocated as relative weights. Therefore, PCS is not so much an estimator of long-run ability, but rather an index providing a within-team relative evaluation in each game, and can be interpreted as describing a performance allocation structure under game-by-game scale adjustment.

Future Directions

The paper suggests future extensions include formalizing “intervention times” as stopping times to describe the moment when the flow begins to reverse. It also points toward using deep time-series models for learning the T-process dynamics and employing causal inference for generating counterfactual evaluations. p. 1

Where the paper says it

In doing so, we develop a generalizable and extensible evaluation system that aims to reconcile “a lightweight statistical infrastructure implementable in practice” with “compatibility for integration with artificial intelligence technologies. ” Although this paper does not attempt a full AI implementation and adopts relatively simple statistical models, various AI-based extensions can naturally be envisioned, including nonlinear evaluation functions, integration with deep time-series models, strategy optimization through reinforcement learning, and connections to causal inference.

The proposed framework aims to bridge mathematical modeling and artificial intelligence in sports data science. p. 1
Where the paper says it

The proposed framework does not compete with AI; rather, it provides an interpretable foundational model that supports further sophistication through AI. For the empirical analysis, we use official game data specially provided for the Sports Data Science Competition organized by the Sports Statistics Section of the Japanese Statistical Society.

Data Analysis Results

The empirical analysis on B.League data shows that the estimated parameters for Chiba Jets indicate that “points (PTs) and defensive rebounds (DR) appear to be the primary drivers of the mT-process”. Furthermore, player evaluation metrics like PTS are shown to be distinct from static metrics like EFF and PER. p. 1

Where the paper says it

Existing performance metrics widely used in practice possess structural limitations. Taking basketball as an example, commonly used indices such as EFF and PER are static measures formed as linear combinations of aggregated quantities such as points and rebounds, and they do not explicitly account for temporal structure or context dependence.

The research concludes that contributions to momentum formation may not be sufficiently reflected in such static indicators. p. 1
Where the paper says it

Taking basketball as an example, commonly used indices such as EFF and PER are static measures formed as linear combinations of aggregated quantities such as points and rebounds, and they do not explicitly account for temporal structure or context dependence. Because their structure is dominated by scoring contributions, they are unable to adequately evaluate contributions that form or sustain the real-time “flow” of a game, particularly in crucial moments, as well as indirect effects that are not explicitly recorded in the statistics.

Acknowledgements

The study expresses gratitude to Data Stadium Inc., which provided the minute-by-minute play-by-play data for each B. p. 0

Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

League team, and to the Institute of Statistical Mathematics, Research Organization of Information and Systems (Statistical Thinking Institute), for their invaluable support. p. 0
Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

The constructive and insightful comments from the organizers, judges, participants, and others greatly contributed to the development of this research. p. 0
Where the paper says it

Keywords: sports statistics, statistical STATS, latent contribution, dynamic win probability, stochastic process model, MSC2010: 62P25; 60G99, 68T05. 1 Introduction In recent years, with the rapid development of artificial intelligence technologies, the use of data in sports has entered a period of major transformation.

References

[Billingsley 99] Billingsley, P. Convergence of Probability Measures, 2nd ed. p. 1

Where the paper says it

Existing performance metrics widely used in practice possess structural limitations. Taking basketball as an example, commonly used indices such as EFF and PER are static measures formed as linear combinations of aggregated quantities such as points and rebounds, and they do not explicitly account for temporal structure or context dependence.

, Wiley, New York, 1999.

[BLeague] B.League official website. p. 1

Where the paper says it

The proposed framework does not compete with AI; rather, it provides an interpretable foundational model that supports further sophistication through AI. For the empirical analysis, we use official game data specially provided for the Sports Data Science Competition organized by the Sports Statistics Section of the Japanese Statistical Society.

https://www.bleague.jp/ (accessed 2021-11-01).

[Hollinger 03] Hollinger, J. Pro Basketball Prospectus: All-new 2003–2004 ed. p. 1

Where the paper says it

Existing performance metrics widely used in practice possess structural limitations. Taking basketball as an example, commonly used indices such as EFF and PER are static measures formed as linear combinations of aggregated quantities such as points and rebounds, and they do not explicitly account for temporal structure or context dependence.

, Brassey’s sports, Washington, D. p. 0
Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

C, 2003.

[Yamanobe 24] Yamanobe, A. p. 0

Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

and Shimizu, Y. p. 0
Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

Win-Lose Prediction in Team Sports and Players’ Potential Assessment Beyond STATS. p. 0
Where the paper says it

Real-time Win Probability and Latent Player Ability via STATS X in Team Sports Yasutaka Shimizu∗ , Atsushi Yamanobe Department of Applied Mathematics, Waseda University February 24, 2026 Abstract This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics.

Sports Data Science Competition Proceedings, 2024, 304–307. p. 1
Where the paper says it

The proposed framework does not compete with AI; rather, it provides an interpretable foundational model that supports further sophistication through AI. For the empirical analysis, we use official game data specially provided for the Sports Data Science Competition organized by the Sports Statistics Section of the Japanese Statistical Society.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Dynamic Win Probability Tracking: AI systems can continuously update win probability using PWt = P pT(1t) > c based on real-time cumulative statistics, allowing for a dynamic evaluation index that shows how the win probability changes with the flow of the game.

  2. Flow-Driven Player Evaluation: AI can calculate Player Total Score (PTS) by combining a player's PCS and STATS X (X-index), which measures how much of the game the player participated in while being in an On Fire state, providing a structural evaluation that separates baseline strength from game-specific performance fluctuations.

  3. Predictive Strategy Optimization: AI can use the modified T-process, mT(t), to predict future T-scores at a time u based on information up to time t, enabling the computation of predicted T-scores pT(ut) and subsequently win probabilities for strategic decision-making during the game.

  4. Anomaly Detection for Intervention Identification: AI can be used for automated detection of intervention intervals by formulating the problem as anomaly detection or change-point detection to estimate "the probability that Delta mT(t) < -epsilon persists for a certain duration."

  5. Nonlinear Win Probability Modeling: Instead of assuming a linear mapping F alpha, AI can replace it with a neural network approximation F theta : R d -> R to capture nonlinear relationships and interaction effects among the eight STATS, substantially enhancing the expressive power of the model.

Abstract

This study proposes a statistically grounded framework for real-time win probability evaluation and player assessment in score-based team sports, based on minute-by-minute cumulative box-score data. We introduce a continuous dominance indicator (T-score) that maps final scores to real values consistent with win/lose outcomes, and formulate it as a time-evolving stochastic representation (T-process) driven by standardized cumulative statistics. This structure captures temporal game dynamics and enables sequential, analytically tractable updates of in-game win probability. Through this stochastic formulation, competitive advantage is decomposed into interpretable statistical components. Furthermore, we define a latent contribution index, STATS X, which quantifies a player's involvement in favorable dominance intervals identified by the T-process. This allows us to separate a team's baseline strength from game-specific performance fluctuations and provides a coherent, structural evaluation framework for both teams and players. While we do not implement AI methods in this paper, our framework is positioned as a foundational step toward hybrid integration with AI. By providing a structured time-series representation of dominance with an explicit probabilistic interpretation, the framework enables flexible learning mechanisms and incorporation of high-dimensional data, while preserving statistical coherence and interpretability. This work provides a basis for advancing AI-driven sports analytics.

Related papers