Right Move, Right Time: Multi-Sport Space Evaluation Platform for Ultimate Frisbee, Basketball, and Soccer

This paper introduces an open, sport-agnostic platform that standardizes player tracking data to evaluate usable space and optimal off-ball run timing across Ultimate Frisbee, basketball, and soccer, demonstrating a practical path for consistent spatial analysis in invasion sports.

The Gist

Imagine you are a coach for three different teams: a Ultimate Frisbee team, a Basketball team, and a Soccer team. Even though they play on different fields with different balls, you all face the exact same nagging question during film review:

"Where is the empty space on the field, and exactly when should my player without the ball start running to get there?"

If they run too early, the defense catches them. If they run too late, the space is gone.

This paper introduces a new "Universal Sports Translator" (a software platform called OpenSTARLab) that helps coaches answer this question for any of these sports using the same set of tools.

Here is the breakdown of how it works, using some everyday analogies:

1. The Problem: Speaking Different Languages

Right now, analyzing a soccer game is like speaking French, and analyzing basketball is like speaking Spanish. The data looks different, the rules are different, and the computer programs used to analyze them are built specifically for one sport.

• The Goal: The authors built a "Rosetta Stone" for sports data. They took messy, sport-specific data (like player coordinates and ball positions) and translated it all into a single, clean "Universal Language." Now, a computer can look at a Frisbee play and a Soccer play and understand them in the same way.

2. The Core Idea: The "Empty Seat" Analogy

Think of the field as a crowded party.

• The Ball/Disc is the person holding the microphone.
• The Runners are people trying to get close to the microphone to hear the speech.
• The Defense are people trying to block the path.

The platform calculates "Usable Space." It's not just asking, "Is there an empty spot?" It asks, "Can my player actually get to that spot before the defense blocks them?"

3. The Secret Sauce: The "Time Machine" (Counterfactuals)

This is the most creative part of the paper. The software doesn't just watch what happened; it acts like a Time Machine.

Imagine you are watching a play where a player runs and gets the ball, but it was a tight squeeze. The software asks:

"What if this player had started running one second earlier? Or one second later?"

It simulates these "What If" scenarios instantly:

• Scenario A (Real life): Player runs now. Defense blocks. Bad result.
• Scenario B (Time Machine - Early): Player runs early. Defense is confused. Great space!
• Scenario C (Time Machine - Late): Player runs late. The space is already filled by a defender. Bad result.

By comparing these timelines, the software tells the coach: "Your player actually ran too early. If they had waited just a heartbeat, they would have had a clear path to the goal."

4. Testing the Translator

The authors tested this "Universal Translator" in three ways:

• The Ultimate Frisbee Testbed: They used a new dataset of professional Ultimate Frisbee (a sport where you can't dribble, so timing is everything). They found that the team that won didn't just make "better" plays; they consistently found the right moments to run, creating more "safe zones" for their players.
• The Soccer Transfer: They took a model originally designed for Basketball (which accounts for dribbling and passing) and tried to use it on Soccer.
  • The Result: It worked better than the old Soccer-only models! It was like taking a Swiss Army Knife designed for camping and realizing it was actually perfect for fixing a car engine, too. It predicted scoring chances more accurately because it looked at the path the ball would take, not just the destination.

5. Why This Matters

Before this, a coach had to hire a different analyst for every sport, or use different software that couldn't talk to each other.

The Big Takeaway:
This platform proves that space and timing are universal concepts. Whether you are throwing a Frisbee, shooting a basketball, or kicking a soccer ball, the math behind "finding the open space at the right time" is the same.

By using this tool, coaches can stop reinventing the wheel for every sport. They can use the same logic to train their players, saying: "Don't just run when you feel like it. Wait for the 'green light' that the data shows is the perfect second to strike."

In short: They built a universal remote control for sports analysis that helps coaches see the invisible clock ticking on every play, no matter what game they are playing.

Technical Summary

Here is a detailed technical summary of the paper "Right Move, Right Time: Multi-Sport Space Evaluation Platform for Ultimate Frisbee, Basketball, and Soccer."

1. Problem Statement

The paper addresses a critical gap in sports analytics: the lack of a unified, sport-agnostic framework for evaluating off-ball movement timing and usable space across different invasion sports (Ultimate Frisbee, Basketball, and Soccer).

• The Core Question: Coaches in these sports face the same tactical dilemma: Where is the usable space, and when should an off-ball run start to maximize scoring potential?
• Current Limitations:
  • Fragmentation: Existing solutions are "sport-specific silos" with bespoke workflows that cannot be reused or compared across sports.
  • Heterogeneity: Differences in field geometry, event vocabularies, and sampling rates hinder "apples-to-apples" comparisons.
  • Timing Blindness: Most models evaluate where space exists but fail to quantify when a run should initiate. Detecting the onset of a cut within dense trajectories and linking it to value signals is difficult.
  • Portability: Models calibrated for one sport (e.g., soccer) often fail to transfer to others (e.g., basketball or Ultimate) without significant retuning.

2. Methodology

The authors propose OpenSTARLab, an open-source, pip-installable Python platform that standardizes tracking data and provides sport-agnostic spatial evaluations. The pipeline consists of three main components:

A. Preprocessing Package (openstarlab-preprocessing)

• Function: Converts heterogeneous raw tracking and event logs from Ultimate, Basketball, and Soccer into a unified Space Data Format.
• Key Features:
  • Unified Space Class: Organizes metadata, team/player identities, pitch dimensions, and frame-by-frame positional data.
  • Standardized Pipelines: Handles merging, timestamp alignment, coordinate normalization, and missing value cleaning.
  • Output: Produces structured DataFrames containing event logs and tracking data (player $x,y$ coordinates and ball/disc position) ready for analysis.

B. SpaceEval Package (openstarlab-spaceEval)

This module implements three distinct frameworks for spatial evaluation:

1. OBSO (Off-Ball Scoring Opportunity) Framework:

   • Based on Spearman (2018).
   • Calculates the probability of scoring from a location $r$ by decomposing it into: Probability of the ball arriving at $r$, Probability of the team controlling the ball at $r$ (using Potential Pitch Control Field - PPCF), and Probability of scoring from $r$.
   • Limitation: Assumes players move toward a static target location, ignoring interception paths.

2. CRSV (Counterfactual-Ready Space Value) Framework:

   • Focus: Specifically designed for timing analysis in Ultimate Frisbee.
   • Mechanism: Uses Weighted Ultimate PPCF (wUPPCF) to account for throw difficulty and marker obstruction.
   • Counterfactual Approach: It generates "what-if" scenarios by shifting the initiation time ($\xi$) of an off-ball run forward or backward while keeping all other players' trajectories identical.
   • Metric: $V_{timing} = V_{scenario}(0) - \max_{\xi \neq 0} V_{scenario}(\xi)$. This quantifies how much better the actual timing was compared to alternative start times.

3. BIMOS (Ball Intercept and Movement for Off-ball Scoring) Framework:

   • Focus: Addresses the interception limitation of OBSO, originally for Basketball.
   • Mechanism: Replaces the static target assumption with a Potential Ball Control Field (PBCF). Instead of calculating control at a fixed point $r$, it evaluates the probability of controlling the ball at the dynamic ball location along the delivery path.
   • Adaptation: The paper adapts BIMOS from Basketball to Soccer, incorporating dribbling and interception logic.

C. Evaluation Metrics

• Ultimate: Team-level distribution of $V_{frame}$ (play quality) and $V_{timing}$ (timing optimality). Analysis of "High-wUPPCF area ratio" vs. turnovers/goals.
• Soccer: Temporal stability (Pearson correlation between consecutive games), predictive power (correlation with future shots/goals), and calibration (Log-Loss against actual goal outcomes).

3. Key Contributions

1. Unified Multi-Sport Platform: The first open platform to standardize tracking data and apply consistent spatial evaluation logic across Ultimate, Basketball, and Soccer.
2. UFATrack Dataset: Release of a professional Ultimate Frisbee dataset (UFA) with synchronized video, tracking, and event annotations, enabling reproducible research in a sport previously lacking high-quality public data.
3. Cross-Sport Transfer Validation: Demonstrated that a model designed for Basketball (BIMOS), when adapted for Soccer, outperforms the standard Soccer baseline (OBSO) in predicting goal outcomes, proving the viability of cross-sport transfer.
4. Timing-Aware Analysis: Introduced a rigorous counterfactual method (CRSV) to isolate and quantify the impact of initiation timing on spatial value, moving beyond static spatial maps.

4. Results

A. Ultimate Frisbee Validation (UFATrack)

• Play Quality vs. Timing: Analysis of a match between Salt Lake Shred (SLC) and Oakland Spiders (OAK) showed that play quality (measured by max $V_{frame}$) correlated with the score margin (SLC generated more consistent medium-to-high quality plays). However, timing optimality ($V_{timing}$) did not significantly differ between teams, suggesting that in this specific match, throwing accuracy and receiver movement mattered more than the precise split-second of the run start.
• Space Control & Turnovers: Possessions where the offense maintained a High-wUPPCF area ratio (>0.4) consistently resulted in goals. Conversely, turnovers often occurred when this ratio remained low (<0.3), indicating a failure to create safe passing lanes under defensive pressure.
• Counterfactual Case Study: In a specific play, delaying the receiver's run by 10 frames (1 second) increased the scenario value ($V_{scenario}$) from 0.371 to 0.544, proving that a slight timing adjustment could have created significantly more usable space.

B. Cross-Sport Transfer (Soccer)

• Temporal Stability: Both OBSO and BIMOS showed moderate temporal stability (correlation ~0.4–0.48) between consecutive games, comparable to goals but lower than shots.
• Predictive Power: Both models correlated strongly with future shot volume (OBSO: 0.63, BIMOS: 0.82), outperforming goals (0.20), indicating they capture upstream attacking intent.
• Calibration (Log-Loss): The adapted BIMOS model achieved a significantly lower Log-Loss (0.3943) compared to the standard OBSO model (0.6508) when predicting goals. This indicates BIMOS provides better-calibrated probability estimates, likely due to its dynamic handling of ball interception paths (PBCF) versus the static target assumption of OBSO.

5. Significance

• Practical Utility: Provides coaches and analysts with a single workflow to answer the same fundamental tactical questions across different sports, reducing the need for bespoke, sport-specific coding.
• Scientific Advancement: Establishes a baseline for counterfactual timing analysis, allowing researchers to isolate the causal effect of "when" a move happens, not just "where."
• Emerging Sports: By using Ultimate Frisbee (a sport with strict no-dribbling rules and high timing sensitivity) as a testbed, the framework offers a template for analyzing other emerging or rapidly professionalizing sports like handball, lacrosse, and rugby sevens.
• Open Science: The release of the code (OpenSTARLab) and the UFATrack dataset democratizes access to professional-grade tracking analysis tools.

In conclusion, the paper successfully demonstrates that a unified, timing-aware spatial evaluation framework is not only technically feasible but also yields more accurate and interpretable insights than traditional, sport-siloed models.