A new preprint on arXiv presents Throwing4, a phase-aware pose-modeling approach for predicting throw distance in elite athletics. The authors argue that elite throwing events provide a structured biomechanical action sequence paired with an objective continuous outcome—the throw distance—making them an ideal testbed for performance-oriented sports analytics.

The proposed method appears to explicitly account for the different phases of the throwing motion, rather than treating the pose sequence as a single block. This phase-aware modeling is intended to better capture the dynamics that determine how far an implement travels.

Because the source is only the abstract, details on the model architecture, training data, or reported accuracy are not yet available. The work signals growing interest in using fine-grained pose information for predictive sports analysis, where the outcome is a precise measurement rather than a win/loss label.