From Ecological Prediction to Mechanistic Validation in Sport-Related Motor Decision-Making: A Scoping Review of Machine Learning, Computational Modeling, and Cross-Level Computational Coupling.

Sport-related motor decision-making offers a naturalistic context for studying action selection under uncertainty, time pressure, and changing perceptual information. Computational methods address different parts of this problem but support distinct inferential claims. Predictive accuracy alone does not explain decision formation, demonstrate cross-level coupling, or establish mechanistic evidence. This scoping review grouped studies by computational role. A dual-path search across five databases, supplemented by a targeted sensitivity search and citation searching, identified 57 eligible studies: 42 ML-only, 10 CM-only, and 5 Integrative. ML covered the broadest range of sport settings and data sources and modeled sport-specific actions, decision-contingent outcomes, action values, and policies. CM studies were fewer and represented decision processes through formal quantities such as evidence accumulation, thresholds, priors, sensory precision, and latent objectives. Five Integrative studies linked predictive and formal components through explicit directional computational dependencies. None linked ecological regularities to formal hypothesis revision, explicit computational coupling, and independent mechanistic validation within the same study. Independent behavioral, motor, physiological, and neural constraints were limited. Computational outputs were already used in assessment, tactical planning, training, and performance support, but these applications alone did not establish a mechanism. Overall, current evidence is concentrated in ecological prediction, while formal decision modeling, cross-level coupling, and convergent validation remain more limited. Based on this synthesis, we propose Bidirectional Interconstruction as a hypothesis-generating conceptual framework for linking ecological patterns to formal hypotheses, computational coupling, and independent validation while separating practical use from mechanistic evidence.