Nonlinear amplification of human decision errors under time pressure and positional ambiguity: An interpretable machine learning analysis of chess

Time pressure and positional ambiguity are two fundamental cognitive constraints that threaten human performance in sequential decision systems like chess. However, the interactive and nonlinear nature of these factors has not yet been sufficiently quantified. In this study, 39,922 positions obtained from blitz games of seven elite chess players on the Lichess platform were processed using Stockfish engine analysis to examine how the probability of blunder changes according to time and positional ambiguity regimes. Cluster-robust logistic regression and histogram-based gradient models were applied comparatively. Permutation importance and SHAP values were used for explainability analyses. The findings reveal that the probability of blunder amplifies nonlinearly under the interaction of low remaining time and high engine evaluation difference. We proposed Amplification Index (AmpInd), used to measure this risk, showed that an additional error multiplier of approximately 5.1% occurred in the extreme time regime at a 300 centipawn ambiguity level. The HGB model achieved the highest discriminatory power (AUC = 0.806), and explainability analyses confirmed time pressure as the dominant determinant in model decisions. These results demonstrate that human errors are not random but are concentrated under specific combinations of cognitive constraints. We offer a quantitative framework for context-sensitive error modeling and provide generalizable findings that can form the basis for developing adaptive decision support systems in human-centered AI research.