CASE: Computerized Adaptive Strength Estimator for Chess

Accurately estimating a player's strength remains a challenging problem for adapting game AI to players of different strengths. Most existing approaches require players to complete or provide approximately 20–30 games before obtaining a reliable estimate, making the assessment time-consuming. Inspired by computerized adaptive testing (CAT), which sequentially selects informative questions according to the current ability estimate, we propose Computerized Adaptive Strength Estimator (CASE), a novel framework that estimates playing strength through adaptively selected board positions. Specifically, CASE uses an estimator to update the player's strength from the question-response history and a selector to choose the next question based on the current history and strength estimate. Experiments on chess show that CASE achieves over 80% rank accuracy with only 34 questions, whereas none of the compared baselines reaches this accuracy even after 200 questions. Compared with the prior game-based strength estimation method, CASE requires approximately 27 times fewer player responses to reach the same accuracy. Further analyses show that the selector identifies informative positions and adapts its question preferences according to the player's current strength estimate. These results demonstrate that accurate strength estimation can be achieved through a short adaptive assessment without requiring multiple complete games.