Attention and learning strategies reveal distinct dimensions of psychiatric diseases.

Precise measurement of individual variation in psychiatric symptoms is essential for developing scalable tools that can ultimately inform treatment development and clinical care. Individuals performing the same cognitive task often adopt qualitatively distinct strategies that may reflect differences in underlying neural processes and psychiatric symptom dimensions. Yet most cognitive paradigms implicitly assume that all individuals rely on the same cognitive strategy, attributing behavioral variability to differences along a shared computational axis. This assumption may obscure meaningful cognitive heterogeneity that cuts across traditional diagnostic categories, which themselves group phenotypically and biologically diverse individuals. Even dimensional approaches may fail to capture distinct computational profiles if they assume a common underlying cognitive process across participants. To test this hypothesis, we combined a computationally grounded Context Generalization task involving different visual stimuli (color, shape, texture, size) with standardized self-report psychiatric questionnaires. We tested how distinct cognitive strategies relate to symptom dimensions commonly observed in autism spectrum disorder (ASD), attention-deficit/hyperactivity disorder (ADHD), obsessive-compulsive disorder (OCD), depression, and schizotypy. Participants (N=744 in session one; N=584 in session two) were recruited online, enriched for self-reported ASD diagnoses, and matched for sex at birth. We identified qualitatively unique strategies related to goal-directed attention and short-term memory, such as a focused goal-directed strategy and a frequency-based strategy that relied on attending to all the stimulus features. The goal-directed strategy was associated with lower behavioral rigidity, while the frequency-based strategy was associated with elevated behavioral rigidity, despite showing the highest task performance. Strategy membership was substantially stable across sessions. Although transitions from the random-like strategy to the frequency-based strategy occurred less often than expected by chance, the subset of individuals who made this transition showed elevated behavioral rigidity and inattention-related symptoms consistent with the broader frequency-based profile. These findings demonstrate that individuals vary substantially in the cognitive strategies used to solve the same task and highlight the importance of measuring inter-individual variation in computational strategy rather than relying solely on aggregate performance metrics. More broadly, our results support a framework for linking psychiatric phenotypes to interpretable patterns of attention and learning, advancing efforts toward computational phenotyping and precision psychiatry.