<p>Psychiatry currently finds itself at a decisive crossroads, either continuing along the phenomenological path that has become the norm in recent decades, or alternatively, truly innovating by applying mechanistic principles in diagnosis, prognosis and treatment. Despite decades of progress in neurosciences contributing to elucidating the basic principles of brain function, the field still lacks a mechanistic framework explaining how mental disorders arise, or why they share features across descriptive diagnostic categories. Current classification systems lack a coherent paradigm of the processes by which the human brain generates psychiatric symptoms, obscuring computational mechanisms that may link neurobiology to subjective experience and targeted intervention. While predictive processing (PP) has provided a powerful theoretical account of brain function in recent decades, its potential clinical applicability has remained neglected. Here, we propose predictive dysfunction (PD) as a potential transcategorical framework that conceptualizes psychiatric disorders as disturbances in hierarchical predictive inference (PI), resulting from miscalibrated priors, prediction errors, and their associated precision weighting, and/or failure of belief updating. Specifically, PD reframes DSM-based categories as phenotypic expressions of underlying inferential configurations, and as such, conditions such as psychosis, trauma-related disorders, depression, or compulsivity reflect distinct but related patterns of imbalance in belief updating and/or precision weighting across levels of the inferential hierarchy. Importantly, PD aims to move beyond prior theoretical accounts by explicitly relating current computational constructs to clinically interpretable dimensions, and potentially also decision-making. However, despite the compelling theoretical appeal of PP, substantial methodological challenges remain before the PD framework can support either clinical research or decision-making in psychiatry. We delineate the current status by outlining the existing biomarker landscape, identifying a number of promising paradigms that may provide a bridge between formal computational theory and clinically relevant phenotypes, and finally, we propose a preliminary research agenda for clinical validation of PD. Only by eliminating the methodological obstacles that currently limit its advancement to the clinic, PD may ultimately facilitate a shift toward a psychiatry in which diagnosis and treatment are anchored in a coherent, computation-based pathophysiological model that complements existing classification systems and provides a foundation for future precision medicine and therapeutic innovation.</p>