Computational Credibility and the Structure of Evidence Evaluation in Criminal Justice

Traditionally, criminal justice systems entrust testimonial credibility assessment to judicial discretion under free evaluation of evidence, creating structural flaws—opaque reasoning, inconsistent standards, and asymmetric error costs borne mainly by defendants, victims, and vulnerable witnesses—which this Article reframes as institutional design failures rather than individual shortcomings. To resolve this uncertainty, it proposes “computational credibility”: an intermediate mechanism formalizing credibility elements, enhancing reasoning transparency, and improving verifiability by repositioning AI tools, per computational law, as supporters of judicial structure—not judgment automation—with core principles of explainability, auditability, and functional limits. A modular AI architecture decomposes analysis into components like questioning patterns, testimony variations, and context, keeping outputs non-evidentiary and subject to judicial synthesis/verification, guided by purpose limitation, minimal intrusion, non-delegation, and strict confinement to preserve judicial responsibility. By shifting from AI adoption debates to institutional allocation of judgment errors, this public-choice framework rethinks credibility assessment in criminal justice.