Transparency as Constraint: The Reasoning-Cost Effects of a Dual-Track Judge–AI Adjudication System — An Institutional Reconstruction Based on “Mandatory Input” and “Full-Process Disclosure”

Judicial discretion (free evaluation of evidence) is both a core value of judicial independence and an institutional breeding ground for abuse of discretion. Any oversight mechanism that attempts to “prevent” the abuse of judicial discretion faces an inherent contradiction: the more forceful the oversight, the greater the threat to judicial independence. This paper does not attempt to resolve this contradiction; instead, it proposes a “weak-constraint” path—imposing a “structural cost” on judges’ reasoning behavior through a dual-track judge–AI adjudication system, without adding any new disciplinary power. This paper designs two core institutions: “mandatory input of valid evidence into the AI system” and “full-process disclosure of AI input information.” Together they require that a judge’s deviation from the AI’s conclusion face a structured reasoning requirement at the level of the individual case; that the cumulative nature of case-by-case public data give rise, in the public domain, to a de facto reasoning-style profile reconstructible by third parties; and that the deviation threshold itself be subject to independent technical verification. This paper candidly acknowledges that this mechanism creates no new disciplinary power, cannot deal with extreme cases of judges who are entirely indifferent to professional reputation, and cannot fully guard against formally compliant reasoning packaged by generative AI. But its institutional goal is not to “catch abusers of power,” but to “make the abuse of power more costly and more visible.” Through the transparency and verifiability of information, and while preserving judicial independence, it forms a continuous but gentle institutional constraint on the abuse of judicial discretion.