Minimum Viable and Value-Optimal Evidence for AI-Assisted Tasks: A Truthfully Collectible and Game-Theoretic Proof-of-Work Assessment Framework

This paper develops a truthfully collectible proof-of-work framework for AI-assisted educational assessment, arguing that instructors should assess whether bounded, auditable evidence makes student reasoning, revision, validation, and ownership visible rather than relying on AI detection or exhaustive surveillance. Building on Minimum Viable Evidence and Value-Optimal Evidence, it specifies a seven-channel student evidence bundle and a green/amber/red instructor decision procedure for determining when AI-assisted work is assessable, when targeted follow-up is warranted, and when evidence is insufficient. The framework further treats evidence submission and instructor response as an incentive-design problem, using game-theoretic reasoning to make honest disclosure easier than concealment while preserving assessment validity, academic integrity, and proportionality.