The Evaluation Trap: Benchmark Design as Theoretical Commitment
For AI researchers and evaluators, this work provides a meta-evaluative framework to detect when benchmarks fail to track independent capabilities, but it is primarily a conceptual contribution with no empirical results.
The paper argues that AI benchmarks embed theoretical assumptions that can create self-reinforcing evaluation traps, narrowing progress. It introduces Epistematics, a methodology to audit benchmark-design coherence, and demonstrates it on a case study showing how a proposed revision reproduces the constraints it aims to overcome.
Every AI benchmark operationalizes theoretical assumptions about the capability it claims to assess. When assumptions function as unexamined commitments, benchmarks stabilize the dominant paradigm by narrowing what counts as progress. Over time, narrow evaluation reorganizes capability concepts: architectures and definitions are selected for benchmark legibility until evaluation ceases to track an independent object and instead produces a version of the target defined by its own operational assumptions. The result is a trap: evaluation frameworks treat self-reinforcing assessments as valid, both creating and obscuring structural limits on what the current paradigm can accomplish. We introduce Epistematics, a methodology for deriving evaluation criteria directly from technical capability claims and auditing whether proposed benchmarks can discriminate the claimed capability from proxy behaviors. The contribution is meta-evaluative: an audit procedure, a failure mode taxonomy, and benchmark-design criteria for evaluating capability-evaluation coherence. We demonstrate the procedure through a worked audit of Dupoux et al. (2026), a proposal that revises the dominant paradigm's theoretical assumptions at the architectural level while reproducing them in its evaluation criteria, thereby entrenching the constraint it seeks to overcome in a form the evaluation cannot detect.