Per-Trial Non-Vanishing Likelihood Divergence Under Objective Channel Reduction in Drift–Diffusion Models

This technical note examines whether likelihood divergence between full trial-level optimization and scalar accuracy-only (ACC) objective configurations in Drift–Diffusion Models (DDM) vanishes as sample size increases. Synthetic datasets are generated under fixed DDM parameters and fitted using full probabilistic likelihood. Parameter configurations constructed solely to match response accuracy proportions are evaluated under the same likelihood. Results show that: likelihood divergence scales approximately linearly with sample size, per-trial divergence remains strictly positive, divergence does not collapse under increasing N. These findings indicate systematic objective mismatch under probabilistic evaluation rather than finite-sample fluctuation. The note is strictly diagnostic. It introduces no alternative modeling framework and advances no normative claims.