Local Fit Asymmetries in Cognitive Evaluation: A Trial-Level Diagnostic Stress Test

This technical note presents a controlled trial-level stress test of parameter recovery in drift–diffusion inference under partial observational objectives. Using simulated data generated from a standard Drift-Diffusion Model (DDM), we compare parameter estimates obtained from full trial-level likelihood optimization with those derived from restricted evaluation channels, including accuracy-only and reaction-time-only fitting. The analysis demonstrates that optimization under partial objectives produces systematic parameter divergence, even when the underlying generative process is stable and well-specified. Cross-evaluation of parameters across objectives reveals objective-dependent likelihood degradation, indicating that different evaluation targets emphasize distinct components of the same data-generating structure. The purpose of this note is diagnostic rather than theoretical. No claims of model inadequacy, impossibility results, or universal generalization are made. The results are confined to controlled simulation settings and illustrate how inferential outcomes depend on the structure of the chosen evaluation interface. This record contributes to the broader investigation of interface-dependent evaluation limits in cognitive modeling, without introducing new theoretical constructs.