Contrast-Dependent Evidential Discrimination: Why a Signal Equally Probable Under Two Accounts Cannot Favor Either

This conceptual paper introduces Contrast-Dependent Evidential Discrimination, the structural constraint that a signal’s evidential weight depends on the alternatives and background against which it is assessed. When two accounts assign the signal the same positive probability under the same conditions, observing it does not change their relative odds. The paper assesses claims of support through selectivity relative to specified alternatives. Fit with one account is insufficient without considering how likely the observation would be under a relevant rival. A signal that cannot distinguish two accounts may still discriminate against a third or bear on another question. The discussion relates the constraint to Royall’s law of likelihood and examines the FBI’s 2004 misidentification of Brandon Mayfield. The author’s analysis emphasizes the relevance of close nonmatches selected by a database search. The case illustrates the need to account for candidate selection; it does not supply a measured likelihood ratio or establish equal likelihoods under the competing accounts. Part of the Possibility and Counterfactual Structure series within On Human Modeling, the paper offers questions for specifying the contrasts on which evidential judgments depend. Its framework can inform human–AI research on whether explanations, retrieved matches, and confidence judgments are assessed against relevant alternatives and the processes that produced the evidence.