Matched Behavior and Source Use

Abstract

Successful action, accurate prediction, and actual use of a particular signal are different properties. Their separation is familiar, but an operational account must specify which organizational relation an intervention can identify and which alternatives remain indistinguishable. We give a finite, reproducible analysis of two consumers: an actuator and a pre-consequence predictor. Independently addressable source carriers have identical values in nominal trials, so different allocations of their consumers produce identical successful behavior and prediction. With fixed pure single-source consumers, a binary intervention code identifies whether the two consumers depend on the same root precisely when its source codewords are distinct. The resulting worst-case requirement is $\lceil\log_2 n\rceil$ probes for an unknown pair of roots, including deterministic adaptive protocols that observe only agreement or disagreement; one probe suffices when one root is independently known. These are applications of classical separation principles. Allowing both consumers independently to aggregate arbitrary nonempty sets of sources by OR changes the exact worst-case requirement to $n$. A four-source construction shows how a protocol valid for pure routes falsely accepts disjoint redundant source sets. Further countermodels separate root dependence from internal branch consumption, common ancestry from intake-occurrence identity, and prediction-state updates from comparator implementation. Exhaustive finite searches and event-model executions accompany the proofs. Under the inherited UCT correspondence axiom, actual constitutive source relations have experiential counterparts, but these results identify neither a particular feeling of agency or ownership nor a new threshold for basic experience. The contribution is a precise combination of target, mechanism class, probe cost, and interpretation boundary suitable for reuse in sensorimotor research. Theoretical preprint. Conditional formal results and finite computational checks are distinguished from empirical evidence. Substantial ChatGPT assistance under the author's direction is disclosed. Not peer reviewed. No new human, animal, or pretrained language-model experiment is reported, and C1 is not claimed to be empirically validated.