The Gate Between Constraint and Identification

This paper studies the boundary between accumulated constraint and licensed identification for temporally extended bounded observers. It distinguishes a source-side gate—where generated histories coincide or access channels merge their differences—from a retention gate, where an evaluator has lost a distinction that was previously available. Using distinguishability partitions and channel kernels, the paper shows that serial processing never recreates an upstream distinction, jointly available channels refine by kernel intersection, every final loss in a finite nested chain has a first-collapse stage, and external artifacts restore exactly the distinctions they encode. It derives verdict-sufficiency and repair criteria and, in the finite unrestricted-map class, an exact minimum repair alphabet. Finite evidence alone guarantees neither singleton nor nonsingleton candidate fibers; identification is relative to the declared candidate architecture, while declared richness forces non-identification. The resulting boundary is therefore not skepticism: effective evidence licenses identification exactly to the distinctions its surviving partition supports.