Mirror Recognition in Stratified Agent Architectures: A Constraint-Governed Recognition Boundary Under Reflection

This companion paper defines mirror recognition in stratified agent architectures as a constraint-governed recognition boundary under reflection, not as direct self-perception, consciousness, or metaphysical selfhood. It argues that reflected perceptual structure should become behaviorally meaningful only through a bounded sequence of admissibility, short-horizon reflex salience, constrained self-relevance evaluation, and governed consequence. Reflected structure may support self-relevant inference only when accessible invariants justify it; reflex alone may not write memory, request promotion, or authorize irreversible consequence. Any durable self-related effect must occur through an explicit governed transition with provenance preservation and transition logging. The paper also includes concrete failure-mode examples and a pass/fail framing intended to support a deterministic companion harness.