Meta-Coherence: A Cross-Domain Empirical Specification for Validating Induced Coherence Weights Without Correspondence Truth

Coherence-Weighted Information Theory and the induced-weights program that grounds it require empirical validation across the cross-domain regime where their universality claims live. Adaptive Realism and Synthetic Epistemology forbid correspondence-truth validation at the foundational level. The real is what persists across recursive coherence arbitration, and external ground-truth oracles violate the framework's ontology before any measurement is taken. I specify a meta-coherence validation architecture that operates inside the framework's commitments. Three within-domain signatures of self-consistent measurement are required jointly: cross-philosophy convergence (R2) across estimator classes with genuinely different inductive biases, persistence prediction (R1) grounding induced weights in Adaptive Realism's persistence criterion, and intervention asymmetry (R3) operationalizing the substrate-symbol distinction. A fourth condition, M5, carries the domain-translation invariance commitment as a structurally distinct admissibility gate sequenced before the within-domain matrix. The program tests these conditions across three domains spanning a substrate-stream transparency gradient, including a hidden semi-Markov source with structured emissions, Pfam protein families, and FOMC statements. An eight-cell outcome interpretation matrix binds in advance what each result configuration means for specific framework claims. The paper specifies the program; running it is the next layer.