Cmi-Based Recoverability Versus Wilson-Loop Diagnostics in Z2 Lattice Gauge Theory (2+1D): Exact Diagonalization Benchmark on Small Open Lattices

We provide a finite-size benchmark testing whether a CMI-based recoverability proxy correlates with Wilson-loop confinement diagnostics in Z2 lattice gauge theory in 2+1 dimensions, computing ground states by sparse exact diagonalization on 2x2 and 2x3 open lattices (link qubits, Gauss-law penalty verified by ~ 1) and evaluating entropic quantities by pure-state Schmidt/SVD. We state a conditional bridge from a spatial area law to an operational "information horizon" via explicit hypotheses (H1)-(H4), with two recovery-length conventions (absolute, and normalized by the boundary prefactor Sigma_B(w) = |dB(w)| log 2). v2 (no v1 number is changed) adds a structural lemma that v1's own Table 1 was exhibiting unnoticed: for a pure global state, when the collar saturates (B = (A u C)^c) purity forces I(A:C|B) = I(A:C|empty) -- this is exactly why v1's Table 1 shows I(w=2) = I(w=0) = 0.4992999 to seven digits; the saturated row carries no buffer information, and the informative range of that benchmark is w in {0, 1}. v2 also anchors the non-monotonicity of CMI under collar enlargement as geometry-dependent (v1's wall geometry shows growth at w=0 to 1; a BFS-patch geometry on the same states decays monotonically -- there is no data-processing theorem in that direction); cross-links the CMI benchmark to the Petz-error twin on the same lattices (densified n = 8 sweep: Spearman rank-trend +1.00, permutation p = 1e-4, against both 1/sigma_eff and the inverse spectral gap -- so, as in the twin, confinement specificity is unresolved at these sizes); harmonizes the fidelity convention (v1 correctly uses root fidelity with I >= -2 log f, equivalent to the companions' squared-convention I >= -log F); removes an internal phase label leaked into v1's Section 1 title and a duplicated heading; and adds series positioning and a regenerable verification suite. The conditional proposition and its hypotheses are unchanged.