Freeze the Court Before the Signal. SACGU v5.3 is a closure-first protocol that forces the model, the nulls, the likelihood, and the evidence threshold to be frozen before any real-data verdict.

# Abstract We report a conservative closure-domain note on real interferometric signals within the R^8 framework. Motivated by the benchmark governance scale k_gov = 24, the Z48 -> Z24 visibility scaffold, and the protective-versus-risk distinction between 24 and effective 8 channels, we study sparse harmonic transport directly in closure-safe observables rather than beginning from image reconstruction. The analysis combines closure-phase harmonic summaries, ablation stress tests, sparse upward transport kernels, factorization checks, held-out consistency tests, and a corrected phase-mixture scan with stiffness/drag penalties and a source-side ultraviolet amplitude tether. For the previously prepared M87 target, the corrected phase-mixture scan is crystal-dominant with C* = 1 and prefers a sparse upward support {diag, +4, +8}. Within this toy language, the M87 readout is strongly 24-seeded: removing source m = 24 hurts the fit much more than removing m = 32, and the full template outperforms single-mode and rogue-only alternatives. For held-out Sgr A* tests performed directly on two real 2017 EHT uvfits lanes, both directions are again crystal-dominant. The LO -> HI direction is strongly 24-seeded, while the HI -> LO direction remains conservative and inconclusive at the seed level. We do not claim a unique physical transport basis, a first-principles curvature theorem, a direct observational proof of exact Z8 x Z3 factorization, or a raw-scale calibration of the ultraviolet anchor from the present normalized runs. The main result is a cross-source, crystal-dominant closure-domain transport pattern that survives the removal of the earlier glass-normalization loophole. The note also includes a compact summary of SACGU, the closure-first, source-agnostic uvfits-domain geometric disentangler developed alongside this analysis.