Across 1316 evaluations (188 materials × 7 grids), not a single material changes regime. The grid sweep varied resolution by 7.6× (17→129 pts), κ_min by 10× (0.01→0.001), and κ_max by 2× (1.0→2.0). The structural regime is fully grid-configuration-invariant. This is a non-trivial result: a coarser grid could in principle misclassify a near-boundary material. Instead, all regime assignments are robust to every tested perturbation — confirming that the FULL_CONTINUITY/FRAGMENTED boundary is well-separated, not marginal.

The five κ_connect values {0.562, 0.750, 1.000, 2.000, 10.000} and their counts {2, 17, 155, 1, 1} are identical in all 7 grids. With k_points ranging from 17 to 129, the log-spaced grid changes entirely — yet the reported κ_connect values do not change. This is possible only if each material's actual connectivity threshold aligns precisely with a shared structural fixed point on the log-κ axis, not a grid artefact. The invariance across GRID_G (κ∈[0.001,2.0]) is especially compelling: even with 64× more range, no new κ classes appear (except Class-IV SnSe = 2.0, now resolved).

The oxide class shows 100% FULL_CONTINUITY across the entire normalized corpus — 69/69 materials, zero exceptions. Every tested oxide (Al₂O₃, Bi₂O₃, CeO₂, Cr₂O₃, Er₂O₃, Fe₂O₃, GeO₂, HfO₂, La₂O₃, MnO₂, Nb₂O₅, RuO₂, SiO₂, TiO₂, V₂O₅, WO₃, ZnO, ZrO₂, and 51 others) percolates under normalized representation. This is the strongest chemistry-class structural result in the corpus: oxide bonding geometry produces universally admissible spectral manifolds under normalization.

The 12 FRAGMENTED materials do not span a continuum of GR values — they are all exactly 0.700, with n = 10 gaps each. This discretization implies a shared normalized ladder structure: 10-point trajectories with exactly 3 gap-outliers that exceed the full ε-budget. The class spans chlorides, fluorides, sulfides, a selenide, an element (W), a sulfate (SO₄), and a mixed compound (SbSI) — chemically diverse but structurally identical under normalization. This suggests the fragmentation is a representational property (normalized ladder length = 10) rather than intrinsic material chemistry.

The raw-ladder corpus (57 materials) showed 24 MARGINAL_CONTINUITY materials (GR ∈ [0.92, 0.999]). The normalized corpus shows 0 marginal materials — every material is either fully percolating (GR = 1.000) or discretely fragmented (GR = 0.700). Normalization appears to resolve the marginal regime by eliminating representation-induced gap heterogeneity. The admissibility boundary sharpens from a wide marginal band into a hard binary: full connectivity or discrete fragmentation. This is consistent with the UNNS theoretical prediction that structural regime is fundamentally binary at the correct representation scale.

Of 368 HARD_FRAGMENTATION verdicts in the glass corpus, 296 (80.4%) have exactly 1 isolated node. The structural interpretation is unambiguous: a single boundary vertex cannot close its admissibility stitch under the ε-budget. The global manifold (n−1 vertices) remains fully cohesive. This reframes HARD_FRAGMENTATION from "fragmented structure" to "near-complete structure with one stitching defect." The pattern is identical to the protein MSM out_strength result: 3 isolated states in 5000.

All 500 glass ladders have TD ≥ 0.967. Yet 132 achieve FULL_PERCOLATION. The ΔTD = 0.0007 between percolating (TD=0.9989) and fragmented (TD=0.9996) populations is structurally significant but small in absolute terms. The deciding factor is whether the outlier gap exceeds κ_max · IQR exactly — a sharp threshold, not a gradual transition. This confirms that admissibility stitching is a boundary phenomenon, not a bulk spectral property.

The COVID-19 protein MSM across 5000 conformational states shows 4/5 ladders at FULL_PERCOLATION. The outgoing transition strength ladder identifies 3 kinetic traps: states that are populated but cannot escape via any admissible transition channel. The stitching_strength composite (TD=0.545, κ_connect=2620) is the most informative structural ladder: it reveals the underlying admissibility backbone without outlier distortion. The protein domain aligns with the chemistry finding: global coherence preserved, localized rupture at specific boundary states.

The dual-representation result is now fully supported by chemistry data: raw ladders expose boundary stitching sensitivity (296 single-isolated-node cases), while normalized ladders expose global manifold topology (100% oxide percolation, discrete κ-classes). The two representations are not contradictory — they are complementary probes at different scales of the admissibility manifold. Raw → local structure. Normalized → global structure. Together they constitute a complete dual-scale structural characterization.