Precision Collapse and Observable Disintegration of Deep Ulam Eigenvalues at the B2 Spray Point for WBA and Lorentzian Kernels (Paper 57 in the Non-Holomorphic Fractal Series)

Paper 57 in the Non-Holomorphic Fractal Series. Building on Paper 56, which established conditioning proxies κ_j ~ 10¹⁶–10¹⁸ for deep tiers j ≥ 3 despite spike-dominated right eigenvectors, this paper attacks the folklore of deep eigenvalue stability from two independent axes. Axis 1 (Precision Collapse): for each of the 40 eigenpairs in the precision scan (ε = 0.70, K ∈ {WBA, Lorentzian}, N ∈ {8192, 16384}, j = 1, …, 10), we estimate the Float64 eigenvalue drift δ_j = κ_j · ε_mach and compare it to |λ_j|. On the executed grids, 32 of 40 tiers have predicted drift δ_j > |λ_j|, meaning the deep eigenvalues are larger than one machine-epsilon perturbation ball in Float64 arithmetic. Axis 2 (Observable Disintegration): for the 20 eigenpairs in the left-geometry scan (K ∈ {WBA, Lorentzian}, N = 16384, j = 1, …, 10), all 20 tiers have left-eigenvector support width k90(w_j) at least 100 times larger than k90(v_j), confirming that the observable dual to every deep tier is spatially structureless. Together, deep Ulam eigenvalues at the B2 spray point are neither reliably computable in Float64 nor associated with any observable spatial structure on the executed grids.