Veronese Collapse in the Measured Trajectory Span: What the O28 Covariance Does and Does Not Identify

Paper O28 reported effective covariance rank reff=3 in the supplied three-coordinate space Heff. This paper audits what that finite-data result identifies. In an extended recomputation, rank three remains modal but is not invariant: it occurs in $32/50$ stored samples at $q=101$ and $103/105$ at $q=211$. Three spaces must be separated. The measured trajectory span lies in Heff; its leading two-dimensional PCA truncation U⁽²⁾c is data-dependent; and the spin-1/2 module Vρ is supplied elsewhere. No calculation here identifies these objects. The exact Weil-block identity ρq-c=ρc motivates, but does not prove, the additional projected-coordinate relation πq-c(v)=πc(v)̄. Under that relation the outer products are complex symmetric; the stored data satisfy it only approximately. At q∈\29,61\, the trajectory fills the three measured coordinates while its symmetric-square evaluation matrix has numerical rank three rather than six. The fitted constraints are numerically traceless; their commutators have numerical span three in the antisymmetric basis, and the computed commutant is one-dimensional. These finite-precision diagnostics are compatible with the supplied adjoint reading of Heff, but they do not derive that carrier or prove exact irreducibility. The tested conjugate-pair samples do not attain the O26 target rank four. Breaking the per-vector lock restores rank four only in End(U⁽²⁾c), not in End(Vρ). Interpretive outlook. The collapse diagnoses geometry within measured coordinates; it does not select a physical representation. A genuine spin-1/2 test still requires an independent observable and a typed bridge to Vρ.