B4: local layer support is not a plateau cancellation

If a correctly specified cubic density has a layer correction, that correction is locally supported where derivatives of ${\sigma }_{\epsilon }$ are nonzero. Retarded propagation of a layer-supported source need not be layer-supported, as the explicit XI-B layer diagnostics show XIB.

Consequently a support statement about a local Noether breaking term is not, by itself, a theorem that its completed response vanishes on the plateau. Conversely, compact plateau parameters do not justify discarding a term before its variation and propagation have been calculated.

The stipulated class remains ${\mathfrak{g}}_U={C_c}^{\infty }(U;TM)$. No contact term is selected, no layer term is deleted, and this class is not shrunk. No new B4 variation has been computed because the new vertex has not passed B0.

The nonzero pairing in Proposition prop:tracelayer is not relabeled as a nonzero plateau pairing. It establishes a source-definition mismatch only. Ending Proved obstructed for B4 would require a separate, correctly typed interior witness for the actual completed new response.