J5: support control is not output admission

The inherited compact graph closure and properness remain support inputs for a support-preserving construction. They do not turn an unnamed composition into a distribution.

The output target is a compact degree-two kernel with

under the fixed flat cotangent identifications. This is a sufficient microcausal bound: a nonzero tuple lying entirely in one pointed closed causal cone cannot have zero covector sum.

Assigning balanced covectors separately to edges is not the output proof. One needs a defined joint extension, its full graph wavefront theorem, and its actual proper pushforward. Since J2 is unresolved, none is supplied. No admission assertion for ${R_1}^{(1)}$ is made. J5 is not opened.