The variation of the operation must also be counted

Thus scalar reordering changes alone cannot cancel a nonzero breaking in the class with a fixed ordinary leading operation and no other image. This is a statement about that declared class, not a verdict about an unevaluated native anomaly.

A zeroth-order covariance replacement $W\mapsto W+S$ is a new choice of free construction. It must be recomputed or transported on a verified domain. It is not automatically a correction of the form $\hslash J(c)$ to a fixed package.