> If an RM that supports 2PC heuristically completes a branch,

As far as I know, we never heuristically complete a branch. Why do you think
that we do?

Roman