We would like to implement an invariant synthesis algorithm that is as strong as the h^2 analysis. But we generally don't keep wishlist items open unless someone plans to work on it in the (reasonably) near future unless there is something important in the discussion that we need to remember. So marking this as "resolved" as a won't-fix, but we'd be happy to eventually implement better invariant synthesis algorithms if someone steps forward to do the work.

Vidal Alcazar writes:

I was using the mutex groups generated by FD's translator but in some domains
not all the binary mutex relationships are found (Trucks from IPC-06 is
particularly problematic). Can I trust the h^m heuristic in the current version?
Is there any other way to compute the mutexes? Thanks!