I apologise for the late reply.
I have been testing further with an input deck that relies on none of my custom computes or fixes that I have used previously. Turns out without them there seems to be no issue with fix deform variable.
The issue seems to only occur with a variable that relies on a custom compute that I use. The compute itself shouldn’t produce any unsuitable numbers (assuming I haven’t made a mistake).
The compute simply replaces cluster IDs with -1 in the cluster/atom ClusterID array if the ID exists more then once in the atom_vector and then use ‘variable atom’ to mask the output atom_vector for values >= 0. I then reduce it using reduce sum to obtain the number of clusterIDs that exist only once in that atom_array. I want to use this scalar to define the change in box dimension.
Obviously the issue is somewhere in my own work and not with LAMMPs and hence I apologise for wasting your time!