This can cause inflation of the repeat number. Example: 2q4g.W has 8 models with 16 repeats each. Running with 8G of memory I can finish this structure, but get 128 repeats!

If I remember correctly the input to the CESymm method is actually the Atom array extracted from the Structure object.
I wonder if this could be a bigger problem related to how the representative Atoms of multi model structures are obtained.

@sbliven since 5.0.0, BioJava will by default produce bioassemblies by expanding into new chains (instead of new models). That is the default now if you do:

StructureIO.getBiologicalAssembly(String pdbId);

or otherwise you can explicitly set the flag (multiMode=true for expanding into models instead of chains):

StructureIO.getBiologicalAssembly(String pdbId, int bioAssemblyId, boolean multiModel);

I think the feature is also available in 4.2.x, but there it is not the default.

So it should be totally safe to remove the concatenation of models in CESymm.

I need to look into this more, since I found a few cases where CE-Symm seems to be duplicating structures. We used to report the currect number of repeats for these, but now it's doubled. None of them have multiple models.