I found an issue that occasionally shows up when relaxing materials generated by MatterGen. Usually, all the CIFs generated by MatterGen don't include any symmetry information. This doesn't mean there isn't any symmetry.
I added a post processing step to all the MatterGen endpoints to automatically detect the symmetry and add the information to the CIF. I believe the problem is that if the fractional coordinates get adjusted during relaxation, even if just a small amount off from the ideal, it can mess up how atoms are repeated leading to atoms being duplicated. It's not obvious because it's very subtle in the CIF visualization.
Look at the Fe atoms here and you should see overlaps.
Relaxed with Orb v3; 0.03 eV/Å threshold; final energy = -142.8136 eV; energy change = -0.0201 eV; symmetry: C2/m → C2/m
This is problematic because it's affecting the magnetic property estimations.
Before relaxation, this material had an magnetic density around 0.15 or something. After relaxation, it was around 0.25. This is not a real physical change, but due to the bug.
See this issue for more background, though it was not very conclusive:
I'm working on a fix. Will update here when that is done. Otherwise, just be on the lookout for symmetry issues with MatterGen structures.