It seems for some of the symmetrized cif files created in materials Project the symmetry cards are wrong.
An example would be mp-773062, where some of the symmetry cards contain odd fractions such as
4 ‘y, x, -z+299/398’
5 ‘x-y, -y, -z+299/398’
6 ‘-x, -x+y, -z+299/398’
10 ‘y+1/3, x+2/3, -z+415/993’
11 ‘x-y+1/3, -y+2/3, -z+415/993’
12 ‘-x+1/3, -x+y+2/3, -z+415/993’
16 ‘y+2/3, x+1/3, -z+73/863’
17 ‘x-y+2/3, -y+1/3, -z+73/863’
18 ‘-x+2/3, -x+y+1/3, -z+73/863’
Another example is mp-756532, where the symmetrized cif file contains cards such as
2 ‘x-z+2/889, -y+1/2, -z+4/889’
4 ‘x-z+2/889, y+1/2, -z+4/889’
If any program reads such a corrupted cif file and utilizes the symmetry card information the results will be nonsense (That is why e.g. these files were identified as outliers in a random ML training set). Some of the odd fractions are close to 0 or 1/2 so they may be rounding errors, but some others are not. Any chance to correct that.