Inconsistent FoldMason alignments with near-identical AF2 inputs

[RumoisteinKartenspiel]

Coming back to the issue of getting slightly different alignments, I found this surprisingly strong input dependency.
It seems that when using input pbs that differ only slightly, (two different AF2 predictions of the same protein), the foldmason alignment looks surprisingly different for some residues. The two proteins show only minimal differences as calculated RMSD, and I just stumbled upon this while calculating per-residue-based replacement scores and could not immediately replicate an older experiment.

Match: assigning 496 x 496 pairwise scores.
 MatchAlign: aligning residues (496 vs 496)...
 MatchAlign: score 2640.000
 ExecutiveAlign: 3923 atoms aligned.
 Executive: RMSD =    0.365 (3923 to 3923 atoms)

Here for example residue F199 aligns in very different columns.