If one removes files while the clustering is running these files can show up duplicated and the file count can get confused.

E.g. when I was clustering 4k files and immediately after hitting the cluster action I deleted all files, the files showed up in the clusters, but file count was down to ~400. Moving the files back to unclustered the file count went up to the original 4k, but files also remained in the clusters.

After the clustering algorithm has run building the actual clusters must ignore deleted files.