Great! The other kind of plot that might be useful for monitoring convergence is a cumulative class percentage plot, kind of like this one for covid variants:
This allows you to monitor easily not just how many particles are changing classes, but how each class is changing in size as a fraction of the entire dataset (obviously this might get a bit tricky with 80 classes, but one could prob show just the top 10 or 20 and have the rest listed as “other”).
The reason why this (or something like it) is useful is that it captures directionality of change - if a class is changing by a small percentage in each iteration, but is consistently gaining particles, that is a very different situation from a class that is gaining a few particles one iteration, then losing them in the next.