That's a lovely script. Really quite mysterious. Reminds me of graph theory and topology.

I wonder if you can do some kind of weird frequency distribution analysis to classify what the extent specimens may be referring to.

I mean given some old civilization, is there a guassian like distribution of say X% documents being financial, Y% religious and so on?

And if there is, can you somehow classify these samples into camps and then maybe get to some confidence interval that one pile is likely say, political statements such as laws.

If you can get say a commerce/accounting pile, your known plaintext is pretty constrained. There's probably some bookkeeping and itemization and maybe some mathematics. Especially if there's addition or subtraction on there. It may not look anything like how we do it: could be concentric circles or something wild but arithmetic is generally the same. X + Y >= (X or Y) - even in systems where they just use a term "many", some really basic rules will still hold

For the religion you have feasts, calendars, holy days, etc. Maybe there's domain specific characters given this approach