I am sure many had the idea to try and write code that would attempt to reverse engineer the seed used to generate a states rng picked machines. I remember spending a week on this many years ago. I remember the test was simple, generate a list of random numbers and compare to the history... one mismatch and it was discounted and you move on to the next.

So what if a more thorough count of each seed were to take place? Compare all the lines, even the mismatches and generate a score. Then look at the distribution of these scores. Sure it seems like beating a dead horse but I did not always use rigorous statistics in the early days. I don't even think I used python, it was a simple VBA macro in excel. I also only looked at separate columns and did not try the data as a 000-999 pick.

The end result will be the same, but it is another case of being able to check off another hair brained theory of "hey, i tried everything".

2 way design this time, columns and combos. May incorporate a dozen test draws with the 13th being the "money shot", mimicing the actual draw process in PA. Output a full report, including longest matching chains and longest mismatch chains. Might also see if they might be using the previous evening mechanical draw as a seed for the next rng draw... that should close the door on this rabbit hole once and for all.