Ideas

state variables

I think you were right about not using "state" variables. It's redundant and unnecessary. I'm changing all of the assembly 'step' functions to have a "force" argument, and will simply check whether a sample has finished a step or not by checking whether it has a finished product for that step yet, e.g., querying the pandas data frame sample.files.isnull().clusters can check whether a sample has a cluster file yet.

...