About the build fail:

Please help me make sense of this: in test_rules_are_applied_deterministically() it appears that rules should be applied several times, as the test checks to compare against outlines of "Q.ss01.alt".

I added a warning to the main code that prints:
WARNING root:instantiator.py:174 Duplicate rule for ('Q', 'Q.ss01') in location {'Weight': 900.0, 'Style': 0.5}: ('Q', 'Q.alt'), meaning that ('Q', 'Q.alt') was already applied (= in the list), and now ('Q', 'Q.ss01') will be rejected for duplication.

If we're on the correct trajectory here, I think I should indeed check for the whole tuple (oldName, newName) to already exist, not just for oldName.

@belluzj added that test in #689 when he fixed some bug about processing rules, maybe he could also take a quick look?

Well, the test makes sense as far as the UFO test data goes. (The .ss01 swap is a bit unfortunate of an example because a stylistic set is typically implemented as a GSUB substitution, but okay, the UFO is clear on this.)

The commit I just posted (checks for whole tuple like in the beginning) fixes the test. But I think we want to go over this one more time and see whether this is truly the correct approach and maybe why the duplicate swap exists there in the first place.

Adding Simon as a reviewer.