"Accuracy" is an interesting word. The waveforms your are working aren't actually square waves - with zero edge transition - but have more or less exponential rising and falling edges.
You don't cares if the rising an falling edges have much the same stable shape, and if each exponential has effectively got to the rail before the next one starts.
Different rise and fall times complicate the error analysis, which is a tedious exercise (and one that I've done from time to time). One of engineers I worked with - and interviewed when he was first hired - did a six-digit DVM based on delta-sigma principles and he talked me through some of the problems during the job interview. I'd met them before on less accurate systems, which meant that he impressed the hell out of me.
He did some very nice work after we hired him - some of it on one of my projects.