> the explanations that don't end up producing an image all seemed to skip over one detail or another.

Implementation can be wonderfully useful as both a test of, and a forcing function for, really understanding something. At least when ground-truth (ie tests) is available.