No, that's a lie. Let's rephrase - everything (apparently) works other than what I'm interested in.
Which is ... wait for it ... finding the medial-axis of an image at a single pass from the inverted Canny edgemap by computing the local maxima of its distance transform.
Head. Meet wall.
I usually never refer to work - except as an euphemism for disaster - on this blog. This diversion may be mostly fueled by some random hope that ranting about it here might help. Somehow.