So here's a small selection of patterned dithers that do not contribute to banding. Easing from one to another is bound to prodouce some, but them the breaks.
Hmm, the middle patterns of each row appear to have visible rows, because they have no diagonal quality. The bottom left and top right and left are good. I'd hesitate to use the middle patterns instead of 50% though because it would introduce lines, or... bands... into my transition.
I'm wouldn't be too picky about it in my personal work though, and all are good patterns which I might use, I just don't think visible bands are any better than the super dense (to the point of negligible) banding that is 50% dither.
I think a cluster needs to have banding across a visible length to be considered banding. In the case of 50% dither the fact that the corners are touching doesn't make any one cluster look any more aligned than another, because the clusters are all single pixels and are as offset as possible at their density. Introducing gaps into the mix to try to separate the corners just introduces clusters and potential rows or columns which are worse bands than any 50% could ever be.
Granted, you're talking about the
corners in B, but if that middle line was only a pixel long it would just connect the two lines in the most efficient way possible, with a single pixel and I don't think that counts as banding in any sense I understand it. If the middle line was 2 pixels long then I'd say it was banding, but 50% isnt double pixels, it's as meshed and integrated as you can get. If it was 50% dither with doublewide pixels, yeah, that's banding. But not with squares.
Interesting stuff though and I kinda rambled without a solid point, but it's the ramblethread, right? Not discouraging the notions you're bringing up, just offering my own reaction to them. Whether I disagree with them or not is irrelevant, exploring pixels like this is EXCITING. :'D