Is the local median in any way similar to Symmetric Nearest Neighbor Median? (I did experiments with SNN averaging here, but the original paper mentions a median implementation) https://observablehq.com/@jobleonard/symmetric-nearest-neighbour-all-the-things

Ah, no it looks like it's distinct (although the SNNM filter already is local to the radius it uses as an input)

Furthermore what I'm doing here is not strictly the local median: it's a posterized average of posterized values. This would be identical to the median (or mode, for that matter…) if the average was uniform over a neighborhood. However, since I took a shortcut with using array-blur, which does a weighted average, the result might be a bit different.

typo? isn't 'traditional posterization' shown in the 'bottom left'?

yes, thanks!