https://gerrit.wikimedia.org/r/#/c/196532/ proposes changing output of images in PHP parser to use <figure>, thus moving parser output closer to Parsoid's semantic markup.
But, based on discussion, it seemed better to get word about this: why this is a good thing to do, what are the implications on tools and scripts out there, how to fix them, etc.
Please share a draft (link to etherpad / google docs, whatever) for others to comment on.