[snarfed]as a data point, the experience we've seen in the wild here with endpoint discovery is that everyone starts with just extracting them from HTML. later, only a subset of those people eventually extract them from headers too, and many get tripped up on parsing the Link: header
