(Sorry, I posted this earlier but had clicked reply to the wrong post)
I suppose both are doable, though the question for me is less “how can we identify a (LN) anchor spend?” and more “how can we identify users of CPFP carve out?” At the end of the day we’re trying to continue supporting CPFP carve out as long as possible, for applications that don’t explicitly set nVersion=3 right?
My view is that have 2 anchors => need CPFP carve out, so I think checking 2x330 might be cleaner. If there is a non-negligible false positive rate though, we’d probably want to template-match something stricter.