Well, no protocol ever truly is completely finished, but it makes sense to document wider consensus, mark mature docs as "good-enough to finalize", and then not expect any broader application of them, until they reach the next "good-enough to finalize" stage.

There will always be some bleeding-edge implementations or experimental docs, but I'm not going to chase down every hare-brained scheme some developer came up with (regardless of how popular his client is), and solidify it in the SDK.

That's not what an SDK is for.