This document addresses some of the less obvious aspects of how pre-shared keys can be used in TLS. A lot of this advice isn't specific to TLS, but it is a helpful document. For someone who might be deploying a protocol that relies on TLS - or might rely on it - the document is a useful resource. My only concern overall, and it is a vague concern, so I don't think action is needed, is that the document could probably use a little trimming. There are some parts of the document that are less useful than other parts. For example, the bit about who has the PSKs is great (one server, one client, don't swap roles); but it is repeated a little across multiple sections. The same applies to a few of the other points. It is probably not worth trying to edit the document down so that each point is made just once, because it isn't that bad, but a shorter document would be more impactful. A specific concern is the somewhat offhand way that early data is treated. The only mention is in a throwaway: "primarily for the purposes of supporting TLS connections with early data" buried in a bullet in Section 6. This is a pretty big topic and having absolutely no mention seems odd. I do think that it needs some treatment in the document. When early data is used with an external PSK, the only additional source of entropy that provides key diversity is the client's random value, which puts a lot of weight on that value containing sufficient entropy. In this case, even if the PSK is good enough, the entropy in the random is significant as it is what ensures traffic key diversity if the PSK is reused. Reusing a PSK for early data also likely leads to poor anti-replay performance if the random is not good enough. I have to apologize to the authors for missing this when it went through the working group. Fresh eyes and all that.