This draft describes an improved variant of the NTP client state machine that can more reliably reject servers that are hostile or are under attack.  It is effectively a summary of a more detailed research paper.  Overall, the proposal appears reasonable, and is presented clearly.  However, I do have two concerns to note:

1. The document's status is "Informational".  The text is largely a summary of a more detailed academic research paper.  The proposal has been implemented, but seemingly only in an academic demonstration codebase.  If the Khronos behavior has not yet been implemented in a widely used NTP client codebase, I think the "Experimental" status would likely be more appropriate.

2. The document claims to defend against MITM attackers, but it also notes that the defense only applies to attackers that can interfere with some fraction of NTP server access.  The security section should be expanded to note explicitly some attackers who are out of scope.  One such attacker appears to be the "nearby MITM", who can selectively block any of the client's traffic.