This document is written very clearly. I much appreciate the phrase "This is not meant to restrict implementations from structuring racing candidates differently."

In 4.1.1.1, "... send DNS queries for both A (IPv4) and AAAA (IPv6) records ...", this assumes no other address types will exist in the future. RFC8499 uses "address records" to refer to A/AAAA and any future variants implicitly, pointing out that RFC2181 informally says "(A, AAAA, etc)". I'm not aware of a more useful formal reference than 8499, so perhaps something like ".. send DNS queries for address records, which currently means sending A (IPv4) and AAAA (IPv6) queries ..."?

While ietf-taps-interface mentions SRV records briefly, this document does not. An SRV record set could cause a single RemoteSpecifier to map to multiple names. HTTPS and SVCB records can also cause indirection to multiple names (and even transport indications). I do not think the document should go into detail on all of these, but, as with address records, perhaps it can mention that DNS lookups can involve an extra layer of indirection, and provide SRV, HTTPS and SVCB as current examples.

Typo in 4.1.1.2: "intefaces"

In 4.3.2, "should follow the Happy Eyeballs algorithm described in [RFC8305]." - should this be a "SHOULD"? (I noticed a few other lowercase "should"s as well.)

I did not review sections 5, 6, 7 and 8 (other than a cursory check that they do not mention DNS or name resolving).

In 9.1, ".. resolution answers (A and AAAA queries, for example)" - the queries are not cached, (part of) their responses is cached. Perhaps ".. resolution answers (from A and AAAA queries, for example"

No comments on section 10, 11.

Should 12.2 mention DNSSEC? I can very much understand it if you say no :-)