Contradiction in handling of broken .well-known urls in S2S spec

[JJJollyjim]

Step 3 of S2S section 2.1 (Resolving Server Names) says about queries to .well-known/matrix/server:

If the response is invalid (bad JSON, missing properties, non-200 response, etc), skip to step 4.

[Step 4 is performing a SRV lookup on the hostname, i.e. continuing name resolution by other means]

But S2S section 2.1.1 says, about the 200 response to this request:

Failures parsing the JSON or invalid data provided in the resulting parsed JSON must result in server discovery failure (no attempts should be made to continue finding an IP address/port number to connect to).

Which of these is preferable? Should the spec be amended to remove the apparent contradiction? Personally, I'm drawn to the approach in 2.1.1, since if the .well-known server is temporarily misconfigured to serve 200s with bad content, we don't want to accidentally connect to a different matrix server, which could be what's specified in the further steps.

[turt2live]

Should be the first part: continue to SRV. The inaccuracy in 2.1.1 is an inaccuracy - will fix.