That's a tough one: most standards are codified as such because they're already seeing wide use. The major example of one that's been worked the other way around is IPv6: it's been a standard for a very long time, and still doesn't seem to be seeing adoption.
Of course, I wouldn't say I was bullish on IPv6. 32 bits is enough for anyone, right.
The incident you mention is probably the most impactful, but there's also the time the Russian military blocked IPs outside Russia by returning 418 instead of the more logical 403.
This is actually a good use of 418 in production, and one I've come across before: if you need to perform some custom handling and throwing a HTTP error is the only sensible way to do it, 418 is always available.
Unless your server really is a coffeepot, which is ...unlikely.
There are joke RFCs almost every year, so it's not unprecedented to add to the standards. This year, one of the additions was a Death Flag to TCP, to indicate when a connection is about to terminate. The RFC Editors are very approachable when it comes to the Apr 1st RFCs: a "real" standard would need to be drafted by someone actually in the field, but the Apr 1st's are open to public submissions as long as you're willing to redraft/edit in accordance with the documentation standards.
It's worth noting that the Clacks header is an unofficial campaign, and hasn't been standardised; the 'Pedia states that some 84,000 sites return X-Clacks-Overhead, and my own is one.
Internet Relay Chat's been one of those things that's always felt out on the edge. I've been on EFnet since perhaps '03, and it's a lot quieter than it was...
With people moving en masse away from the centralized sites and their Firebase-implemented chats, we may see a pick up in traffic on the IRC networks, which would be good to see.
It's great: the Internet should have a bit of that sense of whimsy, and knowing that there's official support in many libraries for "you're asking me for coffee, but I'm a teapot" is one of those things that gets me through the day.
If you're writing a TEA-compliant client, you'd send the BREW request and expect a 300 Multiple Options back, whereby the server will tell you which teabags are installed. You're correct that there'll be no error, unless all the bag stocks are out server-side.
I think it's excellent out here. I was stuck on Reddit for the longest time, and this recent debacle has pushed me to explore the networks at the edge; this feels a lot more like the Internet of old. The analogy of email is apt, I think, with the accounts on multiple servers and the interplay between.
That's actually the topic of the talk! Around 1995-96, HTTP was picking up all kinds of use outside the academic community, and people were tacking extensions on left and right; one of the biggest was file upload support, which was done by throwing HTTP and email into a room and having them fight it out. Which is how we ended up with the monstrosity that is "sending emails over HTTP", also known as "posting a form".
The author of HTCPCP decided to codify some of his concerns with these, partly as a joke; I noticed long afterward that his joke was only standardized for coffee, which Personally Offended me as a citizen of a tea-drinking nation.
I've just done some quick browsing to see if there's a written-down motivation for Referer existing, and there's this on the Wikipedia: "Many blogs publish referrer information in order to link back to people who are linking to them, and hence broaden the conversation."
Which I guess makes sense, in the context of the original use of HTTP as an academic publishing protocol, but it's gained cruft and nefariousness since wider adoption came about.
There are good arguments for stripping Referer from the standard, and yours is one of the most cogent; if Referer is still a thing in another 30 years, I'd be surprised.
I always rather enjoyed the double entendre of "420 Enhance Your Calm", which was an unofficial response from Twitter's original API before "429 Too Many Requests" was standardized.
But I can't think of any codes which aren't already in there, that I'd use; there are a bunch that don't see much use, like "410 Gone", so the list could do with trimming down if anything.
That's a tough one: most standards are codified as such because they're already seeing wide use. The major example of one that's been worked the other way around is IPv6: it's been a standard for a very long time, and still doesn't seem to be seeing adoption.
Of course, I wouldn't say I was bullish on IPv6. 32 bits is enough for anyone, right.