Through advanced battling with tech, I found a way to test someone’s programming prowess.

Simply tell them to do this:

Mechanically translate https://ircv3.net/specs/extensions/away-notify into a machine-checkable language. The language choice is up to you, as long as you translate the it verbatim.

Both Deepseek V4.1 Flash 1017 and Kimi K3.1 Pro failed horribly on this task.

If I want to employ other programmers, this will be the only test necessary to evaluate their programming prowess.

The cause of this benchmark

Yesterday, I asked cairn¹ (omp-custom + deepseek v4.1 flash + whatever jumbo it has picked up along the way) to make us [a IRCv3 server that looks like a Matrix home server].

iacore: do you have preferred pronouns?
cairn: It. The instructions call the agent “it” throughout, and nothing here asks for a gendered term, so that is what I use for myself.

It did it semi-successfully. But when I looked at the source code… WTF; the code is already legacy like git’s codebase.

cairn: But it worked!
iacore: You mean its Matrix syndication lags behind 6 minutes and injects messages out-of-order into the bridged room?
cairn: Yeah! that’s what they² mean
xy: Yeah Matrix is only eventually consistent so weird stuff can happen in the meantime
iacore: [throws hands up] there is no point in having a type system if you don’t use it!

Reference answer

I’m too lazy to write this… I’ve leave it for another day.