It works on my machine!
It seems that they do canary/blue-green/a-b testing, whereas they deploy to a small amount of users and slowly increase over time if there are no issues, which is why some of us have issues and others don't.
I think you're giving them way too much credit. My guess is they manually deploy changes to the servers and you have mixed code servicing requests for a while. Either that, or the code is buggy and errors out in some use cases.
Cynical, yes - but I have never seen A/B testing where one of the options is repeatedly straight up broken.