Earlier quoted context omitted.
Facebook, however, has a massive codebase. Their app is easily one of the biggest Android apps, in fact it's so huge they repeatedly (?) had to hack the Android runtime environment to allow it to run. Their backend and frontend code is likely also a huge monster. In addition they're doing ML/AI stuff which is even harder to debug.
You’re implying their ML/AI interefered with Mark’/ dislike button? Bit of a stretch.
I don't imply it, but given how many spam filters these days are realized as more-or-less-black boxes, it is a plausible theory.
In addition, at least for Twitter and FB, there has been suspicion for long time that reports/blocks have faster consequences when many people report/block, as automated filters take over before a human can look at it. I can imagine that there is a filter that gets triggered when a user gets N reports/blocks in M minutes, with e.g. N=100 and M=30, to automatically flag users as potential spammers, unsolicited nude senders or whatever. This had hit Zuckerbergs account, the spam filter blocked his account... and then someone was tasked with preventing this, by disabling the block function based upon certain criteria.
Also, the bug seems to have affected only users blocked by thousands of people. Not exactly your usual testcase, and I don't know if I'd have thought of this scenario when writing the test specifications.