Observing for last 5-6 months how protection works (or more precisely _fails to work_) leaves me quite disappointed. Unless stats show that my observations somehow miss the big picture, I would say that it is probably broken, at least at the sites I am active at (Programmers and Workplace).

As an example, take a look at [recent question at TWP](http://workplace.stackexchange.com/q/35519/168). It has quickly got three troublesome answers ([1](http://workplace.stackexchange.com/a/35583/168), [2](http://workplace.stackexchange.com/a/35608/168), [3](http://workplace.stackexchange.com/a/35565/168)) - all from new users having less than 10 on-site reputation, all with negative score (one is even already deleted).

Still, protection didn't kick in - simply because when third "answer" was posted, one of two prior ones has picked a random upvote that pushed answerer's reputation to 11 (soon it has fallen back below, because downvotes to that answer continued coming in, but it was too late). It's not the first time I see this mind you, only maybe the most clear cut one.

You know, protection criteria depends on reputation, which is driven by voting. Voting,  in turn, is [somewhat random](http://meta.stackexchange.com/a/215397/165773 "where are Tim's keys?")... and that's fine! Voting may be weird sometimes, but _in the long run, on average, in typical cases_, it tends to be reasonable - that's the way how whole system works and so far it looks sufficiently reliable, despite infrequent _oscillation_.

If protection criteria was as reliable as the rest of the system, I would be fine with that. Thing is though, it is (to a large part) based on voting in questions intended to be protected - in _popular_ ones, and voting in these often is _anomalous_... softly speaking. As far as I can see, that makes it way too fragile.