How confident are you in Elon's ability to identify "chaff" within a week? What makes you think the remaining 50% will continue to work at full capacity after this?

One way is to look at a few most recent perf reviews. Another way - look at some internal productivity metric that many companies introduced during Covid/WFH.

Have you ever seen either of those be remotely well correlated with actual job performance?

Yes. Underperformance is a decent indicator. Number of PRs is not a great one, but it can be somewhat correlated with eng velocity/productivity. Twitter will probably end up using a combination of metrics like that.

They will hopefully also talk to EMs, but at this scale might choose not to. It's going to be an ugly process but I doubt it's going to be a random one. They clearly understand that it's in their interest to retain the most skilled and performant employees.

Also, don't want to sound too cynical, but I would guess that it's actually not that hard to evaluate eng performance with all the data companies are collecting during WFH. I am sure Twitter requires using a VPN, so digging into that data would yield quite a revealing and precise performance metric. The only reason this approach is not widespread is ethical, not technological.