So should they hand out promotions and fire people based on randomness?

Take the manager's word for it and go all in on nepotism?

What's the better solution?

Holy strawman batman.

Note: 'in a standard way across teams'. You have jumped directly to 'can't be measured at all in any circumstances'

Difficulty of project/task seemingly is a big factor. The engineer working in a legacy code base is likely to be less productive due to technical constraints.

How to grade that into any kind of ranking, in a standard way, is seemingly not likely given how subjective the difficulty of a code base is. This also completely elides the clever and best engineers that find ways to simplify and/or avoid solving problems entirely. Further, a team that built a crap pile of bad code, but know it pretty well could seem really productive just because they have built for themselves some job security. Then compound all this with measuring the utility of what was produced.

Honestly if they are targeting percentages of the company, ditching entire products, divisions or slicing diagonally across the org chart is the way to go. Consolidation of like functions and just stop projects. If there are uniquely skilled people in the layoff, give them a chance to interview with other teams. Make it quick and as painless as possible, the longer it drags out with rankings and middle managers politicking the worse morale will be.

Also, any investors out there might hear “one time restructuring charge” in the quarterly results when they talk P&L which will affect the bottom line.

> Take the manager's word for it and go all in on nepotism?

Geez, I mean there are decent managers who do rate on performance and not just buddy-buddy stuff. Frankly I'd expect to be trusted with my reviews of people under me.

Less scaling, smaller focus? I kid, but there's not an actual correct answer to any of this. Nobody likes nepotism except the family that it serves.

This may not be surprising at all actually. Look at how they form their feeds. It is hyper-dependent on a singular metric: engagement. Engagement doesn't mean you're making your users more connected or happier, it just means they are using your product more (e.g. fighting and yelling at one another). Goodhart's Law in action. But I agree that something like human performance is incredibly difficult to quantify and requires a lot of nuance. But people, and especially bureaucrats, don't like nuance.