[virgilp]

I'm not saying "people can't be managed" and "one can't evaluate the performance of a programmer" - that'd be obviously false.

What I _am_ saying is that it's an art not a science - very hard to teach, impossible to scale.

Take your example with "Documentation" - if you measure anything objective (words written, number of functions/features that have associated documentation) and tie it with the performance evaluation, those metrics will soon become meaningless.

[athenot]

Poor metrics are meaningless regardless of what you try to measure.

In the example of documentation a better metric might be a composite value of

(i) how many wiki articles a developer has written, weighed 0.33 and

(ii) how useful on a numerical scale his/her peers rate the documentation produced, weighed 0.67.

This is just an example but it's a composite of perceptual and objective, with the perceptual being a lot more important (and to prevent gaming the system by writting lots of useless articles).

[tstrimple]

The way I've experienced it is that when a measurement becomes a metric, it ceases to be a good measurement. When people's bonus is tied to a measurement, they will find creative ways to influence that measurement which, in many cases, defeats the purpose of what it was trying to measure in the first place. It's not simply a matter of choosing the right measurements. It's also a matter of how you incentivize those measurements.

When I worked at Microsoft one of the teams I was adjacent to had a metric on number of new apps in the Windows Phone app store. So the teams went out and got a bunch of college students to build shitty apps in bootcamp style working groups. Suddenly the number of apps isn't good enough, so they added review metrics. Now those teams add a "let's all rate each other's apps" portion to the bootcamp taking you even further away from the results you're trying to obtain.

[virgilp]

It's all in the details. If I get to pick who rates "how useful", this translates into a general "how well liked by my colleagues am I" (general problem with 360* feedback; I'm not saying it's useless, but it does have its limitations)

[brlewis]

If it's true that one can evaluate the performance of a programmer, what does that mean? Can it be reduced to a vector of scalar values?

I hate doing performance reviews, and I especially didn't like when at a former employer I was asked to stack rank programmers. I feel like it's a task that's so difficult to get right, I can't even think of a wrong way to do it that's useful.

[virgilp]

> what does that mean? Can it be reduced to a vector of scalar values?

No, it certainly doesn't mean that "your performance was [7.23541 8.1241 34.412515 .52632 995.154]". It means "you did good/ you did very good/ you need improvement", with some details like "<these things> you really handled well and I appreciate it; maybe you can work a bit on <this area> though". I definitely agree one shouldn't stack-rank people - especially in a good team (it's entirely possible, desirable even, that everybody did a great job)