[nugget]

I think it's possible, but usually non-cost-effective because it has to be done on a 1:1 basis and therefore doesn't scale well. For example: I can intensively audit a developer's code for a couple weeks (in real time) and usually get a good feel for the blend of skill/time/effort being invested. But it's so time intensive for me to do that, that it's reserved for extreme situations.

[nradov]

Writing code is only a small part of a developer's job. If you focus on that then you'll miss a lot of other key productivity factors, such as contributions to team design discussions, code reviews, defect root cause analysis, etc.

So in short no, it's not possible to accurately measure developer productivity even if cost isn't a factor. Employee evaluations are necessarily subjective and we simply have to accept inaccuracy.