So, a bit like these velocity points in Pivotal Tracker, only... even more detached from reality?

I'll let you in on why measuring mostly anything related to code in weeks is wrong: It promotes short-term thinking and code debt, not unlike what we have seen in the banking blowup.

Yes, there are classes of tasks that can be solved in a week-timeframe. But those are not normally the ones that make or break your project.

For a small, contrived example: You can hire a low-skilled worker to provision new servers. You can then measure how many servers that person gets done per week.

Or you can hire a high-skilled worker to build infrastructure for provisioning servers with a mouseclick. This will take months. But once finished you have saved yourself an ongoing full-time salary and can assign the high-skilled worker to other tasks.

Good luck determining that high-skilled workers "weekly velocity", though...