Reintegrating future changes in the upstream is also made nearly impossible as a result; our tech lead made a change to numpy a few years ago, didn't manage to get it accepted by the project, and we're stuck with this version until the sun burns out.
If there are changes in future numpy versions we want, it's up to us to backport them, which is nowhere near our core business.
There's a lot to be said for standardization and 'boring.'
Well you could estimate the impact of backing out of the changes on the application side, with the upside being continued savings in operational complexity. Or, you could address the operational complexity with processes or tooling - which would be easiest with something like a developer OS image.
If there are changes in future numpy versions we want, it's up to us to backport them, which is nowhere near our core business.
There's a lot to be said for standardization and 'boring.'