[satyrnein]

This is the stuff technical debt is made of.

True, but also ignoring Chesterton's Fence is what catastrophic rewrites are made of.

If you know why the fence is there and have confidence the reasons no longer apply, you can be bold. If you're not sure because the archeology is expensive, you should take baby steps if possible. (Which I think you get at with the cost of regressions.)

For my team, that's often flipping a feature flag where we don't expect any difference, and watching the output for a while to verify. First sign of surprise, we can quickly flip back. We get surprised more than we would like.