I generally create python environments with conda, then populate them with pip or uv. But conda hardlinks files to save space, so I wanted to try to get packages mostly from conda, then top it off with pip.
Turns out it's a mess.
The messiest part of the mess so far: I had to binary search, doing dry installs with increasingly small requirements, until I could pinpoint the one requirement that caused a package to install, but didn't have that package in its recorded dependencies ... and installing that hidden dependency failed when doing it through conda.
It's not just "is that package of that version in conda-forge", it's also "is it in our conda-forge mirror".
In theory these two questions have the same answer.