I think the difference is that the pytest and astropy tweaks to the end were entirely standalone and didn't impact other packages (you can do `conda install pytest=x` without having to change numpy). My new conda env changes numpy/astropy/scipy in ways that mean that you can't downgrade without changing everything and since there are packages in the new env that aren't in the old one (they split numpy and matplotlib into two) I think trying to install an old env on top causes real problems trying to resolve exactly which numpy is being used etc. You can't downgrade numpy, for example, without forcing every package to downgrade and clearly the conda solver couldn't handle it.
I think the difference is that the pytest and astropy tweaks to the end were entirely standalone and didn't impact other packages (you can do `conda install pytest=x` without having to change numpy). My new conda env changes numpy/astropy/scipy in ways that mean that you can't downgrade without changing everything and since there are packages in the new env that aren't in the old one (they split numpy and matplotlib into two) I think trying to install an old env on top causes real problems trying to resolve exactly which numpy is being used etc. You can't downgrade numpy, for example, without forcing every package to downgrade and clearly the conda solver couldn't handle it.