I figured out how to describe my complaint with the @osi's OS AI definition: it treats open source and freely licensed as equivalent in all cases. But they're not.
The checklist says you can provide a black box model under a free license as long as you document it. That isn't open source though because I don't have the original source, i.e. the dataset. It fails the second criteria of the OSD, i e. Source code.
As I've said previously, I think there are situations where freely licensed but well documented models might be justifiable. We as a community should have a way to recognize freely licensed models that do have that legitimate limitation. But it's not "open source"
version 0.0.9 Checklist to evaluate machine learning systems This checklist is based on the paper The Model Openness Framework: Promoting Completeness and Openness for Reproducibility, Transparency and Usability in AI…
Open Source Initiative (opensource.org)