>The only thing is, you can't really publicly admit exactly what dataset you trained/tested on...?

Curious why not? Assuming you only used the metadata. I think they would be considered raw facts and not copyrightable.