in general, training data is considered part of the source code, but ISTM that using mechanically-curated training data would make it object code, whose source code would be the pre-curation data, and the curation tools would be build scripts and tools of sorts.