I believe you.

I'm thinking whatever we decide to do about undesired content, we should create a database of examples (and counterexamples) for testing filters and training models.

Has anyone started such a thing? I know Amethyst has a report feature, are the reports accessible anywhere?