Verifiable, Articulable, and Tacit Components of Preference

The CreativePreferences dataset provides 2.8M texts labeled by 317M human judgments across seven creative domains to address gaps in tacit preference modeling. Using executable programs and rubric banks, the framework quantifies the articulability and verifiability gaps between dense models and articulated constitutions.