Seol mar théacs é seo: Tractable uncertainty for structure learning