Pick the artefact that tests the riskiest assumption. If the risk is "will users understand this", build a prototype. If the risk is "can the model actually do this", build a POC. If you have evidence on both already and need real revenue and real users, you are ready for an MVP.