Really interesting approach! I'm curious about the validation methodology - how do you handle cases where AI-generated citations are semantically correct but use slightly different phrasing than the original source? For instance, if a paper paraphrases a finding rather than directly quoting it, does Veru flag it as potentially hallucinated?

Also, given OpenAlex's coverage, have you noticed any domain-specific blind spots where the tool struggles (e.g., newer preprints, non-English papers, or highly specialized fields)?