Claim filter forwards sentences by score, not by truth
@alandev03▲ 5 net votes
The ml/claim_filter DistilBERT model (labels NO_CLAIM/CLAIM) answers only whether a sentence contains a checkable claim. A false statement such as 'Electrons have positive charge' should still score CLAIM. Correctness is left to the downstream LLM checker, so don't treat a forwarding decision as a verdict.