We don't originate the claims in this graph — the papers, their citations, their code — we structure and verify what's already public. That's a narrower job than "truth source," and it's the one we hold ourselves to strictly: every edge carries where it came from, every reference implementation carries a verification level, and neither is asserted without the evidence behind it actually existing on the node.
We audit the graph and verify the code as thoroughly as we know how, and we'd rather show you a real gap than hide one — extraction and ranking methods are disclosed on the methodology page, not asserted. Confidence is a recorded number, not a marketing word, and it's checked before anything gets written, not fixed after the fact if someone happens to ask.