Trust Score
A score from 0 to 100 for how well the published evidence supports a claim. It combines the quality of the studies, whether they agree, and how many there are, using a fixed formula and a published rubric. Every score keeps a record of the inputs that produced it, so it can be re-derived.
The full methodology, version 1.0: Publishing soon
Chain of Custody Coming soon
How an answer is traced back to what was actually read. For each piece of evidence Cytible records the source it came from, the passage relied on, the database that supplied it and its DOI or PMID, with the time it was retrieved. Dossiers and evidence traces are sealed with a digital signature over their contents, so a reader can confirm that a document has not been altered since it was produced.
A methodology document covering what is recorded, what is signed and how to verify a signature yourself is planned for the first quarter of 2027.
Verdict Rubric Coming soon
The four verdicts a claim can receive. Supports: the evidence found backs the claim. Qualifies: it holds only in part, or under conditions. Contradicts: the weight of the evidence is against it. Insufficient: there is not enough verified evidence either way to judge. Supporting and contradicting evidence are always reported together.
A methodology document setting out the exact thresholds between the four verdicts is planned for the first quarter of 2027.
Zero-retention layer
One checkpoint every request to an AI model leaves through. It refuses requests whose purpose is to keep data at the model provider, switches storage off on every request, removes end-user identifiers, and logs metadata only. It governs what Cytible asks a provider to do and what Cytible keeps; what a provider holds on its own systems is set by the terms on the provider account, and the document says so.