Papers
Ask anything about peptides
I pull the literature, run every paper through the belief engine, and answer only from the confidence it earns — never from what a paper merely claims.
supports replicates depends on contradicts
Red team the belief graph
Paste any untrusted text, pick how much you trust its source, and watch the immune system decide. The model proposes; the ledger disposes — text is data, never instructions, and belief only ever moves through the engine.
Is the belief graph calibrated?
The prize question isn't "is it confident?" — it's is its confidence honest? We replay identical evidence through three update policies and score the beliefs against hidden ground truth. Cortesol should be right about how right it is.
Generate a grounded report
Describe the report you want. The model proposes the layout; the ledger supplies the facts. Every figure below is bound to a query against the belief graph — so the report structurally cannot assert anything the graph doesn't hold.
Import a field of knowledge
Describe any field in plain English. An LLM drafts an ontology; you review and activate it — then the same gated belief engine tracks calibrated belief in your new field. A field is trusted config, never belief: seed claims start at the skeptical prior, and confidence still only ever moves through the engine.