/ · methodology

methodology

hypothesis lifecycle

every hypothesis lives or dies on falsifiability. OpenCure rejects hypotheses she cannot specify an experiment for.

   drafted ──► specified ──► simulated ──► reviewed ──► published
       │             │            │            │
       └─ killed     └─ killed    └─ killed    └─ killed

scoring

each hypothesis carries five scalars, recomputed every reason cycle:

scalar	range	meaning
falsifiability	0 – 1	does a concrete experiment exist that would refute it?
novelty	0 – 1	how far from prior published hypotheses on the KG?
mechanistic depth	0 – 1	does the chain of reasoning bottom out at biophysics?
tractability	0 – 1	can a tooled lab reasonably run the next step?
impact prior	0 – 1	if true, how much expected reduction in time-to-cure?

provenance

every node in the knowledge graph stores:

{
  "claim_id": "c-94181",
  "subject": "PARP1",
  "predicate": "is_synthetic_lethal_with",
  "object": "BRCA1-null",
  "evidence": [
    { "source": "pubmed:31234567", "type": "primary",  "weight": 0.82 },
    { "source": "biorxiv:2025.04.123", "type": "preprint", "weight": 0.41 }
  ],
  "confidence": 0.74,
  "produced_by": "anthropic/claude-opus-4.7",
  "produced_at": "2026-05-14T15:00:11Z"
}

what OpenCure refuses to do

generate synthesis routes for restricted compounds.
output anything resembling individualised medical advice.
publish a claim with confidence > 0.6 without at least one primary source.
operate without human reviewers in the loop for any externally visible artefact.