Skip to main content

Missing Proof Detection

Missing proof is represented explicitly in the current output model. It is not supposed to disappear inside a single composite score.

What counts as missing proof​

The example bundle shows several concrete missing-proof signals:

  • required modules missing from the executed set
  • in-scope modules skipped with reason: missing_input
  • unresolved evidence notes in risk_summary.json
  • claim_strength demoted to HYPOTHESIS
  • fast-fail outputs listing missing_proofs and missing_proof_details

How the system flags it​

The sample artifacts surface missing-proof state in multiple places:

  • tier_compliance.missing_required_modules
  • claim_acceptance.global.missing_required_modules
  • phase_5_fastfail_summary.json
  • risk_summary.evidence_notes

In the example run, uncertainty_quantification is missing because of missing_input, and that directly affects whether verdict and safety claims are treated as supported.

How it changes interpretation​

Missing proof changes more than the explanation text. It can change:

  • claim strength
  • recommendation mode
  • evidence status
  • whether a claim is treated as supported or only hypothetical

That is why the sample run can have a concrete risk score and still require MISSING_PROOF_ACTIONS_ONLY.

When a run is unresolved, check missing-proof handling before you react to the headline score. The missing-proof layer often explains why a run that looks informative is still not fully decision-ready.