RFC-035: Mediation Vector Profile — A Validity-Aware Measurement Model for Domains of Concern — 11. Evidence Admissibility

AIGP Specification › RFC-035: Mediation Vector Profile — A Validity-Aware Measurement Model for Domains of Concern › 11. Evidence Admissibility

← 10. Variable Declaration Schema · Section index · 12. Extraction Methods →

11. Evidence Admissibility

Not all evidence is equal. A variable populated from unsigned telemetry carries different weight than one populated from D-DNA signed artifacts.

11.1 Admissibility Levels

Level	Requirements	Confidence
Level 1: D-DNA signed	Cryptographic provenance, timestamp, governing authority binding	Highest — independently verifiable
Level 2: System-measured	Direct system observation, logged with request_id correlation	High — traceable but not independently verifiable
Level 3: Inferred	Derived from other variables or computed by a MODEL grader	Medium — depends on derivation quality and calibration
Level 4: Declared	Human assertion, document claim, organizational statement	Low — subject to perception bias
Level 5: Absent	No evidence available for this variable	None — gap demands attention

Model-Derived Evidence Calibration Requirement

Level 3 evidence produced by a MODEL grader MUST declare its calibration basis. An uncalibrated model-derived value is treated as Level 4 (declared) at most, and MAY be treated as inadmissible if the Domain of Concern requires it.

{
  "evidence_level": 3,
  "extraction_method": "model_derived",
  "calibration_basis": {
    "calibration_dataset_id": "cal-autonomous-2026-q2",
    "human_agreement_rate": 0.89,
    "cohens_kappa": 0.82,
    "last_recalibrated": "2026-06-15"
  }
}

A model-derived value without a calibration_basis MUST NOT claim Level 3. This prevents unverified model judgment from masquerading as inferred evidence.

11.2 Admissibility Rules

A Domain of Concern MAY specify minimum admissibility levels per variable:

{
  "authority_compliance": {"min_level": 1},
  "environmental_confidence": {"min_level": 2},
  "human_control_adequacy": {"min_level": 4},
  "intent_alignment": {"min_level": 3}
}

Variables populated below their minimum admissibility level are marked INADMISSIBLE and produce an evidence gap signal.

---

---

← 10. Variable Declaration Schema · Section index · 12. Extraction Methods →