# Evidence Sufficiency **Entity class:** Epistemic decision criterion **Domain:** Evidence / artificial intelligence / decision making **Maturity:** Developed ## Definition **Evidence sufficiency** is the threshold at which available observations and sources adequately discriminate among competing claims or causal explanations for a decision or conclusion. ## AI Mechanism An AI system evaluating sufficiency must search for contradiction, distinguish direct support from mere consistency, compare alternatives and abstain when apparently answer-bearing material does not justify choosing among them. ## Article Context [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] identifies over-answering under conflicting evidence as a failure to recognize insufficiency and places calibrated abstention inside the proposed Universal Epistemic Engine. ## Evidence Boundary Sufficiency is claim- and consequence-dependent. Evidence adequate for a provisional low-stakes explanation may remain inadequate for diagnosis, accusation, coercion or consequential causal attribution. ## Relationships - [[wiki/Claim-Level Verification|Claim-Level Verification]] - [[wiki/Epistemic Uncertainty|Epistemic Uncertainty]] - [[wiki/Evidence Weight|Evidence Weight]] - [[wiki/Evidence Boundary|Evidence Boundary]] - [[wiki/Model Calibration|Model Calibration]] - [[wiki/Epistemic Vigilance|Epistemic Vigilance]] ## Sources / Provenance - [[articles/Epistemic Invariance in Conversational AI|Epistemic Invariance in Conversational AI]] — AI application, contradiction handling and abstention research.