AI-assisted evidence-triage VOI

The value_of_ai_assisted_evidence_triage method evaluates whether automated screening or extraction support is worth using when false exclusion, false inclusion, reviewer time, audit sampling, human overrides, extraction error, and model drift affect the decision.

Inputs are deterministic reference labels and triage scores. The result reports sensitivity, specificity, decision-impact-weighted error burden, reviewer time saved, audit value, and the net value against full human review. The method is explicitly human-in-the-loop: audit sampling and override recovery are recorded rather than hidden behind an automation claim.

The current implementation is fixture-backed. A licensed evidence corpus, external model validation, cross-language/Rust parity, and mature/stable promotion review remain required gates.