Preservation of Effective Judgment: A Falsifiable Research Agenda for Separating Examination, Reconsideration, and Material Intervention in AI-Mediated Decisions

The presence of a human reviewer in an AI-mediated decision does not establish that the person could inspect the relevant evidence, reconsider a preconfigured set of alternatives, or legitimately change the course of action. This is not a novel discovery: research in human factors, meaningful human control, signal detection, and human–AI interaction already identifies relevant components. This paper treats Preservation of Effective Judgment (PEJ) as a bounded, falsifiable research problem rather than a validated psychological construct or a proprietary assessment score. It asks whether a state-sensitive reconstruction that distinguishes (i) available epistemic and procedural opportunities, (ii) observable examination and reconsideration, and (iii) feasible interventions and verified material effects produces more informative claims than established comparator frameworks. The contribution is a proposed comparison logic, a public observational grammar, explicit rival explanations, and designs capable of returning null, redundant, or adverse findings. It does not assert construct originality, report human-subject results, establish legal compliance, or validate a diagnostic instrument. Related behavioral-assurance proposals from Behavieural are credited as independent prior work. A failure to demonstrate incremental explanatory or reconstructive value would justify narrowing PEJ to a documentary synthesis or retiring it as a separate construct. Version 1.0 includes a substantive comparison with Daria Lepikhina (2026), section 3.1 and reference [19]. Conceptual and methodological working paper; not peer reviewed. 11 pages, 19 references. English text with a Spanish abstract.