Typed Indeterminacy Engineering: A Practitioner's Design Pattern for Carrying Uncertainty Types Through a Decision Pipeline

Abstract

Decision systems that operate under uncertainty — LLM-as-judge evaluators, multi-criteria ranking procedures, evidence-fusion pipelines, AI alignment mechanisms, expert-consensus methods — routinely reduce heterogeneous epistemic failure modes to a single scalar of “confidence.” Several established literatures already resist this collapse in pieces: Dempster-Shafer theory separates ignorance from conflicting evidence; the aleatoric/epistemic uncertainty distinction separates irreducible from reducible uncertainty; neutrosophic set theory keeps truth, indeterminacy and falsity as three uncollapsed components; imprecise probability represents indeterminacy as a credal set rather than a point; AGM and Katsuno-Mendelzon belief-change theory distinguish how a belief state should change depending on why it changes; selective prediction, conformal prediction, and learning-to-defer already give a system several ways to act when it should not commit to a point answer. None of these literatures is new, and none of the individual distinctions below is claimed as new. This note states a design pattern connecting them: index (a) a 2×3 type grid — failure mode (ignorance, conflict, falsity) × removability (removable, constitutive) — to (b) which belief-change operation applies to each of the six cells (expansion, question reformulation, update, belief merging, revision, or source discounting), and in turn to (c) which decision-output form applies (a tagged point decision, a set-valued decision, or an escalation). We call this pattern Typed Indeterminacy Engineering (TIE). We report two worked examples from the author's own applied practice — disclosed here as published in a journal family the author edits, hence offered as illustration and not as independent evidence — plus a worked application to Delphi-style consensus methods, before stating plainly what remains conceptual and what an independent research program would need to formalize and test.