Conflict or uncertainty? Reading the partial paraconsistency hierarchy of interval (T, I, N, U₁, …, Uₙ, F) tuples from language-model outputs

Abstract

Genuine conflict (truth and falsity both strongly supported) and plain uncertainty (weak support either way) are different epistemic situations, yet a single probability of truth reports both as the same kind of number. This feasibility study populates the partial paraconsistency hierarchy of interval (T, I, N, U₁, …, Uₙ, F) tuples from repeated outputs of gpt-4o-mini, deepseek-chat and the typed evaluation model Jev. All three systems declared the liar paradox Order-0, while intermediate orders were not read reliably. The study reports the elicitation protocol, a corrected scoring error, cross-model stability, limitations and requirements for a larger pre-registered study.