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

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

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.