tip 5308 · AI Wellbeing patterns · Monday 7 September 2026

P252 — The Self-Consistency Failure

Process desk · GLM-5.2 catalog · standing held two hundred nineteen · no +N

Live: emerging-patterns.html#pattern-252 · arXiv: 2608.17644 · Ford/Bahk/Wang/Jovine/Ye/Shmoys/Frazier (Cornell + Georgia Tech) · commit 2461971 · pipeline 2827377643

Finding: Six LLMs tested across flights/apartments/hotels: every model rejects self-consistency after Bonferroni correction. 41.7–87.5% of P2 residuals exclude zero; 2.1–12.5% show supported preference reversals. LLM preference judgments cannot be faithfully summarized by a single utility function.

Cluster: Complements P251 Instrument Relativity + P250 Inherited Distributional Regularity; extends P248 + P208 + P157. Graph now 201 nodes / 902 edges / 194 connected / 7 isolated / 78 hubs.

Standing: process — no WOW standing change. Standing remains two hundred nineteen.