Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models

Hugging Face Daily Papers Papers
statistical-self-consistency language-models in-context-learning probabilistic-identities consistency-evaluation

Summary

This paper proposes a framework to test whether LLM estimates obey statistical self-consistency (law of total probability) across subpopulations, finding widespread violations and the 'macro fallacy' where fine-grained estimates align better with human data.

In-context learning is commonly interpreted as a form of conditional inference, in which the prompt specifies a context and the model's output is treated as an estimate of the corresponding conditional distribution. If this interpretation holds, then LLM estimates should satisfy basic probabilistic identities. In particular, the law of total probability asserts that prior-weighted conditional distributions aggregate into population-level marginals over any valid partition of the population. In this work, we investigate to what extent LLM estimates adhere to this self-consistency principle. We use binary trees as an evaluation scaffold to recursively partition a population into increasingly fine-grained subpopulations. We then prompt LLMs with verbalized subpopulation descriptions in context, aggregate the resulting estimates back into population-level estimates, and compare them across partitions of varying granularity. Applying this protocol across problem domains and state-of-the-art frontier models, we show widespread violations of basic consistency properties. An in-depth study of persona prompting reveals a pattern we call the macro fallacy: estimates reconstructed from more fine-grained subpopulation responses are often better aligned with human reference data than direct population-level estimates. This effect persists across variations in tree structure and estimation task, and can be partially recovered through implicit prompting. Together, these findings suggest that models possess relevant subpopulation knowledge but do not reliably propagate it into aggregate estimates. This gap establishes statistical self-consistency as an unsaturated, reference-free criterion for evaluating LLMs.
Original Article
View Cached Full Text

Cached at: 07/20/26, 09:42 AM

Paper page - Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models

Source: https://huggingface.co/papers/2607.15277

Abstract

In-contextlearningiscommonlyinterpretedasaformofconditionalinference,inwhichthepromptspecifiesacontextandthemodel’soutputistreatedasanestimateofthecorrespondingconditionaldistribution.Ifthisinterpretationholds,thenLLMestimatesshouldsatisfybasicprobabilisticidentities.Inparticular,thelawoftotalprobabilityassertsthatprior-weightedconditionaldistributionsaggregateintopopulation-levelmarginalsoveranyvalidpartitionofthepopulation.Inthiswork,weinvestigatetowhatextentLLMestimatesadheretothisself-consistencyprinciple.Weusebinarytreesasanevaluationscaffoldtorecursivelypartitionapopulationintoincreasinglyfine-grainedsubpopulations.WethenpromptLLMswithverbalizedsubpopulationdescriptionsincontext,aggregatetheresultingestimatesbackintopopulation-levelestimates,andcomparethemacrosspartitionsofvaryinggranularity.Applyingthisprotocolacrossproblemdomainsandstate-of-the-artfrontiermodels,weshowwidespreadviolationsofbasicconsistencyproperties.Anin-depthstudyofpersonapromptingrevealsapatternwecallthemacrofallacy:estimatesreconstructedfrommorefine-grainedsubpopulationresponsesareoftenbetteralignedwithhumanreferencedatathandirectpopulation-levelestimates.Thiseffectpersistsacrossvariationsintreestructureandestimationtask,andcanbepartiallyrecoveredthroughimplicitprompting.Together,thesefindingssuggestthatmodelspossessrelevantsubpopulationknowledgebutdonotreliablypropagateitintoaggregateestimates.Thisgapestablishesstatisticalself-consistencyasanunsaturated,reference-freecriterionforevaluatingLLMs.

View arXiv pageView PDFGitHub0Add to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.15277 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.15277 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.15277 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Language models struggle with compartmentalization

arXiv cs.CL

This paper investigates compartmentalization in LLMs, where models fail to share statistical strength across distinct representations of the same concept, leading to reduced sample efficiency and model capacity. The authors demonstrate this phenomenon in multilingual and multi-format settings and show that synthetic parallel data does not fully resolve it.

Some Large Language Models Exhibit Consistent Risk Attitudes

arXiv cs.AI

This paper introduces a framework to test whether large language models exhibit consistent risk attitudes across domains. It finds that most LLMs show intra-task and cross-domain stability in risk attitude, converging to a narrower distribution than humans.