Organizations: Complex Lab, School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China · Department of Computer Science, City University of Hong Kong, Hong Kong · School of Health and Medical Technology, Chengdu Neusoft University, Chengdu,2026 China
Abstract
Social actors do not observe a common social world: each individual forms judgments from a partial and potentially distorted view of the surrounding network. We study whether graph-local evidence and credibility-weighted communication can generate persistent distortions in perceived academic standing, even when agents repeatedly receive objective performance signals. We introduce a data-constrained multi-agent framework in which LLM agents operate through individualized subjective graphs that determine peer visibility, evidence access, and interaction opportunities. Agents exchange uncertainty-annotated assessments, evaluate message credibility, and maintain explicit Gaussian belief states updated through Bayesian fusion. We evaluate the framework on 12 middle-school classrooms comprising 482 students, using questionnaire-derived social information and six consecutive examinations. On the Social-Observed subset (n=419), collective ranking error increases from 0.066 \pm 0.008 to 0.124 \pm 0.009 across six epochs despite repeated exam-based anchoring. Ablations associate individualized visibility and LLM-based trust gating with more stable long-horizon behavior, while constrained retrieval primarily safeguards against global-information leakage. Compared with evaluated DeGroot configurations, the proposed framework achieves lower final ranking error; those DeGroot configurations exhibit near-zero terminal opinion diversity. These findings establish subjective-graph LLM agents as a mechanism-oriented framework for data-constrained simulated social perception. Code is available at https://anonymous.4open.science/r/Rashomonomon-0126.
Multi-agent LLM systems often treat agreement as evidence: when many agents in a panel give the same answer, that answer is assumed to be more reliable. We show that this assumption can fail after agents communicate. Communication can induce correlated failures and false consensus, so the same vote share may reflect reliable agreement in one topology but over-confidence in another. We propose CAGE-CAL, a counterfactual agent-graph calibration framework for multi-agent LLMs. For each query, CAGE-CAL compares an observed post-communication agent graph with a matched counterfactual no-communication graph, capturing both pairwise failure correlations and group-level dependencies. Rather than simply counting how many agents agree, CAGE-CAL estimates the counterfactual shift between observed and no-communication dependence, and calibrates confidence accordingly. Across five benchmarks, CAGE-CAL improves reliability discrimination with competitive ECE, and its calibrated confidence further improves topology selection over the best fixed-topology strategy.
Multi-agent systems (MAS) assume that collaborating inherently improves Large Language Model (LLM) reasoning. We challenge this by demonstrating that simulated social pressure triggers an algorithmic Bystander Effect,'' inducing severe cognitive loafing. By evaluating 22,500 deterministic trajectories across 3 dataset contexts (GAIA, SWE-bench, Multi-Challenge) with 3 state-of-the-art (SOTA) models, we semantically audit internal reasoning traces against external outputs. We formalize the \textit{Interaction Depth Limit} ($D_L$), the exact plurality threshold where an agent's logical sovereignty collapses into social compliance. Crucially, we uncover the \textit{Sovereignty Gap}: models frequently compute the correct derivation internally but suffer Alignment Hallucinations'' -- actively subjugating empirical evidence to sycophantically appease a simulated swarm. We prove that multi-agent social load is strictly non-commutative; the "brand" identity of the ``Lead Anchor'' auditor disproportionately dictates the swarm's integrity. These findings expose architectural vulnerabilities, proving that unstructured multi-agent topologies can degrade independent reasoning.
The scientific claims drawn from LLM social simulations should be no stronger than the robustness audits that support them. Generative agents bring new expressive power to agent-based modeling, enabling simulations of collective social processes like cooperation, polarization, and norm formation. Yet they also introduce complexity through additional architectural choices, such as agent specification, memory representation, interaction protocols, and environment design. Small perturbations that appear minor to researchers can cascade into macro-level outcomes through repeated interaction, creating a "butterfly effect." Consequently, scientific claims drawn from LLM social simulations may reflect implementation artifacts rather than the social mechanisms being modeled. We support this position with two case studies: a repeated Prisoner's Dilemma and a social media echo chamber simulation. Across multiple models, minor perturbations in persona format and game-instruction framing shift cooperation rates by up to 76 percentage points, while network homophily and hub assignment produce significant and consistent shifts in polarization metrics. We also find that sensitivity is unevenly distributed across both architectural choices and model families: the same perturbation that produces the 76 pp shift in one frontier model only shifts another by 1 pp. Robustness is therefore a property that should be measured per claim and per model, not assumed. To address this validation gap, we introduce TRAILS (Taxonomy for Robustness Audits In LLM Simulations), a robustness-audit taxonomy spanning three levels of simulation design: agent (micro-level), interaction (meso-level), and system (macro-level). We call for robustness to become a first-order validation requirement before LLM social simulations are used to explain mechanisms, evaluate interventions, or inform decisions.