aiqbench analysis
SOCIA-$\nabla$ introduces a diagnosis-driven multi-agent framework that uses LLMs to iteratively synthesize, execute, and repair executable simulator code within a textual computation graph. By attributing behavioral residuals to specific mechanisms and converting them into localized repair signals, the system achieves high fidelity in simulating personal mobility and geo-social diffusion scenarios. This work advances automated simulation generation by integrating diagnostic reasoning with multi-agent orchestration to handle both in-distribution and out-of-distribution tasks.
SOCIA-$\nabla$: Textual Gradient Meets Multi-Agent Orchestration for Automated Simulator Generation
arXiv:2505.12006v5 Announce Type: replace Abstract: In this paper, we present SOCIA-$\nabla$, a diagnosis-driven multi-agent framework for automated simulator-code construction. SOCIA-$\nabla$ iteratively refines executable simulator code within a textual computation graph, where LLM-driven agents synthesize, execute, evaluate, diagnose, and repair simulators. A grounded simulator blueprint attributes behavioral residuals to candidate mechanisms and their spatial or interaction contexts, which are then converted into localized textual repair signals with history-aware memory and constraint projection. We evaluate SOCIA-$\nabla$ on the real-world LLMob personal-mobility benchmark and a controlled semi-synthetic geo-social mask-adoption benchmark. Across in-distribution simulation, out-of-distribution what-if generation, and geo-social diffusion, SOCIA-$\nabla$ achieves the strongest overall simulator fidelity. SOCIA-$\nabla$'s code and data are publicly available at https://github.com/cruiseresearchgroup/SOCIA-Nabla.
Abstract:In this paper, we present SOCIA-$\nabla$, a diagnosis-driven multi-agent framework for automated simulator-code construction. SOCIA-$\nabla$ iteratively refines executable simulator code within a textual computation graph, where LLM-driven agents synthesize, execute, evaluate, diagnose, and repair simulators. A grounded simulator blueprint attributes behavioral residuals to candidate mechanisms and their spatial or interaction contexts, which are then converted into localized textual repair signals with history-aware memory and constraint projection. We evaluate SOCIA-$\nabla$ on the real-world LLMob personal-mobility benchmark and a controlled semi-synthetic geo-social mask-adoption benchmark. Across in-distribution simulation, out-of-distribution what-if generation, and geo-social diffusion, SOCIA-$\nabla$ achieves the strongest overall simulator fidelity. SOCIA-$\nabla$'s code and data are publicly available at this https URL.
14 pages, 2 figure, 3 tables. This paper has been accepted to the ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems 2026 (ACM SIGSPATIAL 2026), the thirty-fourth edition
Artificial Intelligence (cs.AI)
arXiv:2505.12006 [cs.AI]
(or arXiv:2505.12006v5 [cs.AI] for this version)
Full text under the publisher's open licence · 150 words
Related coverage
How Anthropic’s Dynamic Workflows Coordinate Claude Agents (Updated with video!)
Latent Space · Oct 12, 2026
The Thin Orchestrator: How We Made Our AI Agents More Reliable by Giving the Router Less to Do
Meta Engineering · Oct 19, 2026
How to Use Claude Managed Agents?
Analytics Vidhya · Jun 2, 2026
Agentic AI-enabled Semantic Commissioning of a Cognitive Digital Twin for Reconfigurable Manufacturing
arXiv · Sep 10, 2026