Who Gets the Reward, Who Gets the Blame? Evaluation-Aligned Training Signals for Multi-LLM Agents

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Yang, Chih-Hsuan, Mallick, Tanwi, Chen, Le, Raghavan, Krishnan, Wells, Azton, Gueroudji, Amal, Foster, Ian T., Thakur, Rajeev
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909909371584512
author Yang, Chih-Hsuan
Mallick, Tanwi
Chen, Le
Raghavan, Krishnan
Wells, Azton
Gueroudji, Amal
Foster, Ian T.
Thakur, Rajeev
author_facet Yang, Chih-Hsuan
Mallick, Tanwi
Chen, Le
Raghavan, Krishnan
Wells, Azton
Gueroudji, Amal
Foster, Ian T.
Thakur, Rajeev
contents Large Language Models (LLMs) in multi-agent systems (MAS) have shown promise for complex tasks, yet current training methods lack principled ways to connect system-level evaluation with agent-level and message-level learning. We propose a theoretical framework that unifies cooperative game-theoretic attribution with process reward modeling to transform system evaluation into agent credit and then into response-level signals. Unlike prior approaches that rely only on attribution (e.g., Shapley) or step-level labels (e.g., PRM), our method produces local, signed, and credit-conserving signals. In success cases, Shapley-based credit assignment fairly allocates outcomes across agents and is refined into per-message rewards that promote cooperation while discouraging redundancy or sabotage. In failure cases, first-error localization yields repair-aware preferences that penalize harmful steps while rewarding corrective attempts. The resulting signals are bounded, cooperative, and directly compatible with reinforcement-based or preference-based post-training, providing a unified and auditable pathway from global evaluation to local supervision in LLM multi-agent training. Our contribution is conceptual: we present a theoretical foundation and training signals, leaving empirical validation for future work.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10687
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Who Gets the Reward, Who Gets the Blame? Evaluation-Aligned Training Signals for Multi-LLM Agents
Yang, Chih-Hsuan
Mallick, Tanwi
Chen, Le
Raghavan, Krishnan
Wells, Azton
Gueroudji, Amal
Foster, Ian T.
Thakur, Rajeev
Multiagent Systems
Artificial Intelligence
Computation and Language
Computer Science and Game Theory
Large Language Models (LLMs) in multi-agent systems (MAS) have shown promise for complex tasks, yet current training methods lack principled ways to connect system-level evaluation with agent-level and message-level learning. We propose a theoretical framework that unifies cooperative game-theoretic attribution with process reward modeling to transform system evaluation into agent credit and then into response-level signals. Unlike prior approaches that rely only on attribution (e.g., Shapley) or step-level labels (e.g., PRM), our method produces local, signed, and credit-conserving signals. In success cases, Shapley-based credit assignment fairly allocates outcomes across agents and is refined into per-message rewards that promote cooperation while discouraging redundancy or sabotage. In failure cases, first-error localization yields repair-aware preferences that penalize harmful steps while rewarding corrective attempts. The resulting signals are bounded, cooperative, and directly compatible with reinforcement-based or preference-based post-training, providing a unified and auditable pathway from global evaluation to local supervision in LLM multi-agent training. Our contribution is conceptual: we present a theoretical foundation and training signals, leaving empirical validation for future work.
title Who Gets the Reward, Who Gets the Blame? Evaluation-Aligned Training Signals for Multi-LLM Agents
topic Multiagent Systems
Artificial Intelligence
Computation and Language
Computer Science and Game Theory
url https://arxiv.org/abs/2511.10687