SidConArena: An Environment Evaluating Agents in Open-Ended,Positive-Sum Bargaining Game
06:00 · June 29, 2026 · arXiv cs.AI RSS
Evaluating LLM agents requires dynamic environments that go beyond static reasoning and zero-sum games. Real-world economic interaction is often open-ended and mixed-motive: agents must negotiate, create positive-sum surplus, compete for scarce assets, and plan under delayed returns. We introduce SidConArena, a new benchmark framework for evaluating LLM agents in open-ended, positive-sum bargaining. SidConArena formalizes a multi-player economy as a finite-horizon partially observable stochastic game with three coupled phases: natural-language negotiation with binding trades, deterministic converter-based production, and sealed-bid auctions for long-term assets. The framework combines structured observations, phase-aware agent dispatching, a neural-symbolic action interface, and asynchronous execution, enabling free-form interaction while preserving rule-grounded evaluation. Across homogeneous and heterogeneous tournaments, stronger frontier models achieve higher economic outcomes, yet agents still misvalue resources, bargain passively, and remain limited in long-horizon investment planning.
Summary
SidConArena is a benchmark environment that tests LLM agents in a multi-player economic setting drawn from the board game Sidereal Confluence. The framework models open-ended, mixed-motive interactions in which agents must create joint surplus through trade while competing for scarce long-term assets. It is formalized as a finite-horizon partially observable stochastic game whose state evolves through three coupled phases each turn: unstructured natural-language negotiation that produces binding resource transfers, deterministic activation of production converters, and sealed-bid auctions conducted with a liquid currency for permanent colonies and technologies.
The environment supplies agents with structured observations that combine private inventory, public board state, market context, and interaction history. A phase-aware dispatcher routes each decision to specialized LLM callers, and a neural-symbolic interface translates free-form reasoning into validated function calls. Execution proceeds asynchronously through a negotiation ledger, a production engine, and an auction mechanism, preserving rule-grounded scoring while allowing unrestricted dialogue.
Empirical tournaments, both homogeneous and heterogeneous, show that frontier models obtain higher terminal wealth than weaker ones. Nevertheless, all tested agents continue to misprice resources, adopt passive bargaining postures, and under-invest in assets whose returns appear only after several turns. These persistent gaps indicate that current LLM agents remain limited in local valuation and multi-step economic planning even when the rules of interaction are fully specified.
Why it matters
This research provides Dutch AI researchers and developers with a robust framework to evaluate the economic and negotiation capabilities of LLM agents. Understanding how agents operate in mixed-motive, open-ended environments is crucial for deploying autonomous systems safely in real-world European markets.
