AgentFAIR: Multi-Agent Framework for Geospatial Data FAIRness Evaluation

Geospatial datasets are vital for applications ranging from urban planning to climate modeling, yet consistently assessing their compliance with FAIR principles (Findable, Accessible, Interoperable, Reusable) remains a significant challenge. Existing evaluation tools often use differing rubrics, evidence sources, and can fail on dynamic web pages or repository-specific identifiers, leading to substantial discrepancies in scores. To address these issues, researchers introduce AgentFAIR, a multi-agent framework that integrates structured metadata extraction with 13 sub-principle-specific LLM evaluators. Each evaluator generates a maturity score (0-3), cited evidence, and recommendations. A critic agent then reviews the evidence and consistency, capable of requesting targeted re-evaluation. Initial results show average FAIR scores for Findability (79.7%), Accessibility (70.4%), Interoperability (45.3%), and Reusability (72.0%). The system demonstrates high sub-principle agreement with the critic (89%) and strong alignment with expert consensus (82%), supporting its auditability and feasibility, despite limitations in benchmarking and generalization claims.