The FAIR principles are key to reusing Earth system data for analytics and AI. A primary challenge for reuse is judging a dataset's fitness for purpose, which depends on its representational, relational, contextual, and error characteristics, aspects not necessarily captured by traditional curation and documentation. As machine learning and AI become central to Earth system modeling and prediction, unintentional misuse becomes a critical hazard: data unfit for training models or their components can lead to downstream societal harm.

The session covers observations, simulations, reanalyses, and synthetic data. Training, fine-tuning, evaluation, data assimilation, downscaling, bias correction, and attribution each impose distinct data requirements, and provenance of these processes helps trace how data characteristics and uncertainties propagate to results. A key question is how to tell users and automated processes whether a dataset supports the intended use. Topics include:

- Metadata standards and documentation of intended and unsupported uses and dataset limitations
- Assessment methods combining metadata-derived indicators, content checks, lineage, validation, and domain expertise, and the limits of each
- Purpose-specific assessment of resolution, coverage, representativeness (extremes, transferability across regions, periods, and observing systems), physical consistency, and uncertainty
- Time, financial, and energy costs of assessing and improving fitness, how they are distributed, and the consequences of unsuitable reuse

Some reuse challenges can be addressed through curation or supplementary information; others require alternative data or narrower scientific claims. Contributors are encouraged to state the intended use, present assessment evidence, and describe resulting decisions and their effects on conclusions. Comparative studies, negative results, and post-mortems with transferable lessons are particularly welcome, including assessments that find a dataset unsuitable. Contributions may address single assessment steps or complete data-to-discovery workflows.

The session convenes data producers, repositories, research infrastructures, AI practitioners, and domain scientists to identify shared assessment practices that avoid redundant effort and to clarify when scientific judgment is essential. Its focus is how data properties and assessment evidence support or constrain scientific claims.