Every Eval Ever standardizes eval metadata

Hugging Face says Every Eval Ever and Community Evals now share a JSON schema for evaluation results, including runner provenance, model access, generation settings, metric meaning, and optional per-sample outputs. Why it matters: Model and agent evaluation is hard to trust when leaderboards omit run context. A portable result schema makes benchmark claims easier to compare, reproduce, and audit.

Try this: Use the schema as a checklist before accepting any eval result: require provenance, model version or access path, generation settings, metric definitions, and sample-level evidence when possible.