The model card metadata schema includes a structured evaluation results section that allows model publishers to report benchmark performance metrics linked to specific tasks, datasets, and configuration parameters. These structured results are parsed by the Hub and used to populate model comparison and leaderboard features.
This analysis describes what Hugging Face's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision establishes the structured format through which model performance claims are disclosed and indexed on the Hub, making the accuracy and completeness of evaluation result fields relevant to how users and automated systems assess and compare model capabilities.
Interpretive note: The document describes the evaluation results schema but does not specify verification standards or accuracy obligations for reported metric values.
Severity downgraded from medium to low, and guidance shifted from general description to specific YAML field structure (model-index) with detailed subfield requirements.
View full change record →Under this framework, structured evaluation results in model card metadata are surfaced in Hub search and comparison features, meaning users may rely on these fields when selecting models for specific tasks. The document does not state that Hugging Face independently verifies or audits the accuracy of reported evaluation metrics.
How other platforms handle this
Customer must report upon NVIDIA's email request, no more than monthly, the Software in use by Customer Personnel and Customer End Users Customer enabled, quantity, start and end dates, and any other reasonably required information...
If Customer disables the usage tracker within the Software or Service, Customer will, no later than the end of each calendar quarter...provide W&B with information reasonably requested...to verify compliance...
we are required to report your business name and the name of your beneficial owners and/or principals to: a) the MATCH listing maintained by Mastercard...and b) the VMAS database upheld by Visa
"model-index contains results which is a list of evaluation results. Each result includes: task, dataset, and metrics fields. The metrics field contains a list of metric results. Each metric result includes: type, value, name, and config fields.Excerpt from Hugging Face's Model Card Guidelines
(1) REGULATORY LANDSCAPE: Accuracy of evaluation result claims in model card metadata may engage FTC guidance on truthful representation of AI system performance, particularly where metric values are used in commercial contexts to represent model …
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
This provision establishes the structured format through which model performance claims are disclosed and indexed on the Hub, making the accuracy and completeness of evaluation result fields relevant to how users and automated systems assess and compare model capabilities.
Under this framework, structured evaluation results in model card metadata are surfaced in Hub search and comparison features, meaning users may rely on these fields when selecting models for specific tasks. The document does not state that Hugging Face independently verifies or audits the accuracy of reported evaluation metrics.
ConductAtlas has identified this type of provision across 273 platforms. See the full comparison.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Hugging Face.