Release Managed-Care Operations Handbook on Hugging Face

[NielsRogge]

Hi @weirayao 🤗

I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2605.16679.

The paper page lets people find artifacts about it, and I see you've already claimed it and linked the chi-bench dataset—that's awesome! CHI-Bench is a fantastic contribution to evaluating agentic workflows in complex domains like healthcare.

I noticed in the README that the Managed-Care Operations Handbook (the 1,279-document policy corpus) currently lives off-Hub and requires a manual request form for access. Would you like to host this artifact on Hugging Face as well?

We support gated datasets, which would allow you to require users to share their contact information and/or agree to specific terms before they can access the files—similar to your current form, but integrated directly into the HF ecosystem.

Hosting it on Hugging Face would improve visibility and allow researchers to easily load the knowledge base via the datasets library:

from datasets import load_dataset

# After access is granted
handbook = load_dataset("actava/managed-care-handbook")

Besides that, there's the dataset viewer which would allow people to browse the policy documents directly in the browser.

Let me know if you're interested or if you'd like some help setting up a gated repository for this!

Kind regards,

Niels
ML Engineer @ HF 🤗

[weirayao]

Hi @NielsRogge,

Thanks for reaching out 👋 and for featuring our paper! Great to connect with the HF open-source team.

This is a great idea. We've been using a Google Form for collection of access requests, and moving to a gated HF gated dataset would be a much smoother experience for researchers. Thanks for introducing this cool feature to us.

Weiran

[hlnchen]

Hi @NielsRogge,

Thanks for bringing this issue to us. It is a great advice. We plan to prepare a gated dataset on HF so people can access the whole benchmark with lower friction. Furthurmore, we are wondering if it is possible to register our benchmark as a HF benchmark dataset like https://huggingface.co/datasets/openai/gsm8k, with eval results of all models/harnesses registered as well? I found that we/the HF team need to PR all model repos and whitelist our benchmark to do that. It would be appreciated to get some assistance on this thread.

[NielsRogge]

Hi @weirayao and @hlnchen,

Thanks so much for uploading both the CHI-Bench dataset and the Managed-Care Operations Handbook to the Hugging Face Hub. It's great to see both artifacts now linked to the paper page at https://huggingface.co/papers/2605.16679 and prominently featured in your GitHub README.

Regarding your question about registering CHI-Bench as a benchmark dataset: I'll check with the team.

Thanks again for making these artifacts available on the Hub.

Best,
Niels

[haolin-chen-actava]

Thanks @NielsRogge. Let us know any update from the team.