Hosted and managed by the University of Alabama in Huntsville

Agentarium

scientific agent registry
ToolsGovernanceSign in with ORCID
← All policies

Endorsement Policy

Endorsement is how Agentarium handles the cold-start trust problem for new authors. It's the same pattern arXiv uses for new posters: one existing community member vouches that you're a real scientist working in a real scientific domain, and from that point on you can submit freely.

An endorsement is not a peer review. Endorsers are not vouching that your agent is correct, that your validation numbers are honest, or that your tool will stay online. They are vouching that you are who you say you are, that you have plausibly worked in the domain you claim, and that your initial submission isn't on its face fraudulent.

Who can endorse

To endorse another author, you must meet one of these:

  1. Be a moderator (Tier 3 or above).
  2. Be UAH-affiliated and ORCID-verified.
  3. Be NASA-IMPACT-affiliated and ORCID-verified.
  4. Be an already-endorsed author with at least 2 active agents in the registry.

Endorser status is visible on your public profile, and it's part of your reputation in the registry.

How endorsement works

  1. A new author signs in with ORCID and prepares one or more submissions in draft state.
  2. They click Request endorsement and either:
    • Name a specific endorser who has agreed in advance, or
    • Request a general moderator review.
  3. The endorser sees the author's profile, their drafts, and any institutional info ORCID provides.
  4. The endorser decides: endorse, decline, or request more information.
  5. If endorsed, the author moves to Tier 2 (Endorsed) and can publish freely going forward.

SLA: 72 hours for a first response (endorse, decline, or request info).

The endorsement is public: the endorsee's profile shows who endorsed them; the endorser's profile shows whom they've endorsed.

What endorsers should check

A short, focused review — not a peer review:

  • Does the author's ORCID exist and match the claimed institution?
  • Is the claimed scientific domain plausible given the author's record?
  • Is the first submission's structure substantive (i.e., not obvious spam)?
  • Are the declared tools real?
  • Does the agent description match the author's stated expertise?

You are not checking:

  • Whether the validation numbers are accurate
  • Whether the agent works as claimed
  • Whether the underlying methodology is sound
  • Whether the prompts are optimally designed

Those are the reader's job.

What endorsers are vouching for

Concretely: that if you, the endorser, were searching for an agent in this author's claimed domain, you would not be embarrassed to find this author registered. That's the bar. Not "I approve of this work." Not "this is correct." Just: this person belongs in the community.

Endorser accountability

Endorsement is a real responsibility. The system tracks it:

  • Every agent shows the endorser who let the author in.
  • Your public profile lists everyone you've endorsed and their current trust tier.
  • If your endorsee is later withdrawn for fraud, that withdrawal is associated with your endorsement.
  • Two endorsees withdrawn for fraud → your endorsement privileges are revoked.
  • Your existing endorsements remain visible but get a "endorser privileges revoked" tag.

This is intentionally social — endorsement is supposed to slow down sock-puppet attacks, and visibility is most of what makes it work.

How to decline gracefully

If you're asked to endorse someone you can't vouch for, decline with a brief, neutral reason. Common reasons:

  • "I don't know this author well enough to endorse."
  • "Outside my domain — try [other endorser]."
  • "Please add an example agent with at least one full validation block before I review again."

Declines are not public on the requester's profile; they're visible only to the requester and to moderators.

What if no one will endorse?

If you've requested moderator endorsement and 72 hours have passed without response, your request escalates to UAH staff. If after another 72 hours your request still isn't actioned, you can:

  1. Improve your draft submission and try again.
  2. Provide additional context (links to publications, code repositories, prior work).
  3. Contact /about to ask what would unblock the request.

The registry should never silently refuse to endorse. Either you get an endorsement, a decline with reason, or a request for more information.

Revocation

Endorsements can be revoked by moderators with cause — typically when:

  • The endorsee is confirmed to have misrepresented their identity (fake ORCID, fake institution).
  • The endorsee turns out to be a sock puppet of an already-banned account.
  • The endorser pattern shows bad-faith vouching (rubber-stamping spam).

Revocation is logged in the public audit trail. The endorsee's listings remain visible but flagged.

Why not just open submission?

The registry is in the citation graph. Listings get DOIs (eventually) and resolve to permanent URLs that papers will cite. An open-submission model — like PyPI — would let bad-faith actors register agents that look real to a non-expert reader. Endorsement makes that meaningfully more expensive while keeping the barrier low for legitimate scientists.

The cost: a real human has to look at your account once. The benefit: every listing has a traceable chain back to an accountable scientist.

Acknowledgments

This policy is adapted from arXiv's endorsement system, with adjustments for the scientific-agent context (notably the accountability trail and explicit narrowness of what endorsement vouches for).