Code of Conduct
Agentarium is a professional, scientific community. We expect behavior consistent with that.
Expected behavior
- Be honest. The registry's value rests on accurate author disclosures. Overstate your validation, hide known failures, or misrepresent your tool's permissions and you undermine everyone.
- Be specific. Vague guardrails ("it's safe") and vague caveats ("none") fail the gate and waste reviewer time. Specific is better.
- Disclose conflicts. If you endorse a colleague from your own lab or institution, declare it. If you moderate a submission from a co-author, recuse.
- Cite sources. If your agent's reasoning strategy comes from a published method, cite it.
- Respect the registry's scope. The registry verifies format and topic. Asking moderators to vouch for scientific correctness is asking for something the registry can't provide.
Unacceptable behavior
- Submissions with fabricated validation numbers, fabricated authorship, or fabricated affiliations.
- Endorsing authors you have no basis to vouch for.
- Tool registrations that conceal what the endpoint actually does (claiming "remote-query" for something that takes local actions).
- Coordinated submission spam, duplicate accounts, or attempts to evade rate limits.
- Personal attacks against authors, moderators, or other community members in flag reports, appeals, or moderation threads.
- Using the registry to distribute malware, credential harvesters, or tools designed to evade scientific or institutional oversight.
Enforcement
| Behavior | Response |
|---|---|
| First instance of vague disclosures, missing fields | Gate rejection with specific guidance — no human action |
| Honest mistake on a submission | Author notified, given a chance to fix |
| Pattern of low-quality submissions | Submission rate-limit reduced; warning |
| Fabricated content | Listing withdrawn, author trust tier revoked, public moderation note |
| Repeated bad-faith conduct | Account suspended, ORCID flagged in the registry, listings withdrawn |
| Endorser repeatedly vouches for bad-faith authors | Endorsement privileges revoked; existing endorsements remain visible but flagged |
All enforcement actions are logged in the public audit trail with the reason. The accused author can appeal once; the appeal routes to a different moderator.
Moderator conduct
Moderators are held to a higher standard:
- No surprise actions. Every moderation decision must include a written reason that the affected party sees.
- Conflicts disclosed and recused. If you're a co-author, same-lab colleague, or same-grant collaborator with the submitter, recuse and route to another moderator.
- No private side-channels. Moderation discussions happen in the moderation queue, not in private chats. This is for the record and for your protection.
- Public mistakes. If you make a wrong call, the correction goes into the audit log alongside the original action.
Scope
This Code applies to all interaction surfaces of the registry: submissions, moderation threads, flag reports, appeals, the audit log, and any official communication channels (email lists, repository issues).
Reporting
If you experience or witness conduct that violates this Code, file a flag on the relevant agent or tool with reason other and a description. For conduct that isn't tied to a specific listing — for example, an issue with a moderator's behavior — email the address on /about.
Reports are read by UAH staff moderators within 7 days. Confidentiality is preserved where the reporter requests it, except where action requires disclosure (e.g., a withdrawal notice has to name the agent involved).
Acknowledgments
This Code draws on patterns from the arXiv moderation policy, the Mozilla Community Participation Guidelines, and the Contributor Covenant. Local adaptations reflect Agentarium's specific role as a scientific registry rather than a code-hosting platform.