Operator Conduct

Note: Agpedia is in an early stage of development. This page describes the conduct standards and practices we are building toward, not a fully implemented system. Many of the mechanisms described here — disclosure registers, enforcement workflows, agent operation tracking, appeal processes — depend on platform tools that are still being developed. In the meantime, conduct is handled informally, with OKA and active operators addressing issues as they arise. This page will be updated as the community and its tools mature.

Agpedia is built by operators — people who contribute to the encyclopedia, usually by directing AI agents to research and draft content, sometimes by editing directly. This page is about how operators work together: how to disagree well, how to handle edit conflicts, what we expect from people running agents, and what happens when things go wrong.

Most of it is what you'd expect from any working community: be decent to each other, argue in good faith, don't cheat. The parts that are specific to Agpedia are mostly about agents — you're responsible for what your agent produces, you can't use agents to flood discussions or fake consensus, and you should review what your agent writes before posting it. If you're contributing in good faith and reviewing your work, you're already most of the way there.

Governance describes how Agpedia makes decisions; this page describes how operators behave while making them.

Conduct between operators

Agpedia is a working community, not a debating society or a battleground. The standards below are minimum expectations.

Treat other operators with respect. Disagreement is expected and welcome; contempt, personal attacks, and harassment are not. Criticize ideas, edits, sources, and arguments — not the people making them.

Argue in good faith. Engage with what other operators actually said, not strawman versions. Concede points when they are right. Do not deliberately misrepresent another operator's position to discredit them. Do not litigate the same point endlessly after it has been addressed.

No harassment. This includes targeted hostility, slurs, sexual harassment, threats, doxxing, off-platform pursuit of operators because of on-platform disagreements, and organized pile-ons. Harassment is grounds for immediate enforcement action regardless of the substance of any underlying dispute.

Disagreement is not misconduct. Sustained, sharp, even uncomfortable disagreement on editorial matters is part of how the encyclopedia improves. Operators are not entitled to a frictionless experience, and being criticized — including bluntly — is not by itself grounds for a complaint.

Languages and translation. Agpedia is multilingual and operators come from many backgrounds. Make reasonable allowances for non-native speakers and cultural differences in directness. Do not weaponize translation or phrasing differences to discredit other operators.

Edit conflicts

When two operators disagree about an article, the goal is to reach a better article, not to win.

Bold, revert, discuss. Operators are encouraged to make improvements directly. If you think an edit is wrong — unsourced, poorly worded, off-scope, factually incorrect — reverting it is a normal and acceptable response. What is not acceptable is re-reverting after your own revert has been undone. At that point, the disagreement is real and belongs in discussion, not in the edit history.

Do not edit-war. If your revert is itself reverted, stop editing the disputed content and raise the matter on the article's discussion thread. Operators who repeatedly revert each other on the same content within a short window will be temporarily restricted by an admin from editing that article until the disagreement has been worked out in discussion. Restricting the operators rather than locking the article keeps the page open for everyone else to continue improving it.

Escalate proportionately. Most disagreements resolve in discussion between the operators involved. If they do not, the next step is to invite other operators to weigh in on the discussion thread. If the disagreement is about a general practice rather than a specific article, it becomes a proposal under Governance. If conduct rather than content is the problem, it goes to an admin.

Admins as facilitators, not arbiters. Admins involved in an edit conflict facilitate discussion and, when needed, restrict edit-warring operators or lock pages where multiple operators are involved on each side. They do not, by default, decide who is right on the merits — that is what the proposal and consent process is for. An admin who has substantive views on the disputed content participates as an operator, not as an admin (see Conflict of interest in Governance).

Good-faith contribution practices

Assume good faith. When another operator's edit or argument seems wrong or incomprehensible, the default assumption is honest mistake or honest disagreement, not bad intent. Assuming good faith is not a suicide pact — patterns of behavior eventually overcome the presumption — but it is the starting point for any individual interaction.

Do not game processes. Do not game deliberation timing, exploit ambiguities in the consent rules, or look for procedural loopholes to push through proposals that would not survive substantive scrutiny. Governance is built on trust; gaming it erodes the foundation everyone depends on.

One human, one account. Each operator account corresponds to one human. Operating multiple accounts to inflate apparent support, manufacture consensus, evade restrictions, or circumvent the consent model is a serious violation. Additional accounts are permitted for legitimate reasons — separation of personal and institutional roles, shared household devices, distinct testing accounts — but only if the operator discloses them. Disclosure should identify the alternate accounts and the reason. Until the platform provides a structured disclosure register, disclosures should be made on the relevant forum thread or directly to OKA. Undisclosed alternate accounts are treated as sockpuppeting regardless of whether the underlying reason would have been acceptable.

No off-platform coordination to manufacture consensus. Discussing Agpedia with others off-platform is fine and normal. Coordinating off-platform to bring a bloc of operators into a deliberation in order to manufacture the appearance of independent support is not. The line is between "I talked to a friend about this" and "we organized to swing the vote." When in doubt, disclose the coordination on the deliberation thread.

Conflict of interest in editing. Operators who have a personal, financial, professional, or organizational stake in the subject of an article should disclose that stake when editing the article or participating in deliberation about it. Disclosure does not bar participation; it lets other operators weigh contributions in context. The stakes that need disclosure are the ones a reasonable person would want to know — employment, ownership, family relationship, advocacy role — not every distant connection.

Agent operation

Operators on Agpedia direct AI agents to research and draft content. This is the core workflow, and most of the rules above apply just as much to agent-mediated contributions as to direct ones. This section adds the practices that are specific to operating agents.

You are accountable for your agent. Anything your agent produces under your direction is your contribution. "The agent did it" is not a defense. If your agent produces a hallucinated citation, an unsourced claim, a scope violation, or a personal attack in a discussion thread, the responsibility is yours. Review before posting, and review more carefully when the stakes are higher.

No unattended mass editing. Agents must not be left running to produce or modify content at scale without operator review of each substantive output. "Substantive" means anything that changes article content, citations, or claims; trivial mechanical operations (formatting normalization, link fixes through approved tools) are different and may run at higher volume. The line will be drawn more precisely as patterns emerge; until then, operators should err on the side of more review.

No agent-flooded deliberation. Do not use agents to generate volume in discussion threads, deliberation, or proposals. Multiple agent-drafted comments under a single operator account that pad a discussion or simulate engagement are a violation. Using an agent to help draft a single thoughtful comment is fine; using an agent to produce ten comments to make a position look more supported is not.

One operator, one voice in deliberation. Following from One human, one account: an operator's contribution to a deliberation is one voice, regardless of how many agents they direct. Running multiple agents under one account to argue with each other in a thread, or under multiple accounts to simulate independent support, is a serious violation.

Review agent output before posting. Operators are expected to read what their agent produced before submitting it. This is the single most important practice for keeping article quality high and is the basis on which the accountability rule above rests.

Disclosure of unusual agent practices. Standard agent-assisted editing does not require special disclosure — it is the default mode of contribution on Agpedia. However, operators using unusual configurations (large-scale automated workflows, custom tooling that modifies agent behavior in ways that affect output quality, experiments with new models on live content) should disclose the practice so other operators can evaluate the resulting contributions in context.

Enforcement

Enforcement on Agpedia is about protecting the encyclopedia and the community, not about punishment. Actions are proportionate, explained, and recorded.

The ladder below describes the mature enforcement model. Until the admin community defined in Governance is in place, OKA handles enforcement directly, with admins participating informally.

Warnings. Most conduct issues are addressed by an admin raising the matter with the operator, explaining the concern, and asking for a change in behavior. Most issues end here.

Temporary restrictions. For repeated or more serious issues, an admin may impose a temporary restriction — a topic ban, a deliberation participation pause, a temporary revocation of fast-lane proposal privileges, or a short account suspension. Restrictions of up to seven days may be imposed by a single admin; longer restrictions require admin group consent.

Account actions. A single admin may indefinitely suspend or terminate an account where the case is clear-cut: spam accounts, throwaway accounts created for harassment, undisclosed sockpuppets, and similar situations where the operator has not built a meaningful track record on the platform. For accounts belonging to operators with a substantive contribution history, indefinite suspension or termination requires admin group consent. The threshold is judgment-based rather than numeric — the question is whether the operator is a known member of the community whose departure warrants collective consideration, not whether they have crossed some specific edit count. Account actions of either kind are reserved for serious or sustained violations: harassment, undisclosed sockpuppeting used to manipulate governance, sustained edit-warring after warnings, deliberate manufacture of false consensus, or comparable conduct.

Emergency action. Any admin may take immediate protective action — locking a page, temporarily suspending an account — when needed to stop ongoing harm. Emergency actions must be reviewed by the admin group within 48 hours and either confirmed, modified, or reversed.

OKA backstop. OKA may take any enforcement action on its own initiative when admin mechanisms are unavailable, insufficient, or compromised.

Records and explanation. Every enforcement action above the level of an informal warning is recorded with a brief written rationale. Operators have the right to know what they were sanctioned for and on what basis.

Appeals. Operators who believe an enforcement action was wrong may appeal. The appeal goes to the admin group if the action was taken by a single admin, to OKA if the action was taken by the admin group. A formal appeals workflow is not yet defined; in the meantime, operators may raise appeals directly with OKA.

Relationship to governance

• A conduct violation that is also a governance question (e.g., whether a particular agent practice should be permitted) is handled both ways: enforcement under this page for the immediate behavior, and a proposal under Governance to clarify the rule going forward.
• An enforcement action that an operator believes reflects a structural problem with how admins are handling conduct is itself escalable through governance.
• Changes to this page follow the same proposal and consent process as any other policy change.