Moderation principles
The platform’s ethos is curiosity with guardrails. Moderation is those guardrails — explicit, inspectable, and appealable.
Protect people & sites
No doxxing, stalking, threats, or harassment. No instructions that materially increase the risk of looting, vandalism, or harm. “It’s public already” is not a free pass.
Protect evidence integrity
Claims are allowed; fabricated evidence is not. We label uncertainty, demand provenance, and require citations for “this proves X.” Bad reasoning is permitted; bad faith is not.
Make decisions explainable
When we take action, we provide a plain-language reason, the rule invoked, and how to appeal. If we can’t explain it, we shouldn’t enforce it.
What we moderate
- Content: posts, evidence uploads, citations, comments, profiles, collections, and claim threads.
- Conduct: harassment, brigading, impersonation, spam, manipulation of voting/reputation.
- Safety metadata: location precision, access routes, and any “how-to” that enables harm.
What we don’t moderate
- Disagreement: “That’s wrong” is allowed. “You’re subhuman” is not.
- Speculation: labeled speculation is fine; presenting it as verified fact without evidence is not.
- Unpopular views: being unpopular is not a violation; being abusive is.
Trust roles & who can do what
Moderation is easier when responsibility is earned. Trust roles limit the blast radius of new accounts while keeping contribution open.
Explorer
Can browse, save sites, follow tags, and report issues. Limited posting until basic trust is earned (email verified + light friction).
Contributor
Can submit evidence, propose edits, and join claim threads. Contributions are reviewed faster as reputation grows.
Curator / Steward
Can resolve basic disputes, lock heated threads, request provenance, and escalate cases. High-trust actions are logged publicly.
Why roles exist
Research communities attract both brilliant obsessives and chaos goblins. Trust roles keep the door open without leaving the lab unattended. Most users never notice roles — they just experience calmer threads and better evidence hygiene.
What earns trust
- Accurate citations and clean provenance.
- Good-faith engagement (steelman before you attack).
- Helpful edits and de-escalation behavior.
- Consistency over time (not one viral post).
Enforcement ladder
We use the least-restrictive action that reasonably reduces harm — except where safety demands immediate removal.
-
Level 0Label / Context
Soft intervention: labels, prompts, and citations
Used when content is allowed but needs clarity: “Speculation,” “Needs citation,” “Provenance unclear,” “Location precision reduced.” This is the default for messy but non-abusive research.
-
Level 1Warning
Warning + education
Used for first-time rule breaks that don’t create immediate danger: mild harassment, repeated low-effort spam, sloppy attribution. Warnings include the rule and what to do differently.
-
Level 2Limited reach
Rate limits, thread cooldowns, and temporary locks
Used when behavior shows escalation: dogpiling, brigading signals, repeated insults, or “reply storms.” Threads can be slowed, temporarily locked, or moved to “structured mode.”
-
Level 3Removal
Content removal
Used for doxxing, threats, hate, impersonation, fraud, deliberate evidence fabrication, and heritage endangerment details. Removal includes a reason + appeal route.
-
Level 4Account action
Suspension / ban
Used for repeated abuse, evasion, coordinated manipulation, serious threats, or sustained deception. Permanent bans are reserved for patterns, not misunderstandings.
Immediate removals
Some categories skip straight to Level 3/4: credible threats, doxxing, sexual exploitation content, explicit looting instructions, and repeat impersonation/fraud. Speed matters.
Evidence disputes ≠ policy violations
“This scan is misinterpreted” is a normal scientific argument. We’ll push for provenance and better citations, but we don’t remove content just because it’s controversial.
Rule categories
These are the buckets we use for decisions (so “why was this moderated?” has a real answer).
Safety & harm
Threats, harassment, stalking, doxxing, and “go bother this person” calls-to-action. Also: instructions enabling physical harm.
Heritage endangerment
Looting routes, precise hidden locations, access instructions to restricted areas, or encouragement to damage/deface protected sites.
Fraud & fabrication
Forged documents, fake citations, manipulated media presented as authentic, impersonation of experts/institutions, coordinated deception.
Hate & dehumanization
Attacks on protected groups, slurs, dehumanizing language, and content that celebrates violence against groups or individuals.
Spam & manipulation
Repeated promotional dumping, link farms, bot-like behavior, vote/reputation gaming, brigading, and coordinated harassment.
Privacy & sensitive data
Personal data posted without consent, private comms dumps, and sensitive details that aren’t necessary for evaluating a claim.
AI-generated media
AI imagery is allowed if it is labeled and not presented as primary evidence. “AI concept art” is not “site documentation.” See AI Policy.
Copyright & rights
We prefer linking to sources over rehosting. If a takedown request is valid, we remove and log. If it’s a dispute, we follow a process. (And we still require citations—because citations are not copying.)
Report & appeal
Reporting is how the community protects itself. Appeals are how we avoid becoming an unaccountable priesthood.
Founding Access
Help us tune the moderation system before launch: evidence labels, appeal UX, heritage-protection defaults, and transparency logs. Founding Access ships in waves.
Moderation log (public receipts)
This transparency log shows how we capture what changed, why, and which rule was invoked.
-
Jan 8, 2026Label
MOD-2026-001 — Added “Needs provenance” label to an evidence upload
Action: Label only. Reason: missing source metadata for a scan. User prompted to add provenance and citation. Rule: Evidence integrity → provenance required for “primary evidence” status.
-
Jan 8, 2026Cooldown
MOD-2026-002 — Slowed a claim thread (cooldown mode)
Action: Thread cooldown. Reason: rapid-fire replies + escalating insults. Users instructed to steelman and cite before rebuttal. Rule: Conduct → harassment / incivility escalation.
-
Jan 8, 2026Removal
MOD-2026-003 — Removed doxxing details from a comment
Action: Partial redaction + removal of personal identifiers. Reason: posted private contact details without consent. Rule: Privacy → personal data / doxxing.
-
Jan 8, 2026Heritage
MOD-2026-004 — Reduced location precision for a sensitive site
Action: Location precision reduced + access route removed. Reason: content materially increased looting risk. Rule: Heritage protection → actionable access instructions prohibited.
-
Jan 8, 2026Reversed
MOD-2026-005 — Appeal reversal (label changed, content restored)
Action: Restored content, kept “Speculation” label. Reason: initial removal misclassified a hypothetical as “instruction.” Rule: Appeals → second review; resolved via less-restrictive action.
Transparency cadence
- Public log entries for high-impact actions (removals, suspensions, thread locks, precision reductions).
- Aggregated stats (“how many removals for X”) as a periodic transparency report.
- Clear “statement of reasons” for affected users + complaint/appeal route.
Coordination rules
- One voice: moderators don’t freestyle policy in comments; we link to the rule + ID.
- Bridge back to process: “Here’s the rule, here’s the appeal, here’s how to fix it.”
- No public pile-ons: enforcement happens quietly; explanations are public but not humiliating.
How this connects to the rest of the trust stack
Moderation is one leg of the table. The other legs: evidence standards, credibility policy, disclosures, corrections, privacy, and AI policy.
Evidence Standards
What counts as evidence, how provenance works, and how we label certainty and uncertainty.
Corrections
How we change our minds in public: corrections log, receipts, and what happens after an error is found.
Disclosures
How we label money, partnerships, and material relationships — so “trust” isn’t vibes-based.