Content Safety

How submitted and crawled links are screened. We name what we screen for; we don't publish the thresholds, since that would be a bypass guide.

Hard-Blocked at Submit

Rejected outright. No link record is created and the submitter sees a generic error.

• Sanctioned-entity domains. Hosts on the US Treasury OFAC Specially Designated Nationals list cannot be hosted here as a matter of law.
• Malware and phishing. Hostnames and URLs with an existing malicious verdict from a third-party reputation service.
• Known adult and high-harm domains. Submissions whose host appears on a curated, regularly-refreshed list of pornography, malware, warez, and similar domains. This catches the known long tail; it does not catch a new or obscure adult site, and it reads the domain, not the page — so a porn link on a clean-looking host can still get through. We don't index adult content; what slips past here is removed when noticed.
• CSAM. Multiple independent signals; a hit on any one blocks. See the reporting obligation below.

Flagged for Review

Accepted into the database but hidden from public surfaces until a moderator clears them. Signals include:

• Lower-confidence reputation signals from the same reputation service.
• Admin-managed keyword patterns for scams, doxxing, and abuse.
• Inference that the link belongs to a small set of safety-sensitive topics. The list of sensitive topics is in source code; the inference itself is not.
• A family-filtering DNS resolver that flags the host as adult. Family DNS over-blocks (it routinely catches art, medical, and fitness sites), so an adult flag from it routes to a moderator rather than blocking the submission.
• Curated domain lists for lower-stakes-but-unwanted categories — drugs, gambling, aggressive content, and similar. A host on one of these is held for review rather than rejected outright.
• An AI moderation classifier on the submission's title, description, and curator notes. Abuse-adjacent categories flag; sexual/minors blocks. See below.

The AI Moderation Exception

The user-facing rule is "no AI deciding what you see." The ranker is a published formula, descriptions are written by curators, no page on the reading side is generated. The one exception is the moderation classifier above. It runs server-side, produces a category verdict, and a moderator clears or confirms every flag before anything goes live. We took the dependency because under-flagging CSAM-tier material costs more than the dependency does.

CSAM Reporting

Under 18 USC § 2258A, providers with actual knowledge of child sexual abuse material must report to the NCMEC CyberTipline. We do. The internal runbook covers what gets sent.

Reporting a Link That Got Through

Use the Report button on the link's detail page. Screening is mechanical; curatorial judgement is what catches what screening misses.