Methodology

Uses the Unconsenting Media flags and the Does the Dog Die warnings, and takes the more severe result.

Harmful content: Unconsenting Media flags rape on screen, rape off screen, attempted rape, child sexual abuse or incest, or a Does the Dog Die warning about rape, sexual assault, sexual abuse, molestation or incest has more yes votes than no votes.

Watch with caution: no harmful flag, but Unconsenting Media flags on-screen sexual harassment, sexual content between an adult and a teenager, or rape that is mentioned or implied.

Safer to watch: neither source flags sexual violence.

Uses up to four signals: the share of women in the cast, in the top five billed, and in the key crew departments (directing, writing, production, editing and camera), plus the Bechdel Test rating for films.

A share of 70% or more scores 2, 40% or more scores 1, lower scores 0. The Bechdel Test scores 2 for 3/3, 1 for 2/3 and 0 below. The scores are averaged: 1.75 or more is Strong, 1.0 or more is Mixed, lower is Poor. With fewer than two signals the verdict is Not enough data.

Sexual violence: Not shown, Present or Not sure. Present opens the 8 Unconsenting Media concern flags, each Yes, No or Not sure, so the counts line up with that source's entry. No level is derived from the flags.

Partner or family abuse, physical violence or murder and stalking or harassment: Not shown, Mentioned or implied, On screen or Not sure.

Sexual objectification: Not present, In some scenes, Throughout or Not sure. It means the camera dwelling on women's bodies, or women written only as sexual objects. It is not an Unconsenting Media flag.

An optional question on how the title treats the violence is asked when sexual violence is Present, or when another violence category is answered Mentioned or implied or On screen. Its counts cover the reports that answered it.

A signed-in account that is not muted or banned can report a film or series once, after confirming it watched the title, and can edit or delete the report. Not sure answers are counted and shown as such. The concern flag counts cover the reports that answered Present.

A report counts once its account is 7 days old; reports from moderators and administrators count at once and are never held. A title shows counts once 3 reports count; below that the page shows the number of reports only. Counts are cached briefly, so a new report or a lapsed wait can take a short time to show. A muted or banned account's reports are not counted while the mute or ban lasts, and a report a moderator rejected is not counted unless a moderator counts it later.

Three rules hold a report for a moderator instead of counting it. They run on every save, an edit included, and the first that applies is recorded. The author sees the report as waiting for review.

Contradicts Unconsenting Media: the title has an Unconsenting Media entry with at least one concern flag, and the report answers Not shown for sexual violence or No to a flag that entry marks. Not sure never contradicts.

Burst: 5 or more reports on one title, this one included, fall within 24 hours. A report made before that window counts only the reports inside it.

Busy new account: the account is under 30 days old and has more than 10 reports, this one included.

A moderator counts a held or rejected report, or rejects a held or counted one, and can reject every report from an account. An edit does not reinstate a rejected report.

A signed-in account that is not muted can mark a title's counts with Counts look wrong, with an optional note. An account has one per title; sending it again replaces the note and reopens it. Moderators see these titles with every report on the title. They can reject or count a report, reject all reports from an account, or mute or ban the account, then resolve or dismiss each Counts look wrong. Resolving one does not change the counts.

A short blocklist of words and phrases runs first. A match rejects the comment with no further check. The site also keeps a private list that extends it.

Otherwise Claude Haiku 5.5, a model from Anthropic, reads the comment. Describing, quoting, analysing and criticising misogyny, sexual violence or other harm in a film or series is allowed, including a reader's own experience of it, and so are disagreement, harsh criticism of a film and strong language. Endorsing or celebrating misogyny or sexual violence, mocking or doubting survivors, attacking, insulting or harassing a person or a group, slurs, threats, sexual content involving minors, explicit sexual content, glorifying or calling for violence, demeaning a person or a group without a slur, bad-faith bait meant to provoke, sharing personal information, and spam, advertising or links with no relation to the title are not. The model answers with a decision, the categories that apply, a severity from 0 to 3 and a reason.

Anthropic receives the comment text, the title's name and, for a reply, the comment it answers. It never receives the author's email or account.

Published: the model allows the comment. It appears at once.

Awaiting review: the model asks for a person to look, or finds a clear break below severity 2. The same applies when the check gives no verdict, or when the site has no key for it. Only the author and moderators see the comment until a moderator decides.

Rejected: the blocklist matches, or the model finds a clear break of severity 2 or 3. The author sees the comment with the reason.

Moderators approve, hide or reject comments from a queue. Three different readers reporting a published comment send it back to review. A moderator can also mute or ban an account.