Rating methodology

Updated October 12, 2026

Our goal is to show both content cautions and positive qualities. Ratings describe prominence or intensity; they do not label a person, identity, belief or topic as inherently good or bad.

What the scores mean

• 0: best assessment is not present in the identified work.
• 1: slight or brief.
• 2: light or occasional.
• 3: moderate or clearly present.
• 4: strong or repeated.
• 5: central, very strong or pervasive.
• Unknown: available information does not support a defensible score. Unknown is not zero.

Sources

Depending on the title, evidence may include the full public-domain text, publisher descriptions, publisher or teacher discussion guides, Wikipedia plot information, Google Books, Open Library, Project Gutenberg, TMDB, official certification explanations, reader reports and a reviewer’s direct reading or viewing. Licensed or restricted third-party review services are excluded from collection and evidence.

AI-assisted ratings

For eligible books, independent models assess the same 18-category rubric. The normal sequence is Luna and Haiku first; Gemini is added when they disagree, confidence is low or one provider fails; DeepSeek may serve as a final independent tie-breaker; and Luna may perform a bounded adjudication of categories still unresolved. Providers do not automatically become correct by agreeing.

Automatic publication

A title is eligible for automatic publication only when every required category has a numeric score and the configured source-check and confidence rules pass. Identity uncertainty, missing categories, low confidence, or a failed check sends the title to a held queue rather than treating missing information as zero.

Labels shown on the site

• Full work checked: a reviewer reports reading or watching the complete identified work.
• Source checked: ratings are supported by identified source material but may not cover the complete work.
• Source checked: automated ratings passed the required source and consistency checks before publication.
• AI-guessed: an unverified preview shown only when no checked catalogue assessment is available.

Important limitations

Different editions and adaptations may vary. Summaries can omit brief content, model knowledge can be wrong, and evidence may describe presence more reliably than absence. Scores reflect the rubric and available evidence, not a guarantee that every reader will respond the same way.

Corrections

Every title panel includes a “Report an incorrect rating” link. Reports should identify the exact title, category, proposed correction and evidence. Reports enter review; they do not immediately change a published score.

Commercial independence

Affiliate relationships do not determine category scores or publication decisions. Retail links are added after the assessment and are disclosed separately.