1. Confirm the creator
Every record resolves to YouTube’s canonical channel ID. Names, handles, and URLs are aliases—not separate creators. Parent requests are deduplicated, rate-limited, and never create an indexable page automatically.
2. Select videos separately from selecting creators
The V1 launch cohort contains 75 high-interest creator candidates. That creator count is separate from each channel’s evidence sample. A standard sample contains at least six regular videos: three recent and three popular from the fetched catalog. If findings conflict or transcript coverage is weak, the sample expands up to twelve standard videos. Three Shorts are added when Shorts form a material part of the available channel catalog. Livestreams are outside V1.
3. Store the full transcript evidence
For this checker’s small sample, ScreenSignals stores the full available transcript plus timestamped segments, language, source, and a content fingerprint. That differs from the private high-volume viewing-history pipeline, where hundreds of videos make bounded storage important for speed and cost.
4. Classify age-neutral observations
A pinned production model classifies transcript evidence for language, violence and intensity, sexual or suggestive content, substances and risky behavior, harassment and role modeling, fear and disturbing themes, and commercial pressure. It also records positive, educational, and engagement patterns. Transcript text, titles, and descriptions are explicitly treated as untrusted data and cannot give instructions to the model.
5. Derive four age views
The same stored observations produce scores for ages 5–7, 8–10, 11–13, and 14–17. Younger age bands apply greater sensitivity to relevant concerns. The 8–10 view is shown by default, but it is not a universal recommendation. ScreenSignals is for adult parents and legal guardians; children do not operate the checker or hold accounts.
6. Explain scores with evidence and uncertainty
The overall score and seven category scores use deterministic, versioned rules. A serious, high-confidence pattern caps the overall score. Confidence depends on the number of standard videos and usable transcripts. Each page shows sample size, analysis date, video titles, paraphrased findings, and limitations. ScreenSignals does not claim to assess visuals it has not seen.
7. Review before publication and refresh after it
New pages are stable but unlisted and marked noindex. An administrator can publish only after the quality floor is met and all four age views exist. Administrators can hide, block, retry, refresh, or permanently delete only never-published mistakes. Published pages enter a refresh queue and preserve their prior public revision until a replacement passes review.