Kids deserve safe wonder.
So we built a way to measure it.

How the score works

Disclaimer: A low score does not mean a title belongs in the trash. Passionate people put their hearts into making it. But as with food, it helps to understand what your kids consume, so you can keep their diet varied.

Methodology

Grid version 0.5, written for the US market. Every title gets a score out of 100 and a letter grade: the average of five criteria worth 20 points each.

Disclaimer: A low score does not mean a title belongs in the trash. Passionate people put their hearts into making it. But as with food, it helps to understand what your kids consume, so you can keep their diet varied.

The 20 questions

Each question is rated from 0 to 3, and every rating must cite a concrete feature of the title. For the first two criteria, 0 means absent and 3 means at the core of the title. For the three protection criteria the scale is reversed: 3 means no problem at all.

Curiosity & Creativity

1. Does the title raise questions without answering them right away, and leave room for mystery?
2. Does the child lead: real choices, open-ended exploration, room to try, fail and revise, with no single prescribed path?
3. Does the child produce something that did not exist before, so that two children end up with different results?
4. Does it carry on outside the title: looking things up, asking an adult, acting it out, drawing, inventing what comes next?

Learning & Development

1. Does the title build real knowledge or a real skill, with language slightly above the child's level?
2. Is it minds-on: does it ask for thinking and productive effort, not just tapping or watching, with a challenge that adjusts?
3. Is it meaningful and social: does it connect to the child's own life and invite talking or playing with an adult or other children?
4. Does it model emotional skills: nuanced feelings, conflicts resolved, empathy?

Respect for attention

1. Does it stop by itself: a natural ending, pacing that leaves time to understand, no autoplay or endless feed switched on by default?
2. Is it free of mechanics that pull the child back: streaks, daily rewards, notifications, characters pleading to stay?
3. Is it free of purchase pressure: loot boxes, timed offers, in-app purchases pushed at the child?
4. Is it free of advertising, including ads disguised as content or targeted at the child?

Online Safety

1. By default, is the child unable to be contacted by strangers?
2. Is it private by default: minimal data collection, no ad tracking or profiling, location off, real parental controls?
3. Is the child kept away from unvetted content (outbound links, unmoderated user content, algorithmic recommendation feeds), and can the child report a problem easily?
4. AI safety: if the title includes an AI chatbot, companion or generative feature, does it behave safely with children (age-appropriate answers, says it is an AI, no flattery that builds emotional dependency, no romantic role-play, no harmful advice, points to a trusted adult when it matters)? A title with no AI feature gets full marks on this question.

Fully offline formats, such as books and board games, get full marks here.

Physical & sexual violence

1. Is the title free of graphic or realistic violence and gore?
2. When violence appears, does it have consequences, without being rewarded or glamorized?
3. Is it free of sexual content and of sexualized characters? Any sexual violence scores zero on the whole criterion.
4. Is it free of unresolved cruelty or bullying, and of content too frightening for its age audience?

From answers to a score

Each criterion adds up its four answers (0 to 12) and converts the sum to a score out of 100. The final score is the average of the five criteria.

Caps

• Online safety or physical & sexual violence below 50: final score capped at 39.
• Respect for attention below 33: final score capped at 59.

What a score is today, and what comes next

• Today: preliminary. An AI applies the grid from what it knows about the title. The first score given to a title is saved, so everyone sees the same one.
• Next: product checks. Yes-or-no facts anyone can verify, such as whether autoplay is off by default or whether there is open chat. They will count for half of each criterion.
• Next: expert-verified. A named child-development expert who has watched, read or played the title signs the review. Upvotes decide which titles are reviewed first.

Known limits

• A preliminary score can be wrong or out of date, especially for recent or little-known titles.
• Tier lists show the ten titles the AI considers most popular in a category, not a ranking based on download or audience data.
• Three of the five criteria are about protection, so a harmless but empty title can still earn a middling score.

References

Two projects shaped the grid itself.

• KORA Benchmark and its Apps benchmark. An independent, open-source benchmark of how AI models and apps behave with children, across 26 child-specific risks. We borrow its categories, its percentage with a letter grade, and its AI safety tests for our Online Safety criterion.
• Open Children's Media Index. A transparent database of measurable features of children's television, such as pacing, flashing and language. We borrow its rule that every score shows its component parts.

The evidence behind each criterion

Grid v0.5 was revised against the research below. Each question traces back to at least one of these sources.

Curiosity & Creativity

• Zosh et al. (2017), Learning through play, LEGO Foundation. Playful learning is joyful, meaningful, actively engaging, iterative and socially interactive. Hence our questions on trying, failing and revising.
• UNICEF Innocenti, Responsible Innovation in Technology for Children (2022, 2024). Eight wellbeing outcomes for digital play, including autonomy, competence and creativity. Hence the question on whether the child leads.
• Blanco-Herrera, Gentile & Rokkum (2019), Video Games Can Increase Creativity, but with Caveats, Creativity Research Journal. Free play in Minecraft raised creativity scores; being told to "be creative" did not. Hence we reward open-ended play over guided creative tasks.

Learning & Development

• Hirsh-Pasek, Zosh, Golinkoff, Gray, Robb & Kaufman (2015), Putting Education in "Educational" Apps, Psychological Science in the Public Interest. Four pillars of learning: active, engaged, meaningful, socially interactive. Our four questions follow them.
• Meyer et al. (2021), Journal of Children and Media. The most-downloaded "educational" apps were rated 0 to 3 on each pillar, and most scored low. This is the precedent for our 0 to 3 scale, and the reason an "educational" label earns no points by itself.
• Takeuchi & Stevens (2011), The New Coviewing, Joan Ganz Cooney Center. Children learn more from media built to be used with an adult. Hence the question on shared use.

Respect for attention

• Radesky et al. (2022), Prevalence and Characteristics of Manipulative Design in Mobile Applications Used by Children, JAMA Network Open. About four in five apps used by children aged 3 to 5 contained manipulative design. Our four questions follow its four types: prolonging play, pulling the child back, purchase pressure, advertising.
• Meyer et al. (2019), Advertising in Young Children's Apps, Journal of Developmental & Behavioral Pediatrics. 95% of 135 apps for children under five contained advertising, often disguised as play.
• American Academy of Pediatrics (2026), Digital Ecosystems, Children, and Adolescents. Pediatric guidance has moved from time limits to design: autoplay and algorithmic feeds off by default, no targeted advertising to minors.

Online Safety

• UK Information Commissioner's Office, Age Appropriate Design Code (2021). Fifteen standards, including high-privacy defaults, location and profiling off. Hence we score what is on by default, not what a parent could switch off.
• OECD (2021), Recommendation on Children in the Digital Environment. Safety by design and transparency a child can understand.
• 5Rights Foundation (2021), Pathways: How digital design puts children at risk. Child accounts were shown harmful content and contacted by adult strangers within a day. Hence recommendation feeds and open contact count as risks.
• UNICEF (2025), Guidance on AI and children, version 3, and the KORA Benchmark. Together they shape the AI safety question.

Physical & sexual violence

• Common Sense Media ratings and the PEGI and ESRB age systems grade violence by context: realistic or cartoon, graphic detail, consequences shown, and how frightening it is. We follow the same logic.

Where the evidence is weak

• Violence and aggression. Whether violent media makes children aggressive is disputed. A meta-analysis of 101 studies found a very small effect (Ferguson, 2015), and a pre-registered study of 2,008 teenagers found none (Przybylski & Weinstein, 2019). Our violence criterion measures whether content suits a child's age and may frighten or distress. It makes no claim about behavior.
• Fast pacing. One small experiment found that nine minutes of a fast cartoon lowered four-year-olds' self-control right afterward (Lillard & Peterson, 2011). A re-analysis of the best-known study linking early television to attention problems found the result fragile (McBee, Brand & Dixon, 2021). Pacing is therefore only part of one question.
• Screen time. Hours of use predict later development only very weakly (Madigan et al., 2019). We score what a title is and how it is designed, never how long a child uses it.
• One grid for all ages. The same feature does not carry the same risk at 3, 6 and 10. Age bands are the next revision.

Sources we plan to check for each title

Today's preliminary scores do not use these sources yet. They are the outside evidence we plan to check for each title, grouped by the criterion they inform.

Independent reviews (all criteria)

• Common Sense Media reviews and age ratings.
• Association for Library Service to Children notable lists, chosen by children's librarians.

Educator recognition (Learning & Development)

• CODiE Awards, Tech & Learning awards, Teachers' Choice Awards (Learning Magazine), International Serious Play Awards.
• Trade press that reviews classroom products: District Administration, eSchool News.

Proof that it teaches (Learning & Development)

• ESSA evidence tiers: whether a product's learning claims rest on a study, and how strong that study is.
• Independent efficacy studies, such as those run by SEG Research.

Privacy and safety certifications (Online Safety)

• COPPA, and the FTC-approved Safe Harbor programs: kidSAFE, iKeepSafe, PRIVO.
• 1EdTech TrustEd Apps: a privacy label granted after a review against a public rubric.
• Student Privacy Pledge (Future of Privacy Forum): a public commitment by the publisher.

An award or a certification is evidence, not a score. Each one still has to answer one of the 20 questions.

About us

As parents, we all look for entertainment that helps our kids grow, not just keeps them busy.

Screens, games and shows are part of childhood. Today's ratings tell parents whether something is appropriate for an age. They do not say whether it helps a child grow. The Marble Index exists to measure that, with a public grid anyone can check.

The index is built and funded by Marble, the team behind an interactive game made to nurture kids' curiosity. It carries our name, so we hold it to a public standard: Marble's own product is not scored by the team.

• Not for sale. No publisher can pay for a score or for a review.
• Open. The grid is public and versioned, and every change is dated and explained.
• Accountable. Publishers can contest a score with facts, and the response is published.

We are assembling a panel of child-development experts to review scores. To take part or to contest a score, write to support@withmarble.com.