Calmness
Gameplay recordings measure motion, visual change, brightness variation, audio intensity and frequency. We also record reward activity and attention prompts. The public score reverses attention intensity: 10 means calmer.
Ratings · Methodology v0.1
How we test kids apps
A repeatable framework built from recorded gameplay, a structured review, automated measurements, and human confirmation. Every public score runs from 0 to 10; higher is better.
- versioned
- evidence-based
- five scores
A score starts with two records: clean child-facing gameplay with system audio, and a guided review covering access, pricing, offline use, permissions, rewards, and representative learning activities.
The testing process
- Sample representative play. We aim for 60–120 seconds per materially different activity and at least three activity types when they exist.
- Measure the recording. Deterministic video and audio analysis produces raw metrics; the structured review supplies facts that cannot be read reliably from pixels alone.
- Confirm the observations. Draft findings are checked and corrected by a human before scores become canonical.
- Calculate versioned scores. Raw evidence, normalized components, formula version, and final scores remain separate so a future calibration can be rerun without re-recording.
The five public scores
Travel
We test offline function and reliability, preparation burden, dependence on audio, commercial interruptions, and how much content remains accessible.
Content value
We combine content breadth, depth, meaningful variety, and the amount of useful content available without payment.
Learning value
Representative activities are scored for skill demand, agency, depth, and transferability. This measures the interaction—not educational marketing—and does not claim proven learning outcomes.
Commercial experience
We document advertising, purchase pressure in the child flow, visible locked content, pricing transparency, and purchase-model friction.
Exact score formulas
Calmness
10 − Attention Intensity| Component | Weight |
|---|---|
| Visual motion | 25% |
| Visual change frequency | 15% |
| Brightness variability | 10% |
| Audio intensity | 15% |
| Audio event frequency | 10% |
| Reward activity | 15% |
| Attention prompts / interruptions | 10% |
Travel
Σ(normalized component × weight)| Component | Weight |
|---|---|
| Offline functionality | 45% |
| Offline reliability | 10% |
| Offline preparation burden | 15% |
| Audio independence | 10% |
| Child-flow commercial interruptions | 10% |
| Accessible content breadth | 10% |
Content value
Σ(normalized component × weight)| Component | Weight |
|---|---|
| Content breadth | 35% |
| Content depth | 25% |
| Meaningful variety | 20% |
| Free-access value | 20% |
Learning value
Per activity: (skill demand + agency + depth + transferability) ÷ 8 × 10. App: mean weighted by accessible-experience share.| Component | Weight |
|---|---|
| Skill demand | 25% |
| Agency | 25% |
| Depth | 25% |
| Transferability | 25% |
Commercial experience
Σ(normalized component × weight)| Component | Weight |
|---|---|
| Third-party advertising | 25% |
| Child-flow purchase pressure | 25% |
| Locked-content exposure | 15% |
| Pricing transparency | 20% |
| Purchase-model friction | 15% |
Measurement and normalization
Capture parameters
| Video sampling | 2 frames/second |
|---|---|
| Analysis width | 320 px |
| Changed-pixel luma threshold | 24 |
| Motion magnitude threshold | 1.25 |
| Brightness-change threshold | 18.0 |
| Large visual-change fraction | 0.28 |
| Silence threshold | −45 dB |
| Audio-event threshold | −22 dB |
| Audio outlier margin | 8 dB |
| Audio-event window | 0.5 seconds |
| Evidence screenshots | 3 per category |
Raw metric → 0–10
| Metric | Good | Poor | Meaning |
|---|---|---|---|
| Visual motion | ≤ 0.015 | ≥ 0.18 | Mean optical-flow motion fraction |
| Visual changes | ≤ 2/min | ≥ 18/min | Large frame-change events |
| Brightness variability | ≤ 18 | ≥ 65 | Luminance standard deviation |
| Audio intensity | ≤ −28 LUFS | ≥ −12 LUFS | Less negative is louder |
| Audio events | ≤ 4/min | ≥ 35/min | High-energy audio events |
Values between the good and poor anchors are linearly normalized. These thresholds are provisional for the first ten-app calibration pass.
Observation rubric
How observed answers become component values
| Component | Mapping to 0–10 |
|---|---|
| Reward activity | none 0 · light 3 · moderate 6 · heavy 10 |
| Attention prompts | no 0 · yes 8 |
| Offline functionality | yes 10 · partial/unreliable 4 · no 0 |
| Offline reliability | reliable 10 · sometimes fails 5 · always fails 0 |
| Offline preparation burden | no 10 · partial 4 · yes 2 |
| Audio independence | yes 10 · no 2 |
| Commercial interruptions | no 10 · yes 1 |
| Third-party advertising | no 10 · yes 0 |
| Purchase pressure | no 10 · yes 1 |
| Locked-content exposure | no 10 · yes 3 |
| Pricing transparency | yes 10 · partially hidden option 5 · no 2 |
| Purchase model | one-time/lifetime 10 · subscription 5 · hidden lifetime/subscription front 4 |
| Free access | genuinely free 10 · meaningful tier 6 · limited demo 4 · trial 3 · effectively paid 1 |
| Meaningful variety | low 3 · moderate 6 · high 8 · excellent 10 |
For accessible content breadth, 0 items maps to 0 and 12+ maps to 10. Content breadth maps from 1 to 8 activity types; content depth from 3 to 40 items. Unknown always remains null and makes the affected score incomplete.
What the scores do not say
The framework does not label an app addictive, non-addictive, globally safe, unsafe, developmentally beneficial, or proven to produce learning outcomes. Permissions and privacy observations are factual indicators, not a single safety score.
Missing evidence stays unknown. It is never silently converted into “no” or a zero. An incomplete score remains incomplete until the required evidence exists.
Denny's Maze
The app I built around these priorities
I built Denny's Maze to be optimized for calmness, reliable offline play, and educational value through focused maze-solving. It is my practical attempt to apply the same principles this methodology measures.