# How Jogg Mini Builds High-Quality Questions

**Blog Article | MokingBird**

---

The most important thing in a quiz app isn't the animations or the level-up sounds. It's the questions.

A bad question confuses instead of educating. It tests vocabulary rather than understanding. It's ambiguous, unfair, or pitched at the wrong level. A bad question is worse than no question — it teaches children the wrong thing, or trains them to rely on guessing rather than genuine comprehension.

At Jogg Mini, questions aren't an afterthought. They're the product. Jogg Mini is a quiz-first AI learning app, which means the quality of its questions directly determines the quality of the learning experience. Here's how MokingBird builds them, quality-checks them, and continues to grow the question library as the platform scales.

---

## The Question Philosophy

Questions in Jogg Mini are designed to be:

- **Age-appropriate** — vocabulary, sentence length, and abstraction level calibrated to the learner's age band
- **Concept-first, not jargon-first** — teaching the idea before introducing technical terminology
- **Encouraging rather than discouraging** — wrong answers are learning opportunities, not punishments
- **Suitable for short, repeatable mobile sessions** — clear, focused, completable in under 30 seconds
- **Aligned to worlds and progression** — each question reinforces the mental model of its world
- **Useful for parents and teachers** — explanations should be readable by adults reviewing a child's progress

The goal is not just to ask trivia. The goal is to teach.

---

## The Challenge: One Subject, Ten Years of Learners

Building questions about AI for children is harder than it sounds. Not because AI is inherently complicated — but because the right way to explain a concept to a 6-year-old is completely different from the right way to explain it to a 14-year-old.

Consider "What is a neural network?"

**For a 7-year-old:**
"Imagine you taught a robot to recognise your face by showing it 1,000 photos of you. The robot doesn't memorise the photos — it learns what *you* look like. That's a bit like what a neural network does."

**For a 14-year-old:**
"A neural network is a computational model inspired by biological neurons — layers of interconnected nodes that transform input data through learned weights to produce an output."

Both are accurate. Neither is a dumbed-down version of the other. They're genuinely different formulations for genuinely different learners. Building a question system that handles this range is the core challenge Jogg Mini's question architecture is designed to solve.

---

## The Question Model

Every question in Jogg Mini is a rich data object with attributes that ensure it's delivered to the right learner, at the right time, at the right level.

### Core Content
- **Question text** — the question the child sees
- **Options** — answer choices (A/B/C/D for MCQ; True/False; multiple select)
- **Correct answer** — the right choice
- **Explanation** — a clear, age-appropriate explanation of why the correct answer is right (and often why the wrong answers are wrong)

### Classification
- **World (1–6)** — which learning world this question belongs to
- **Difficulty level** — Easy, Medium, Hard, Smart, or Genius
- **Question type** — Multiple Choice, True/False, Multiple Select
- **Topic layer** — the specific AI/ML concept area (Foundation, Data, Embeddings, Model Training, Applications, AI Safety)
- **Tags** — keyword tags for filtering

### Age and Grade Targeting
- **Min age / Max age** — the recommended age range
- **Grade levels** — specific grade level targeting within each education system
- **Grade system** — US (K-12), UK (Year 1-13), India (Class 1-12), or IB
- **Reading level** — Flesch-Kincaid grade level score

### Quality and Analytics
- **IRT difficulty** — Item Response Theory difficulty score (0.0–1.0), derived from actual performance data
- **Attempts count** and **correct count** — live performance metrics
- **Average time** — average seconds to answer, per question
- **Emoji used** — emoji in the question text (increase engagement for younger learners)

---

## The Topic Taxonomy

Jogg Mini's questions are organised around a six-layer topic taxonomy, designed for children's AI/ML education:

### Layer 0: Foundation
*What is AI? How does it differ from regular programs? What are robots?*

The most accessible layer — concrete language, familiar examples, minimal assumed knowledge. A 5-year-old starts here.

### Layer 1: Data
*What is data? How does AI learn from examples? What is training?*

Data is the fuel of modern AI. This layer covers what data is, how AI trains on it, the training/testing split, and the critical issue of data quality and bias.

### Layer 2: Embeddings and Patterns
*How does AI represent information? How does it find patterns?*

The bridge between data and models. How does an AI system go from raw data (pixels, words, numbers) to a meaningful internal representation it can reason about?

### Layer 3: Model Training
*How does AI learn? What is a model? How does training work?*

Inside the learning process. What happens when you show an AI system thousands of examples? How does accuracy improve? What is overfitting?

### Layer 4: Applications
*Where is AI used in the real world? How do familiar systems work?*

Connecting earlier concepts to real-world technology: voice assistants, recommendation engines, self-driving cars, medical AI.

### Layer 5: AI Safety and Ethics
*How should AI be used? What can go wrong? Who is responsible?*

Bias, fairness, privacy, surveillance, accountability, environmental impact. Always unlocked because these questions are relevant regardless of technical background.

---

## Age-Appropriate Content Design

Jogg Mini's question system adjusts the entire conceptual approach by age, not just the vocabulary:

### Ages 5–8
Questions should be:
- Concrete — grounded in direct experience
- Visual where possible
- Short in length — low word count
- Simple in vocabulary
- Focused on recognition and basic understanding

### Ages 9–12
Questions can introduce:
- Simple reasoning and cause-and-effect
- Categories and patterns
- Basic AI vocabulary with supporting context
- Multiple-step thinking

### Ages 13–15
Questions can engage with:
- Model behaviour and training logic
- Bias and fairness concepts
- Applied AI in specific industries
- Safety and responsible AI use
- Technical vocabulary used correctly

---

## Reading Level Calibration

One of the most important (and least visible) quality criteria for Jogg Mini questions is readability.

We use the **Flesch-Kincaid Grade Level formula** to score every question. This produces a number corresponding to a US grade level — a score of 3 means readable by a typical third-grader; a score of 9 means ninth-grade proficiency.

Jogg Mini reading level targets by difficulty:

| Age Range | Difficulty | Target FK Grade |
|-----------|-----------|----------------|
| 5–8 | Easy | Grade 1–3 |
| 8–12 | Medium | Grade 4–6 |
| 12–15 | Hard | Grade 7–9 |
| 12–15 | Smart/Genius | Grade 7–10 |

Questions that exceed their target reading level — even if conceptually appropriate — are flagged for revision. A 7-year-old shouldn't be blocked from an age-appropriate concept by a sentence structure that's too complex.

---

## What Makes a Question Good

A good Jogg Mini question has the following qualities:

**Clear learning target.** The question tests a specific concept, not a vague impression.

**Age-appropriate wording.** Vocabulary, sentence length, and abstraction match the learner's age band.

**Strong distractors.** Wrong answers are plausible enough to test understanding, but not confusing or misleading.

**Useful explanation.** The explanation corrects misunderstandings, makes abstract AI ideas easier to remember, and sounds supportive rather than punishing. For a children's app, explanations should feel encouraging.

**World alignment.** The question reinforces the mental model of the world it belongs to.

**Difficulty integrity.** An easy question should actually be easy. A hard question should challenge without becoming unclear.

---

## The MVP Question Library

Jogg Mini launches with **650+ carefully curated, educator-reviewed questions** across six worlds:

| World | Questions | Focus |
|-------|-----------|-------|
| Robot Valley | ~110 | Foundational AI concepts |
| Data Valley | ~125 | Data and learning |
| Pattern Mountain | ~130 | Patterns and models |
| Smart City | ~135 | AI applications |
| Future Lab | ~150 | Advanced AI |
| AI Safety & Ethics | ~40+ | Ethics and responsibility |

Every question in the MVP library has been:
1. Written by subject matter specialists with AI/ML backgrounds
2. Reviewed for age-appropriateness and reading level
3. Cross-checked for factual accuracy
4. Checked for ambiguity — no question has a defensible second correct answer
5. Tagged with complete metadata

This curated foundation ensures that every question a child sees has been through a rigorous process before reaching them.

---

## Scaling with MokingBird DataGen

650 questions is a strong MVP. But as Jogg Mini grows — more users, more grade systems, more advanced content — the platform needs a scalable generation pipeline.

The **MokingBird DataGen** system is a purpose-built educational question generation framework developed by MokingBird. It is not a general-purpose AI writing tool — it's a specialised pipeline designed to produce questions that meet every quality criterion in the Jogg Mini question model.

### How DataGen Works

**1. Specification Input**
DataGen receives a detailed specification: topic layer, target age, grade level, difficulty, question type, reading level target, required metadata, and specific concepts to cover or avoid.

**2. Generation**
Using large language models constrained by the Jogg Mini question schema, DataGen produces candidate questions — including question text, options, correct answer, and explanation.

**3. Automated Quality Checks**
Every generated question passes through automated checks:
- Flesch-Kincaid reading level scoring
- Factual accuracy verification
- Ambiguity detection
- Age-appropriateness screening
- Duplicate detection
- Safety screening (no harmful stereotypes, misleading AI claims, or inappropriate scenarios)

**4. Human Review Queue**
Questions passing automated checks enter a human review queue. Subject matter reviewers with AI/ML and education backgrounds evaluate each question before it enters the live bank.

**5. Live Performance Monitoring**
Once deployed, every question generates performance data:
- IRT difficulty scores update from real student performance
- Questions with unexpectedly high or low accuracy are flagged for review
- Average time statistics identify questions that may be too ambiguous or trivial
- Questions that consistently produce wrong answers at high rates are investigated

This compounding feedback loop means the question bank becomes more accurate and better calibrated over time as more children use the app.

---

## Question Generation and Child Safety

Because Jogg Mini teaches AI to children, question design must also be responsible. MokingBird's content review process specifically screens for:

- Harmful stereotypes or cultural bias
- Misleading claims about AI capabilities or limitations
- Manipulative framing
- Age-inappropriate scenarios
- Unsafe advice, especially in the AI Safety & Ethics world

This is especially important in World 6. Questions about AI ethics need to be accurate, balanced, and age-appropriate — not alarmist, not dismissive.

---

## Multi-System Grade Alignment

Jogg Mini supports four major education systems. Questions can be targeted to specific grade bands within each:

| System | Range |
|--------|-------|
| US | Kindergarten – Grade 12 |
| UK | Year 1 – Year 13 |
| India | Class 1 – Class 12 |
| IB | PYP – Diploma Programme |

When a parent sets their child's education system and grade in the profile, the question filtering system automatically surfaces questions calibrated to that child's curriculum level. A Class 7 student in India and a Grade 7 student in the US may encounter different questions — both correct, both appropriate — because the right calibration genuinely differs.

---

## The Explanation Is Part of the Question

Most quiz apps show whether you were right or wrong and move on. Jogg Mini treats the explanation as equally important as the question itself.

After every answer — correct or incorrect — the child sees:
- The correct answer (highlighted clearly)
- An age-appropriate explanation of why it's correct
- For wrong answers: brief context on why the most common wrong choices are incorrect

The explanation is the most powerful learning moment in the interaction. It happens at exactly the point of highest engagement — right after the child has committed to an answer. We invest heavily in ensuring it's clear, accurate, encouraging, and genuinely educational.

---

## How This Connects to the App Experience

Question quality directly powers everything in Jogg Mini:

- **World progression** — better questions produce more meaningful XP milestones
- **Quiz sessions** — question clarity determines whether practice feels fair
- **Daily challenges** — well-calibrated difficulty keeps daily engagement challenging but achievable
- **Teacher quizzes** — teachers depend on questions being reliable for classroom assessment
- **Parent reports** — accuracy metrics only mean something if question quality is consistent
- **Streak maintenance** — children stay engaged when questions feel fair and educational

Better questions create better learning, better motivation, and more trustworthy reporting for every user type.

---

*MokingBird — Jogg Mini. Teach AI. One question at a time.*