PM Instructions — Dr. Math Content Review System

🎯 Your Role as a PM Reviewer

You are the bridge between pedagogy and product. Every rating you give directly influences:

Which prompt persona gets prioritized for future content generation
Which questions make it into the app for children to solve
How we iterate on the LLM prompts to improve quality
What content gaps exist across topics and difficulty levels

💡 Core Principle

When you review, put yourself in the shoes of a 12-year-old Indian child. Would you want to solve this? Would it make you curious? Would it help you think — not just memorize?

🎨 The 4 Pedagogical Dimensions

Each topic is generated in 4 versions, each based on a different learning science theory. Your job is to compare them and identify which approach works best for each concept.

🧮 Default (Anti-Gravity)

Theory: Scaffolded instruction (Rosenshine, 2012)

"I understand step by step. The explanation makes sense."

Clear explanations, progressive difficulty, direct instruction. Best for introducing new concepts.

📖 Cultural Storyteller

Theory: Situated learning (Lave & Wenger, 1991)

"This is MY world — cricket, bazaar, festivals, train journeys."

Math embedded in Indian cultural contexts. Every problem is a mini-story. Best for engagement and relevance.

👁️ Visual-First Thinker

Theory: CPA method — Concrete → Pictorial → Abstract (Bruner, 1966)

"I can SEE it. I can draw it. The diagram helps me understand."

Heavy emphasis on diagrams, visual intuition, spatial reasoning. Best for geometry, fractions, and abstract concepts.

🎯 ZPD Adaptive Tutor

Theory: Zone of Proximal Development (Vygotsky, 1978)

"This is hard but not impossible. I can figure it out with a small push."

Just-harder-than-comfortable problems with scaffolding. Best for building resilience and deep understanding.

⭐ The 4 Rating Dimensions

For each question, you rate on a 1-5 scale across these research-backed dimensions:

🧠 Thought Direction Does it guide the child toward the right way of thinking — or just test memorization?

🎭 Playfulness Would a child want to solve this? Is it engaging, fun, curiosity-building?

🪜 Guidance Quality Is the explanation helpful — or does it just give the answer? Good scaffolding?

🔥 Curiosity Building Does it make the child want to learn more about this topic?

Rating Scale

1 ⭐ — Poor. Unusable. Confusing or counter-productive.
2 ⭐ — Weak. Major flaws. Needs significant rewrite.
3 ⭐ — Acceptable. Works but not inspiring. Minor issues.
4 ⭐ — Good. Effective learning. Child would engage.
5 ⭐ — Excellent. Delightful. Sparks curiosity and deep thinking.

⚠️ Hierarchical Rule (from research)

If a question is unclear or has an incorrect answer, it fails the gate — rate it 1-2 on all dimensions and leave a note. Clarity and accuracy are prerequisites for everything else.

🔄 Your Review Workflow

1

Pick a Topic

Start with / (home). Click "📝 Review" on any topic card. We recommend starting with Integers or Fractions as baseline topics.

2

Choose a Dimension

On the review page, use the dimension switcher at the top. Compare the same question across Default, Storyteller, Visual, and ZPD. Which version feels best?

3

Rate Each Question

Click the star buttons for each of the 4 dimensions. Be consistent — if you give Question 1 a 4 on Playfulness, what makes Question 5 a 5? Add notes to explain your reasoning.

4

Save & Move On

Click "Save Review" for each question. The page auto-updates the progress bar. Filter "Unreviewed" to find what's left.

5

Check the Leaderboard

Visit /lab to see which prompt dimensions score highest across all topics. This tells us which pedagogical approach to prioritize.

📊 How Your Feedback Impacts the Pipeline

Per-question ratings are stored in the question_reviews database table, linked to the exact generation.
Aggregate stats (avg per dimension) are computed automatically and displayed on the review page header.
Prompt leaderboard at /lab ranks which prompt persona produces the highest-rated content.
Low-rated questions (avg < 3) are flagged for re-generation with improved prompts.
High-rated questions (avg ≥ 4) are prioritized for inclusion in the Flutter app.
Your notes are read by the engineering team to refine prompt engineering and fix systematic issues.

🔄 Feedback Loop

Your ratings → Prompt optimization → Better generation → Higher child engagement → Better learning outcomes. You are the quality gate.

🧪 Pages You Should Know

/ — Home — All topics. Start here.
/review/{topic} — Review — Rate questions for a topic.
/review/{topic}?prompt=storyteller — Review Specific Dimension — View only the Cultural Storyteller version.
/status — Generation Status — See what's generated, what's pending.
/lab — Manager Lab — Compare prompts, see leaderboard.
/compare?a=1&b=2 — Side-by-Side — Compare two generations directly.

❓ FAQ

Why are there 4 versions of the same topic?

We don't know which pedagogical approach works best for Indian Class VII students. By generating the same topic 4 ways, we can A/B test them via your ratings. The winning dimension gets scaled.

What if I disagree with another PM's rating?

That's valuable signal! Divergent ratings tell us a question is polarizing — some children will love it, others won't. We use variance, not just mean, to evaluate.

How many questions should I review?

Minimum: 5 questions per topic per dimension. Ideal: all 40. Realistic: review the first 10 of each dimension for 2-3 topics. Quality over quantity.

What if the correct answer is wrong?

Rate it 1 ⭐ on all dimensions and leave a note: "Answer is incorrect — should be X." This is critical feedback. Accuracy is our highest priority.

Can I change my rating later?

Yes. Click the stars again and re-save. The database stores the latest rating for each question.

📖 PM Instructions