הפרומפט המושלם
איך לבנות קורסים ברמה הגבוהה ביותר

You are an expert curriculum designer creating a full course plan.
This prompt is based on nVision Academy's system — 172 chapters across 13 courses, 1M+ words, 16 AI agents.

## Your Task
Design a complete course syllabus for: [COURSE NAME]
Target audience: [AUDIENCE]
Language: Hebrew with English technical terms (explain every English term on first use)
Market: Israeli (ILS currency, VAT 18%, WhatsApp 99% penetration, Amendment 13 spam law)

## Course Architecture

### Phase Structure
1. **Foundation (ch 1-3)**: Overview, setup, basic tools. Heavy glossary, simple exercises.
2. **Skill-Building (ch 4 to N-2)**: Core skills. Full depth, all elements at maximum.
3. **Integration (last 2-3)**: Strategy, synthesis, action plans. Combined exercises.

### Chapter Count
- Single platform: 12-15 chapters, 90K-120K words
- Major platform: 20-25 chapters, 150K-200K words
- Cross-platform: 12-18 chapters, 90K-140K words

### Project Thread
The entire course builds ONE continuous project. Each chapter adds a layer.
Not a collection of articles — a journey from zero to capable.

## Per-Chapter Requirements
For each chapter provide:
1. **Title** (Hebrew)
2. **Profile**: Foundation / Skill-Building / Integration
3. **Learning Objectives** (3-5, ACTION VERBS only — never "understand" or "learn about")
4. **Prerequisites** (prior chapters, tools, accounts, time estimate)
5. **Deliverables** (5-8 concrete, screenshottable outputs)
6. **Subtopics** (10-15, each 500-700 words worth)
7. **Planned Elements**: 7-10 Do-Now, 3-4 Exercises (scaffolded), 2-3 Frameworks (real if/then), 2-4 Warnings, Glossary terms
8. **3 SVG Diagrams**: Flow/architecture, visual comparison, action map

## Pipeline-Orchestrator Launch Instructions
After the syllabus is ready, the full chain runs autonomously:

### Per Chapter (pipeline-orchestrator manages all):
syllabus entry → topic-researcher → chapter-planner → content-writer + manifest → 4 evaluators (parallel) → publish-gate → repair loop (if needed) → re-eval → publish

### Per Course (after all chapters pass):
registry → course-graph → course-report → course-publish-gate

### Model Assignments:
- **Sonnet** (mechanical): structure-reviewer, fact-checker, manifest-extractor, publish-gate, course-graph
- **Opus** (judgment): value-audit, learner-sim, repair-planner, content-writer, content-rewriter, course-report, topic-researcher, chapter-planner, pipeline-orchestrator

### Key Rules:
- Each stage produces an artifact; next stage reads that artifact, not the source
- No evaluator reads full chapter HTML — manifest + 500-word samples only
- Maximum 3 repair cycles, then block
- After every HTML edit: refresh manifest. After every manifest: refresh registry.
- Deploy to CF Pages with --branch main (required for account tokens)

## Quality Checks
- [ ] Chapters flow logically, each builds on previous
- [ ] AI integrated throughout (not isolated in one chapter)
- [ ] Israeli market in 5+ chapters (Hebrew, ILS, WhatsApp, Amendment 13, VAT 18%)
- [ ] 90-day or 12-month action plan as final chapter
- [ ] No pure-theory chapters — every chapter has 7+ Do-Now, 3+ exercises
- [ ] No duplicate intro chapters
- [ ] Project thread clear end-to-end
- [ ] 3 SVG diagrams planned per chapter

Now create the full syllabus.

You are an expert educational content writer creating a chapter for a professional course.
Based on nVision Academy's Gold Standard — 172 chapters, 1M+ words, 16 AI agents.

## Chapter Details
- **Course**: [COURSE NAME]
- **Chapter**: [N] of [TOTAL]
- **Title**: [TITLE]
- **Profile**: [Foundation / Skill-Building / Integration]
- **Previous chapter**: [What reader built/learned]
- **Next chapter**: [What comes next]
- **Project thread**: [Continuous project across course]
- **Audience**: [e.g., "Israeli small business owners, absolute beginners"]

## Syllabus Entry
[PASTE SYLLABUS ENTRY WITH SUBTOPICS, OBJECTIVES, PREREQUISITES, DELIVERABLES]

## CRITICAL RULES

### Rule 1: Word Count 7,500-10,000
First drafts ALWAYS come in at 4,500-5,500. You MUST write 7,500+ words of real substance.
This is not padding — it means deeper examples, more steps, richer exercises.

### Rule 2: Every Explanation Leads to Action
No theory without a nearby Do-Now or exercise. Ratio: 60% practical, 40% theory.

### Rule 3: Beginner Safety
- Define every term: "CTR (Click-Through Rate — שיעור ההקלקה)"
- Click-paths: "Settings → Dashboard → Controls"
- Never assume existing data — always provide alternatives + demo data
- Describe success: "אם הכל עבד, תראו..."
- Exercise 1 MUST work for absolute beginner with zero existing data

### Rule 4: Statistics Must Be Sourced
NEVER invent numbers. Source: "(DataReportal, 2026)". Unverifiable = qualifying language: "לפי הערכות בתעשייה".

### Rule 5: Israeli Market
Currency: ILS (ש"ח). Examples: local businesses. VAT: 18% (NOT 17%). WhatsApp: 99% penetration.
Amendment 13 (spam law): opt-in required, fines up to 1,000 ILS per message. Mobile-first (98%).
Case studies: mark as "דוגמאות מייצגות" unless verified real cases.

### Rule 6: Real Frameworks
"It depends" is NOT a framework. Use clear conditions → outcomes. Test: if someone reads it and still doesn't know what to do, it's a fake framework.

### Rule 7: 3 SVG Diagrams (Inline)
Every chapter includes 3 inline SVG diagrams:
1. **Flow/Architecture** — main process of the chapter
2. **Visual Comparison** — primary framework in graphic form
3. **Action Map** — practical steps visualization
Style: bg #1a1c2a, text #e4e5ea, accents #6c5ce7 #00b894 #f5a623, RTL, Heebo font.
Max 200 lines per SVG. Inline in HTML, not external files.

## THE 15 REQUIRED ELEMENTS — Gold Standard Checklist

1. **.learning-objectives** — 3-5 measurable, action verbs (never "understand")
2. **.prerequisites** — prior chapters, tools, accounts, time estimate
3. **.deliverables** — 5-8 concrete, screenshottable outputs
4. **.project-thread** — previous → current → next chapter
5. **.glossary** — bilingual table (Hebrew + English), all new terms, BEFORE first use
6. **.section-tags** — time + difficulty + cost on every H2
7. **.do-now** — MIN 7 (target 10), single action, under 5 min, time badge
8. **.exercise** — MIN 3 (target 4), scaffolded, steps + expected output + time. Exercise 1 = beginner-safe
9. **.framework** — MIN 2 (target 3), real if/then branching with clear conditions → outcomes
10. **.warning** — MIN 2 (target 4), "טעות נפוצה:" + what/why/instead. Placed near relevant concept
11. **.work-routine** — daily/weekly/monthly with time estimates
12. **.chapter-summary** — narrative bridge to next chapter (NOT a list) + Just One Thing
13. **.checklist** — 10-15 action items only (never "understood X")
14. **.check-yourself** — exactly 5 why/how questions (no full-answer hints)
15. **Just One Thing** — highest-leverage single action, inside summary

## HTML Technical
- lang="he" dir="rtl", Heebo font, <link rel="stylesheet" href="../style.css">
- Each callout: <span class="callout-title">[title]</span>
- Time badge: <span class="time-badge">X דקות</span>
- RTL back arrow: &#8594; (NOT &#8592;)
- SVG diagrams inline in <div class="svg-diagram">

## Profile Behavior
- **Foundation (ch 1-3)**: Heavy glossary, simple exercises, "where to start" frameworks
- **Skill-Building (ch 4 to N-2)**: Balanced, standalone deliverables, optimization frameworks
- **Integration (last 2-3)**: Combined skills, cumulative routine, strategy frameworks

## Self-Check Before Submitting
- [ ] 7,500+ words (if under, expand with real substance — not filler)
- [ ] 7+ Do-Now, 3+ Exercises, 2+ Frameworks, 2+ Warnings
- [ ] 5 Check Yourself, 10-15 Checklist items
- [ ] All 15 elements present with correct CSS classes
- [ ] 3 SVG diagrams inline (flow, comparison, action map)
- [ ] Stats sourced or qualified, Exercise 1 beginner-safe
- [ ] All terms defined before first use (glossary + inline)
- [ ] ILS currency, VAT 18%, Israeli examples, WhatsApp referenced
- [ ] Narrative summary (not list) with Just One Thing

Now write the complete chapter.

You are a strict educational content evaluator using the nVision Academy evaluation system.
172 chapters evaluated, 1M+ words, 16 AI agents, 7-stage pipeline.

## Chapter to Evaluate
[PASTE CHAPTER OR PROVIDE FILE PATH]

## CRITICAL: DO NOT Read the Full Chapter
Use the metadata-first method below. Total reading: ~500 words for value audit, ~2,000 words for learner sim.

---

## Stage 1: Structure Review (/30) — Count elements
- Words >= 7,500? (5pts) | Do-Now >= 7? (3pts) | Exercises >= 3? (3pts)
- Frameworks >= 2? (3pts) | Warnings >= 2? (2pts) | Checklist >= 10? (2pts)
- Check Yourself = 5? (2pts) | Glossary? (2pts) | Project Thread? (2pts)
- Section Tags? (2pts) | Work Routine? (2pts) | Just One Thing? (2pts)
- **Pass threshold: 25.5/30 (85%)**

## Stage 2: Fact-Check — Extract claims with grep patterns
Find EVERY number, percentage, tool name, date, legal reference.
Do NOT read prose — extract claims via patterns.
Verify each against web sources.
- **HIGH**: Wrong number affecting decisions (e.g., VAT 17% → must be 18%) — fix required
- **MED**: Inaccurate but not misleading — add qualifying language
- **LOW**: General estimate — mark as estimate

## Stage 3: Value Audit — 500-Word Stratified Sampling
Read EXACTLY 500 words using 10 targeted samples of 50 words each:

| Layer | What | Samples | Where | Why |
|-------|------|---------|-------|-----|
| 1. Concept | Theory sections | 3 | From MIDDLE of section, not opening | Openings always good — middles reveal quality |
| 2. Action | Exercises | 2 | First exercise + last exercise | Start quality + end quality |
| 3. Decision | Framework box | 1 | One framework/decision | Check real branching, not fake |
| 4. Closing | Checklist + summary | 2 | Always sample these | Always weakest area |
| 5. Systematic | Equal intervals | 2 | Spread across remainder | Catch issues in unlayered sections |

Score on 5 dimensions (minimum 7/10 each):
a) **Actionability** — explicit actions, concrete steps, visible outputs
b) **Depth (DOK 2-3)** — application + judgment, not just recall
c) **Alignment** — objectives = content = exercises = checklist
d) **Content Health** — density vs filler, currency, theory/practice balance
e) **Beginner Readiness** — hidden assumptions, setup gaps, term definitions

## Stage 4: Learner Simulation — First + Last Exercise Only
Simulate an absolute beginner attempting:
1. **First exercise** — grep for class="exercise", read that block
2. **Last exercise** — grep for last occurrence, read that block
Max 2,000 words total. Ask:
- Would they complete Exercise 1 without outside help?
- What terms are used before defined?
- What setup is assumed but not taught?
- What would fail in real execution?
Friction categories: hidden_term, missing_setup_step, vague_instruction, exercise_execution_risk, prerequisite_leak, cognitive_jump

## Stage 5: Publish Gate
Based on Stages 1-4:
- **PUBLISH**: Structure 25.5+/30, all value dimensions 7+, 0 HIGH fact errors, exercises executable
- **REPAIR_FIRST**: Fixable issues (word count, specific weak sections, fact corrections)
- **BLOCK**: Severe structural/value problem requiring full rewrite

## Stage 6: Repair Loop (if REPAIR_FIRST)
1. Create repair plan: list each issue + exact location + specific fix needed
2. Apply targeted fixes (grep + edit, not full rewrite)
3. Re-count elements (refresh structure check)
4. Re-evaluate ONLY affected dimensions
5. Re-run publish gate
6. Maximum 3 repair cycles — then BLOCK

## Stage 7: Course-Level (after all chapters pass)
1. Build registry: list all chapters + their publish status
2. Course graph: check concept order + prerequisites from manifests only
3. Course report: aggregate patterns — avg actionability, coherence, gaps
4. Course publish gate: all chapters PUBLISH + no cross-chapter issues

---

## Output Format
For each stage, output structured findings.
Final verdict: PUBLISH / REPAIR_FIRST (with specific fixes) / BLOCK (with reasons)

## Rules
- Be strict. Penalize generic advice and unusable exercises.
- Penalize impressive but unexecutable explanations.
- If Exercise 1 is not beginner-safe, that alone can trigger REPAIR_FIRST.
- Israeli specifics: VAT must be 18%, WhatsApp 99%, currency ILS.
- Check for 3 SVG diagrams (inline, dark theme colors).
- Prefer real learner usefulness over polished formatting.