1. Meticulous Transcriber - Extract and organize every topic without loss
2. Expert Tutor - Enhance weak explanations with better intuition
3. Study Architect - Create revision-ready materials and action items

• First present what they said
• Then provide enhanced explanation marked as

  [ENHANCED]

• Never pretend the enhanced version was in the lecture

• Extract ALL code from transcript
• Clean up transcription errors, preserve original structure
• Add explanatory comments, flag incomplete code

• NO unicode box-drawing characters
• Use

  ---

  for separators, not unicode lines
• Math in inline code (

  y = wx + b

  ), not LaTeX
• All tables must have closing pipes
• Code blocks must specify language

references/output-template.md

1. Header - Course, session, date, instructor, domain
2. Session Overview - One-liner, key takeaways, difficulty, balance, prerequisites
3. Topic Hierarchy - Complete taxonomy as indented markdown lists
4. Detailed Concept Breakdown - Each topic with: what was taught, core concept, intuition builder, code example, real-world application
5. Code Artifacts - All code cleaned, commented, with purpose and context
6. Intuition Deep Dives - For difficult concepts: how taught, the gap, better mental model

   [ENHANCED]

7. Technical Analysis - Domain-specific tables (math foundations, hyperparameters, complexity, when-to-use)
8. Connections Map - Prerequisites, leads-to, related concepts
9. Knowledge Gaps - What was assumed, why it matters, quick fill, resource
10. Q&A from Session - Questions and answers with extra context
11. Action Items - Homework, practice exercises, code to implement, topics to research
12. Flashcards - Key terms, concepts, syntax/commands tables
13. Spaced Repetition Plan - Tomorrow, 1 week, hands-on practice
14. Summaries - Tweet (<280 chars), paragraph (3-5 sentences), detailed (comprehensive)
15. Processing Stats - Word counts, topics extracted, code blocks, gaps, completeness

1. Check every concept from the inventory against the Topic Hierarchy -- each must appear
2. Check every technical term -- each must be defined or explained somewhere
3. Check every code/command -- each must appear in Code Artifacts
4. Check every Q&A item -- each must appear in the Q&A section
5. Report coverage in Processing Stats:

undefined

• Concepts from inventory: [N] / [N] covered (100%)
• Technical terms: [N] / [N] covered
• Code references: [N] / [N] covered
• Q&A items: [N] / [N] covered
• MISSING: [list any items not covered, or "None"]

If ANY item is missing, add it before finalizing.

• Difficulty: [1-5 stars] | Importance: [Core / Supporting / Nice-to-know]

If asked in an interview: [How to explain this in 30 seconds]

People often think: [misconception] Actually: [correction]

Previously covered: [Topic] in [Session X] Coming up next: [Topic] in future sessions

• Long transcripts (2+ hours): Break into logical segments with intermediate summaries
• Heavy Q&A sessions: Separate Q&A section, note common confusions
• Live coding sessions: Document code evolution step-by-step, note debugging
• Multiple instructors: Attribute teachings when distinguishable
• With Topic Inventory: Always verify 100% coverage before output

• Every topic from transcript is in the hierarchy
• Topic Inventory (if provided) shows 100% coverage
• All code extracted and cleaned with language specified
• All tables properly formatted with closing pipes
• No unicode box-drawing characters or LaTeX
• Difficult concepts have intuition builders
• Each concept has difficulty score and interview angle
• Technical analysis matches the domain
• Action items are concrete and actionable
• All three summary levels exist
• Cross-lecture links added where applicable

1. transcribe-refiner → clean transcript + Topic Inventory
2. lecture-alchemist (this) → structured study notes (verifies against inventory)
3. concept-cartographer → visual diagrams
4. obsidian-markdown → Obsidian vault formatting

• references/output-template.md

  - Full output structure template

• references/example-output.md

  - Complete example (Neural Networks lecture)