Nelson Pediatrics Anki Flashcard Deck

Comprehensive Anki flashcard deck generated from a 265-page Hebrew medical summary of Nelson's Textbook of Pediatrics (22nd Edition), using Claude Code CLI with an automated multi-phase pipeline.

Final Output

• 4,557 notes across 14,471 cloze review cards in the merged deck (verified 2026-09-27 via python scripts/verify.py output/nelson_complete.apkg; earlier batch-level counts before the gap-fill merge do not sum to this exactly, so this is the number to trust)
• 265/265 pages covered (100%)
• 37 APKG files in output/ — all pass strict verification (+ nelson_complete.apkg merged deck)
• 21 medical specialties organized into sub-decks
• External audit: Gemini Pro 3.1 table-coverage audit identified and patched 142 supplementary cards across 12 pages

How to Use

1. Download output/nelson_complete.apkg (single merged file with all 4,557 notes) — OR import individual batch APKGs from output/
2. Import into Anki (File → Import)
3. Cards appear under the deck נלסון 21:: with sub-decks per specialty and disease

Card Model (7 Fields)

Field	Description
Text	Hebrew medical content with cloze deletions
Context	Disease | Subtopic | Page reference
Extra	Clinical pearl or exam-relevant insight
NelsonChapter	"Ch. NNN. Chapter Name"
SearchKeywords	English medical terms for Google
GoogleURL	Auto-generated Google search link
ChapterURL	Link to Nelson online chapter

Specialties Covered

#	Specialty	Pages	Notes
1	מחלות זיהומיות (Infectious Diseases)	1-47	901
2	נוירולוגיה (Neurology)	48-66	254
3	קרדיולוגיה (Cardiology)	67-86	252
4	ריאות (Pulmonology)	87-95	135
5	אימונולוגיה (Immunology & Allergy)	96-111	184
6	המטולוגיה (Hematology)	112-128	290
7	אונקולוגיה (Oncology)	129-136	174
8	גסטרו-אנטרולוגיה (Gastroenterology)	137-155	296
9	תזונה (Nutrition)	156-161	97
10	נפרולוגיה (Nephrology)	162-178	297
11	אנדוקרינולוגיה (Endocrinology)	179-205	555
12	ראומטולוגיה (Rheumatology)	206-219	215
13	מחלות מטבוליות (Metabolic Diseases)	220-233	264
14	נאונטולוגיה (Neonatology)	234-246	239
15	אורתופדיה + עיניים + עור (Ortho/Ophth/Derm)	247-259	275
16	התנהגות + גנטיקה + חירום (Behavioral/Genetics/Emergency)	260-265	109

Pipeline Architecture

PDF → Extract (PyMuPDF) → Chapter Map → Card Writing (Claude sub-agents)
    → Build APKG (genanki) → Verify (strict) → output/*.apkg

Key Design Decisions

• Sub-agent architecture: Each card-writing phase runs in isolated context (3-4 pages max)
• Gold standard reference: 8 example cards from best batch included in every sub-agent
• Anti-degradation: Formula-based density minimums, cloze distribution validation
• Two-level verification: validate_density.py + verify.py --strict
• Progress tracking: progress.json enables interrupt recovery at any phase
• External audit: Post-completion table-coverage audit via audit_tables.py + Gemini review to catch structural gaps
• Merged output: merge_apkg.py produces a single complete deck for easy distribution

Project Structure

├── output/           ← 37 APKG files + nelson_complete.apkg merged deck + card markdown archives
├── input/            ← Source PDF + Nelson chapter index
├── work/             ← Page images (PNG), gold standard, audit reports, last batch state
├── scripts/          ← Pipeline scripts (extract, build, verify, fix, audit, merge)
├── CLAUDE.md         ← Claude Code instructions
├── prompt.md         ← Card writing rules (11 rules)
├── orchestrator.md   ← Phase-by-phase pipeline logic
├── progress.json     ← Machine-readable project state
├── lessons.md        ← Comprehensive lessons learned (11 critical lessons)
└── todo.md           ← Human-readable progress tracker

Scripts

Script	Purpose
extract.py	PDF text extraction + image page PNG export
build_apkg.py	Parse markdown → 7-field APKG with sub-decks
verify.py	APKG structure + content verification (strict mode)
validate_density.py	Card density + cloze distribution checker
fix_missing_chapters.py	Patch missing NelsonChapter/ChapterURL in APKGs
fix_cloze_distribution.py	Surgically reduce excess cloze counts
audit_tables.py	Table content coverage audit (PyMuPDF table detection vs cards)
merge_apkg.py	Merge all batch APKGs into single nelson_complete.apkg

Dependencies

pip install pymupdf genanki

Quality Metrics

• Avg clozes/note: 3.2 (target 2.5-3.2)
• Cards ≤3 clozes: >70% (target ≥50%)
• Cards at max 5 clozes: <5% (target ≤20%)
• All 7 fields populated: 100%
• Strict verification: 37/37 PASS (including supplementary gap-fill)
• Table coverage audit: 96.6% (0 RED, 1 YELLOW, 196 GREEN pages)
• External audit: Gemini Pro 3.1 confirmed 142 gap-fill cards across 12 pages

Built With

• Claude Code CLI (Anthropic) — card generation and pipeline orchestration
• genanki — APKG creation
• PyMuPDF — PDF text and image extraction

License

This project generates study materials from copyrighted medical textbook content. The flashcards are for personal educational use only.