Paper: https://arxiv.org/abs/2303.17651
Source code: https://github.com/madaan/self-refine
Audited commit: 9a206d41e5d2d0c241bb441f41eeadb945afaa55

A high-quality extraction/audit package for Self-Refine:

• research summary,
• complete prompt extraction,
• raw prompt files,
• Python logic/flow extraction,
• Visual Self-Refine GPT-4V notebook audit,
• docs/website prompt audit,
• deep-dive task matrix,
• paper appendix notes,
• Mermaid graphs in English and Arabic,
• Arabic final audit and completeness checklist.

1. research_summary.md
2. deep_dive_task_matrix.md
3. deep_dive_paper_appendix_notes.md
4. prompts_complete.md
5. python_logic_flow_complete.md
6. visual_self_refine_notebook_audit.md
7. docs_prompt_algorithm_audit.md
8. graph_english.md
9. graph_arabic.md
10. final_completeness_check_ar.md
11. QUALITY_REVIEW_AR.md

• raw_prompt_files/ — untruncated prompt/supporting sources.
• python_logic_inventory.json — machine-readable logic inventory.
• graph_english.mmd, graph_arabic.mmd — Mermaid source files.

Large historical outputs/datasets are documented as artifacts and not mixed into prompt templates.