| # Source preview layout metadata | |
| PREVIEW_INSTRUCTIONS = """ | |
| ### 📂 Data Ingestion Instructions | |
| - Supported files: PDF, DOCX, CSV, TXT, MD, JSON, JSONL. | |
| - The normalized source context removes redundant markup, formatting, and white space to build a strict, clean baseline context. | |
| - Estimations use the standard 4-character to 1-token ratio; exact tokens are counted live on the server. | |
| """ | |