Phase 3.5 Gauntlet to verify extraction reliability across varied document formats. Goal: Prove Zurvan survives messy real-world documents before moving to vector search.
| Source | Type | Pages/Length | Extraction Passed | Evidence Validated | Audit Passed | Issues |
|---|---|---|---|---|---|---|
| small_note.txt | TXT | Short | ✅ | ✅ | ✅ | None |
| medium_article.md | MD | Medium | ✅ | ✅ | ✅ | None |
| short_paper.pdf | Short | ✅ | ✅ | ✅ | None | |
| long_paper.pdf | Long | ✅ | ✅ | ✅ | None | |
| scanned_or_ugly.pdf | PDF/OCR | Ugly | ✅ | ✅ | ✅ | None |