| 1 | """ |
| 2 | oversight_core.formats |
| 3 | ===================== |
| 4 | |
| 5 | Format-specific watermarking adapters. |
| 6 | |
| 7 | Each adapter knows how to embed and extract a mark_id for one file family. |
| 8 | The core protocol (container.py, crypto.py, manifest.py, beacon.py) is |
| 9 | format-agnostic; these adapters let watermarking work on more than plain text. |
| 10 | |
| 11 | MVP adapters: |
| 12 | text - L1 zero-width + L2 whitespace + L3 semantic (already in watermark.py + semantic.py) |
| 13 | image - DCT-domain frequency watermark (robust to recompression, resize, moderate crop) |
| 14 | pdf - per-recipient metadata + text-layer marks |
| 15 | docx - Office XML metadata injection |
| 16 | |
| 17 | Not in MVP (roadmap): |
| 18 | video - per-keyframe DCT + audio echo-hiding |
| 19 | audio - echo-hiding + spread-spectrum |
| 20 | xlsx - cell-comment marks + invisible columns/rows |
| 21 | pptx - slide-note marks + image DCT on each slide image |
| 22 | """ |
| 23 | |
| 24 | from . import text as text |