nomarklisted

# NoMark Watermarks in text live in four layers. They are independent: a document can be clean at one layer and heavily marked at another, so work through all four rather than stopping at the first hit. | Layer | What it is | Tool | |---|---|---| | 1. Invisible characters | Zero-width, tag block, variation selectors, bidi controls | `scripts/clean_text.py` | | 2. Confusables | Exotic spaces, Cyrillic/Greek homoglyphs | `scripts/clean_text.py` | | 3. Metadata | Authors, tools, GUIDs, revision IDs, timestamps, leaked paths | `scripts/clean_docs.py` | | 4. Style | Lexical and structural signatures of generated prose | Rewrite by hand | Layers 1-3 are deterministic and belong to the scripts. Layer 4 is judgment and belongs to you. Never try to do layer 4 with a regex, and never try to do layers 1-3 by eye -- you cannot see a zero-width character in a file you have read. `scripts/nomark.py` dispatches all three tools (`scan`, `text`, `docs`, and `check` for a CI gate); the individual scripts also work directly. ## Always scan before cleaning Show the user what is in their file before changing it. The finding is often the point: a decoded hidden payload tells them they were tracked and by what. ```bash python scripts/scan.py FILE # text, or a document's metadata python scripts/scan.py DIR --ext .md,.txt # sweep a tree python scripts/scan.py FILE --show-low # include typographic tells python scripts/scan.py FILE --json # machine-readable ``` R