Text-Hygiene Engine
The Text-Hygiene Engine is our MIT-licensed engine (Node and Python) and protocol for shipping text and documents clean: zero zero-width characters, no stray variation selectors, purged document and image metadata, provenance manifests removed. It covers the layers that can be verified; statistical watermark desynchronisation is declared unverifiable in writing since 2026-08-14.
What it does: Audit and clean the text, documents and images you publish of invisible characters, metadata and provenance manifests, in the formats listed below, and know exactly which layer you cannot verify.
What it will never do: Will never claim to defeat an AI detector. Detector evasion is not a feature and not a promise; the statistical layer is declared unverifiable.
Instrument: the engine's own audit log per publication, kept with each article's production record; PDF metadata purge script in the toolkit.
Proof files (private audit folder, dated): AUDIT/linkgraph-2026-09-03/. See the ledger for the public reading.
UnC2PA, Documami, RemoveAILabel, Scanly: free browser tools for metadata and C2PA removal, one file at a time PRIMARY source
Neighbouring category, not a substitute. Undetectable AI: $9.99 to $42.50 / month by word volume PRIMARY source
Text-Hygiene Engine: $99 once. The free tools do one file in a browser; the kit does a whole publication pipeline from the terminal, logs it, and tells you which layer it cannot see.
Primary = read on the vendor’s own pricing page on that date. Secondary = a third-party read, not confirmed on the vendor page. No calculator, no “breakeven”, no “profit”: the readings are context, not a swap.
What is inside
The engine (JavaScript and Python versions, MIT, ours), the four-layer protocol, the per-publication checklist we run, the metadata purge scripts for PDF and images, and the written register of what cannot be verified and why.
Waitlist: Text-Hygiene Engine
No ship date is set: a system goes on sale the day its payment test passes. You get one dated note then, with the price frozen and the run proof attached. No sequence, no countdown.
- Value against the free tools: Their base function is free. What is sold is the pipeline, the log and the register; whether buyers pay for that is the open question of the distribution test.
Text-Hygiene Engine: questions
Does it hide text from AI detectors?
No, and no page here says it does. It removes the verifiable marks (invisible characters, metadata, provenance) and declares the statistical layer unverifiable.
What formats?
Plain text and Markdown, DOCX, PDF and common image formats, from the command line, with a per-file log.