A simple, direct prompt builder
Paste. Build. Check.
Have a Claude-assisted draft built from your ideas? Rewrite Room masks exact values and builds a prompt locally. A non-Anthropic model does the writing. Paste its draft back here for mechanical checks.
Why this
Full manifestoIdeas shouldn't carry watermarks.
Rewrite Room changes nothing by itself. It builds a prompt and checks surface differences. The target is wording, never thought. A statistical trace cannot tell who formed an idea or took responsibility for it.
Provenance must not become a witch-hunt verdict.
A detector may flag statistical influence. Turning that flag into proof of authorship, fraud or intellectual laziness is judgment outsourced to a score.
Enforcement punishes the people writing tools help most.
Dyslexia, dysgraphia, motor limitations and second-language writing can separate the quality of an idea from the ease of typing it. Treating detectable assistance as cheating measures surface production, not intellectual contribution.
Mark synthetic evidence. Not ideas.
An image can pretend to be a camera record. Text is a symbolic carrier: a false claim is harmful because of the claim, not because software helped phrase it.
Evidence you can actually use
See what changed and what survived
“So is the watermark gone?” Nobody outside Anthropic can certify that from this page. What you can see is whether exact facts survived and whether too much of the old wording, openings or structure remains.
The usable output
Your draft, with exact values restored
This is the external model's draft with every surviving [PV-XX] marker replaced locally by the exact original value. Read it for meaning, tone and unsupported claims before you use it.