Two deterministic checks read what your AI wrote and mark the exact lines that give it away, the same way every run. And DOMAINS shows you what your prose actually is, its register and its voice, so you can steer it back to yours. They never rewrite for you: a tool you can read, not another AI you have to trust.
AI text has three diseases a reader feels even when they cannot name them. This surface names them and points at the exact line.
Flat sentences that all run the same length. Power words doing the work evidence should do. Openers that clear the throat before the point. Hedges that refuse to commit. The reader cannot say why, but it feels machine made.
The prose turns to point at its own parts ("that is the map") or count its own list ("the second point is"). It aims the reader's eye ("notice the shape of it") or names itself ("this section argues"). The reader notices the pointing more than the point. Coined by Gaurav Rastogi, 2026.
Words the writer understands and a newcomer does not. Terms used before they are defined. All we-and-our and no you. Copy written to the people who made it, not the person reading it.
Each check does one thing; the writing check bundles several. Give it the same text and you get the same verdict back.
Flat cadence, escalation words, throat-clearing openers, hedge density: the prose that reads as machine made. Scored on four named dimensions of voice. Pure rules, no model.
The six forms of the stoolworm (prose that points at, grades, and announces itself), plus jargon a newcomer cannot follow. Fixed rules, plus a cheap reader that checks whether an outsider understands. Every term it learns becomes a rule for next time.
The writing check run on a paragraph stuffed with stoolworms. Every line it flags is a place the prose stopped delivering and started managing the reader. Run it on your own text in the playground →
$ check "In this essay, we lay out the argument. That is the map.
The second position is the one that matters most. Notice the
shape of it. So the fact can be conceded. The honest word is
compromise. Having established that, let me now turn to the cure."
verdict: FAIL stoolworms: 6 inward: 100%
"In this essay" names the container self-reference
"That is the map" points at its own parts self-pointing
"The second position is" counts its own list enumeration
"Notice the shape of it" aims the reader's eye reader-gaze
"So the fact can be conceded" announces its own move move-announce
"The honest word is" grades its own diction diction-grading
cure: open inside the substance, hand off through the material,
trust the reader to feel the shape without being told.
Same paragraph in, the same six findings out, every time.
Call a check on its own, or run it inside the doloop machine (our hosted service), which keeps your prior verdicts and the rules learned from them. Same checks either way. The difference is whether the run carries state.
Send a block of text, get the findings back. Nothing to set up, nothing kept between calls. The fast way to check one piece of writing.
Connect your own AI to the doloop machine in writing mode and the checks run on each draft. Your AI drafts, the checks judge, your AI fixes the exact lines, and only clean copy ships.
Want this on your writing pipeline? Talk to us, or see the other surfaces.
The slop checks catch what is wrong. DOMAINS measures what the prose actually is: which of eight writing registers it sits in, and how far it is from where you want to be.
Paste a draft. Score it on seven dimensions of voice: diction, rhythm, mechanism, personal address, imagery, narrative arc, and suppression. It names your nearest register and gives you the one change that closes the gap.
Routes you to a real doloop page, asks when your question is ambiguous, or tells you when there is no answer. No model runs on the answer path, so it cannot invent one.