CNVXTY builds calibrated instruments for people whose writing has to hold up: MultiWriter shapes prose to a writer's own decisions, and Geiger measures language against a standard the writer does not control. Neither is tuned to satisfy the other. That separation is the product.
General AI tools produce prose that is lucid on the surface and falls apart on inspection. A skilled reader can detect it. Almost nobody can prevent it — which is why capable people use these tools for workflow and not for the work that matters.
The deeper problem is that sound and invented text read identically. A confident paragraph looks the same whether it rests on something or on nothing. You cannot audit for that by reading the output, because the failed version and the sound version are indistinguishable on the page.
This is not a design preference. It is a constraint that was proved in the 1930s and restated plainly by the cyberneticians in the 1950s: any language rich enough to talk about language is incomplete, and the evaluation has to happen from somewhere outside.
Most AI writing tools ignore it. They generate text and assess their own output, or hand the assessment to the user and call that independence. The seat gets moved and renamed rather than built.
Geiger judges MultiWriter's output. Geiger is never adjusted to make that output score better. If it were, the measurement would be worthless and everyone involved would know it — so the discipline is structural rather than promised.
A calibrated instrument, not a chatbot. You set the dials; it renders your decisions. It offers alternatives and keeps the record of what you chose and why. It sharpens writing that has an edge to find; it will not manufacture a voice. A honing stone, not a forge.
The same engine, aimed at policy. Where a writing instrument is mostly choices, a policy instrument is mostly floor — a dense set of things it will not do, with a few genuine drafting decisions on top. The loop is yours and it is explicit: you write, you submit to Geiger when you choose to, you revise. Nothing optimises on your behalf.
An independent instrument for reading institutional language — what a document commits to, what it only appears to commit to, and where a reader has been left no room to disagree. It scores. It does not write, and it is not adjusted to make anyone's text pass, including ours. Its internals stay closed, because an instrument you can see inside is an instrument you can tune.
A page that concedes nothing, bounds nothing, and names no uncertainty is not a strong page. It is a page built to be agreed with. We would rather hand you the handles.
We are more interested in hearing where the argument breaks than in hearing that it lands. If you work on institutional language, governance review, or teaching people to write, there is a conversation worth having.
write to contact at cnvxty.com