About
An assay is a test of what something is actually made of.
That is the job here: not to guess which machine produced a piece of work, but to establish what it contains before it goes out under someone's name.
A sign-off nobody has to take on faith.
Why this exists
The volume of work being produced with AI assistance went up by an order of magnitude in about two years. The amount of review capacity did not. Most of it now happens in the same person's head, minutes before publishing, with no record and no second opinion.
ES6 QA AI is the missing layer between generation and publication: mechanical checks that catch what is objectively wrong, and a route to a named professional for everything that requires judgement. Neither half is new on its own. Putting them in one place, with a written record, is.
We are deliberately not an AI detector, not a plagiarism service and not a legal clearance product. Those are different businesses, and pretending otherwise would make this one worse.
What we refuse to do
- Claim to detect authorship. It cannot be done reliably, and the errors fall on real people.
- Sell a green tick that means nothing. A pass means the rules ran and passed, and it says so.
- Warrant a reviewer's judgement. They state their scope, and you read it.
- Store your full draft to serve you an ad later. Reports keep a short excerpt, by design.
Four beliefs that shape the product
Output is not authorship
Nobody needs another tool that guesses whether a machine wrote something. What people need is a way to decide whether it is good enough to publish.
A score you can reconstruct
If you cannot see why a number moved, the number is decoration. Every deduction here traces to a named rule and a sentence you can read.
Local first, by default
Verification should not require handing over the draft. The checks are plain code that runs in the browser, so the default is that nothing moves.
Humans sign, machines flag
Automation is cheap and literal. Judgement is neither. Keeping those two jobs separate is the whole design.
How the standard is maintained
Rules are written by people who do the work: engineers for code, editors for copy, accessibility specialists for markup, media reviewers for images. Each rule needs a stated failure mode before it can ship, and a rule that produces false positives in practice gets retired rather than defended.
The rule set carries a version number, and every saved report records the version that produced it. When a rule changes, old reports still resolve to the old behaviour.
Read the full methodWorking with us
If you review professionally and want to take work through the network, write to us with your discipline and a sample of your sign-offs. We list reviewers with a real name, a real rate and a scope they will defend.
Get in touchStart with one file
Find out what your AI output is hiding.
The console is free, needs no account and uploads nothing. See the findings first, then decide whether you want a specialist to sign it off.