A student, a researcher, a one-person business. Work that matters to you, done with an assistant, with nobody checking it but you.
The engine reconciles an answer against your own documents: every figure traced to its stated source, identifiers run through their own check digits, findings ranked by whether a number stayed in scratch work or went into the deliverable. The famous examples below are just the part you can verify yourself in one second — which is the point of them.
Three. Counting, not a lookup, and you can check it yourself in a second.
68. One multiplication.
7 × 13. Trial division.
Fails its own check digit, so no book has that number. Not hard to find — impossible.
Not a date that exists, and no calendar system disagrees about that.
Credited to the thing you pasted, and absent from it — checkable by you in seconds.
None of this needs a reference file or a model call. The check is the fact, which is why nothing here can go stale or be curated wrongly. And where a check depends on a definition — which gallon, which calendar — it names the definition it used, and declines where it cannot pick one. The same engine reconciles every figure in an answer against the document it claims to come from, and ranks what it finds by where the number went — scratch reasoning, the deliverable, or out through a tool.
If your work is ever questioned — and for students it increasingly is, by detectors whose false positives land hardest on writers whose first language is not English — there is a signed record of what was asked and what came back. Not because you were preparing a defence. Because you were checking your work, and the record is what checking produced.
That order is deliberate. A tool sold as an alibi attracts people who want an alibi, and they have every incentive to game it. This one flags and discloses, and never rewrites.
Verification is not an accusation. A teacher can trust you and still be required to check — that is the job. The receipt lets them do it without you handing over everything else you said.
You choose what to show. Three exchanges out of seventeen, verbatim — and the receipt proves each belongs to the sealed conversation and where it sat. The other fourteen are proven to exist and never read. An all-or-nothing transcript would force a choice between privacy and standing, and most people would choose privacy by not installing.
Verified in a browser, with nothing installed. The page has no backend, and the receipt travels in the part of the address browsers never send to a server — so whoever hosts that page cannot see what is being checked, and neither can we. Mirror it. A university running its own copy is the point.
And on the face of every receipt: that disclosure is selective by design, so you could have chosen the flattering exchanges — and that it proves use and cannot prove the absence of other use. Both facts are written into every receipt.
A teacher is allowed to ask that, and often has to. What they usually get is a conversation about whether you can be trusted. What you can give them instead is a record of the work.
A project receipt covers one piece of work rather than one exchange: the sealed conversations you include for it — which assistant, when, what was asked, what came back — and the findings the checks raised along the way. [In development: marking which findings you acted on. A citation that failed its own check digit and that you removed is the best evidence of judgement there is, and the receipt should show it.]
You still choose what to show. The receipt covers the project, not your life: exchanges from other work stay sealed and are proven to exist without being read. A teacher verifies it in a browser with nothing installed, and learns exactly what you disclosed about that project and nothing else.
The request goes to you, and you answer it. There is no way for a teacher, a department or a platform to pull your history without you handing it over — and that is a property of the architecture rather than a setting someone could change.
Institutions buy per seat, from a budget that already exists, and they are looking — detectors are being deprecated for producing false positives nobody can defend.
The line that decides whether this stays honest: the institution pays for the capability, you hold the key, and disclosure is you handing over a receipt per submission. The day a dashboard is built, the receipt stops being a defence and becomes evidence they already hold.