Four features. No verdict.
Each student works on their own figures, under their own name. And you know how their submission was composed.
Four features, and how each one works.
Four features, delivered and live. They individualise the numerical data in the exam, verify who is sitting it, and transmit to the teacher factual information about how the submission was written. None produces a verdict — and no model intervenes: these are deterministic checks and timestamped browser events.
Individual variants
The numerical values in the Instructions vary from one student to another. The Instructions, questions, their order, and the Marking scheme remain identical for the whole Class: grading fairness is preserved. The Solution for each variant is not written by an AI — it is recalculated from the Teacher's formulas. Assignment is calculated from the student's address, never by chance: they always retrieve the same subject, even if they switch devices mid-Exam.
The values that vary are declared by the teacher in the solution, with their step and bounds. A student's exam is calculated from their email address, never by chance: they retrieve the same variant after changing device, and their solution is recalculated accordingly.
Knowing who is sitting the exam
The student no longer simply declares their identity: they present a secret that is their own. A code you distribute in the room, or that the system sends them by email at the time of the exam. For file-based hand-ins, anonymous submission is replaced by a personal link, valid once. Your enrolment list determines who may sit the exam — and you keep, student by student, everything needed to unblock someone in three seconds. You can also restrict access to addresses from your domain: the code and the link are then sent to the institution's mailbox, and to read them the student goes through your email provider's authentication — Microsoft 365, Google Workspace — without any CORRAICT password being created for them.
Three modes, at the teacher's choice: no verification, a named code, or a personal submission link consumed on first use. The code is distributed in the room or by email, depending on whether phones are permitted.
Documented composition records, restricted where necessary
The teacher receives facts: the number of paste events and their length, the typing rhythm, the composition duration. The content of what is pasted is never recorded — no keystrokes are stored, no tab changes are observed, nothing is collected from the student's device. The list of recorded data is fixed, collection is disabled by default and enabled exam by exam, and the student is informed before they begin. You can also block pasting during the exam: each blocked attempt is then logged, along with its length. Blocking without logging anything would mean trading real information for mere friction.
Pastes and their length, typing rhythm, durations: server-side timestamped facts, recorded only if the institution has granted permission. No tab switching is observed, and nothing is collected from the student's device.
The decision remains human
No score, no indicator, no risk level, no colour badge. All submissions display the same information: only the value changes, never the presence of the marker. The product proposes no action and singles out no one — it is the teacher who reads, and who decides.
The submission list displays the same column for all submissions, with its value. None is flagged, sorted, or surfaced to the top: the teacher notices for themselves — no one has pointed anything out on their behalf.
Individual variants and the composition record require an exam taken online. A handwritten submission uploaded as a photo is marked normally, without these two features.
What this is worth, precisely
A useful feature is one whose boundaries are known. Here are ours, with their rationale rather than a promise.
Individualisation makes a pre-obtained answer useless from as few as three variants. It also makes copying from a neighbour more difficult, but to a degree that depends on the number of variants: with four variants in a class of thirty, eight students share the same figures. We prefer to give you the calculation rather than a promise.
The composition record describes a way of writing; it proves nothing. A student who manually copies a text prepared elsewhere produces a perfectly ordinary record; a student who pastes a quote from their course material produces a paste event. Individualisation and the composition record also require online composition: a submission handed in as a file or in manuscript carries neither a variant nor a record.
Verifying identity addresses impersonation — submitting on behalf of an absent student, overwriting a classmate's Submission. It does not address substitution: a code can be dictated over the phone, a link transferred in one click. If a third party submits with the student's agreement, none of these measures will show it — only physical invigilation or an oral follow-up covers that case. Restricting addresses to your domain does not move this limit: it proves that the address belongs to your institution, never the identity of whoever submitted.
Likewise, blocking paste is a speed bump, not a barrier: voice dictation and manual retyping remain possible. What you gain is less the restriction itself than the trace it leaves — no one accidentally pastes a thousand characters.
We do not integrate any AI-generated text recognition tool. This is a deliberate choice — here is why.
These tools do not deliver on their promise: OpenAI withdrew its own in 2023 due to insufficient accuracy — on its own output. More significantly, Stanford research (Liang et al., 2023) shows they massively classify texts written by non-native authors as generated — on an international programme, this is not an incident, it is a structural bias. And an "87% likely AI" flag on a graded submission cannot be defended before a panel: several universities have disabled theirs for this reason. We prefer to give you an exam that no one else has, and facts a student can read.
See exactly what the teacher receives.
Submit an assessment and a few submissions. You will see the facts transmitted — and, just as usefully, what they do not say.