From submitted submission to approved grade.
The question is not whether a machine can assign a grade. It's whether it assigns yours. CORRAICT does not mark with a generic marking scheme: it applies the grid you submitted, criterion by criterion, and returns control to you before anything reaches a student.
Five steps, three of which are yours.
The same sequence for a scanned in-class exam and for an online submission.
You set the framework
The instructions, your grading context, the maximum grade, and the solution if you have one. The solution can be text or an attached file: it is read.
Once per assessment
Submissions come in
Via a public link you distribute, or composed online in the answer form. A hand-in replaces the previous one for the same student.
During the submission window you have set
The AI proposes
Each submission is assessed against your grid. The result is a grade per criterion, each accompanied by the justification supporting it, and a draft feedback.
While you get on with other things
You decide
You adjust a mark, override a justification, rewrite the feedback. The total recalculates. Nothing is locked.
The only truly irreducible step
You approve
Approval is what makes a grade exist. Until it has taken place, the submission remains a proposal the student cannot see.
Submission by submission, or in bulk
Accepted formats for a submission: PDF, photo (JPG, PNG), spreadsheet (Excel, Numbers). A handwritten submission is read as-is, from a photo or a scan.
Not a generic marking scheme. Yours.
It is the only thing that distinguishes usable marking from a plausible grade.
You write your expectations in plain language
No tick-box form, no imposed structure: you write your marking instructions as you would explain them to a colleague marking in your place. Weightings, tolerances, what earns a mark and what does not.
Every mark is tied to a criterion
A single overall grade cannot be discussed: you either accept it or start over. A grade broken down criterion by criterion, with the passage from the submission that justifies it, can be reviewed in seconds — and any disagreement is visible without re-reading the whole thing.
What you see, and what you can override.
The submission on the left, the proposal on the right. Everything is editable.
- The student's submission, displayed alongside the marking — not in another tab.
- The score for each criterion, editable, with the total recalculated automatically.
- The justification for each mark, rewritable: it's your wording the student will read.
- The overall feedback, editable or fully replaceable.
Approving makes the grade visible to the student. Sending the notification is a separate action: you choose when the class is informed, and you can withdraw an approved grade if it needs to be revised.
What really happens in difficult cases.
None of these cases produces a silent grade. They produce a status you can see.
What can be verified without AI
For an exam composed online, numerical answers are compared cell by cell against your solution, with no model involved. The result is correct or it is not. Everything else — what requires judgement — goes back to the AI.
When reading fails
Illegible handwriting, corrupted file, format we cannot read: the submission is flagged as an error, with the reason. It is never given an estimated grade or set to zero by default. You resubmit, or mark that one by hand.
A log, at every step
Queuing, starting, reading files, calling the model, writing the result: every step is timestamped and viewable. You can see whether a marking job is waiting, running, or has failed — and you can reset it.