AI Transparency

No. SuperGrader does not train, fine-tune or build AI models. We have no training pipeline, and no student work is collected, retained or repurposed for one. Uploaded papers exist to be graded for the teacher who uploaded them, and for nothing else.

We also do not sell, share or license student data, and we do not use it for analytics, marketing or product research.

Student work is processed by a third-party AI provider in order to be read and graded. If your district requires written confirmation of that provider’s own training and retention terms, contact us and we will supply their current terms directly.

Grading uses a leading commercial multimodal (vision and language) model from a major AI provider, accessed over its API. Essay marking additionally uses an image model to draw red-pen annotations onto a copy of the page.

We do not publish the specific model name, because models are selected for accuracy and change as better ones become available — naming one here would be out of date within months and would tell a district less than it appears to.

If your district’s review process requires the specific provider and model in writing, we will provide it on request.

Partly known, partly not. Uploaded images are stored on Cloudflare R2 and account records in a Neon PostgreSQL database; the application runs on Railway. Those are US-configurable and contractually bound.

The grading step sends student work to our AI provider — named in full in our Privacy Policy alongside every other subprocessor — and we do not currently publish a guaranteed processing region for that leg. If your district requires data residency commitments, tell us before you upload student work. This is the question most likely to matter to you and least likely to have a comfortable answer today.

By default, until the teacher deletes it. There is no automatic expiry: assessments, grades, student names and images stay in the account until removed.

Deleting a grade, an assessment or a folder removes it and its images. Deleting the account removes every stored image and every record, then the login itself. What survives is a minimal receipt that the deletion happened — account id, action, timestamp, no student data — so we can show the request was honoured.

The service supports an enforced retention window, and we can enable a fixed period for an account on request.

Yes, and this is the part we are most confident about.

• Every question carries a confidence score, and low-confidence answers are flagged for review rather than quietly scored.
• Each mark shows what the model read from the paper, so a teacher can see the transcription the score was based on.
• Questions with an exact answer key are scored by direct comparison in ordinary code, not by the model — the AI reads the handwriting, arithmetic decides the mark.
• Ambiguous cases are surfaced rather than guessed. Two options ticked on a multiple-choice question is reported as ambiguous, with both recorded.
• Editing the answer key recalculates affected grades from the stored transcription, so a key correction does not require regrading.

Yes. Every grade is a draft until a teacher accepts it. Any score can be changed per question, feedback can be edited, and a student’s name can be corrected if it was misread.

Manual adjustments are respected permanently: once a teacher has overridden a mark, later recalculations leave it alone rather than overwriting the human decision.

Our terms state the corollary plainly — the teacher is responsible for the marks they issue, and no grade should be released without review.

No. The content sent to the AI — the images of student work and the grading instructions — is not written to our application logs, and neither are the model’s responses.

When a grading call fails we record the diagnostic shape of the failure: which model, the HTTP status, the reason the model gave for stopping. Not the paper, not the transcription, not the student.

No, and it cannot be. The paper is sent as a photograph, and a student’s name is usually written on it. We also deliberately ask the model to read that name, because grades have to be attached to the right student.

So the honest answer is that identifiable student work leaves our servers and is processed by a third-party AI provider. Any district review of this tool should start from that fact rather than around it. If your policy prohibits sending identifiable student work to third-party AI services, this product is not compatible with it, and we would rather you knew that now.