Skip to content
LEAPDocs
2026.08

Paper Quizzes

For cohorts that sit exams on paper: print, photograph, extract, grade, override.

The Paper Quiz option is available while an assessment is a draft. Publishing hides it, because delivery mode cannot change once students can see the assessment.

End to end

The steps

  1. 1
    Generate and print the paper

    A question paper is produced from the quiz group, with a QR code identifying the assessment.

  2. 2
    Upload the marking scheme

    This is what grading is grounded in. Without it the AI infers the scheme, which is where earlier versions went wrong.

  3. 3
    Sit the exam on paper

    Normal invigilation. Nothing digital happens in the room.

  4. 4
    Photograph the scripts with a phone

    Scan the QR code to open a capture session on your phone — no app install, and no separate login. Photograph each script.

  5. 5
    Extraction and grading run

    Each script is OCR-ed, the responses are extracted, and each is graded against the extracted answer sheet. Every graded response shows the criterion it was matched against.

  6. 6
    Review and override

    Work through the roster. Anything you disagree with is a one-click override, and the override is what counts.

Why the marking scheme matters

A blank sheet once scored full marks

Before grading was grounded in the extracted answer sheet, the model could hallucinate responses that were never on the page — a blank script scored 12 out of 12. The fix was to ground grading in the extracted text and show the criterion each mark was matched against. Upload the scheme; the grounding is what makes the marks defensible.

Capture limits

Per fileValue: 10 MB: A phone photo is well under this; a raw scan may not be.
Per capture sessionValue: 500 MB: Covers a full class of scripts.
Uploads per minuteValue: 60: Paced so a whole cohort can be captured without tripping rate limits.
Files per sessionValue: 200: One session per class sitting.

When a script will not read

Extraction produced nothing for one script.

Usually the photograph: shadow across the page, an angle steep enough to distort lines, or handwriting too faint. Re-photograph that script in even light, flat on, filling the frame.

The marks look wrong for one student.

Open the response and read the criterion it was matched against. If the extracted text does not match what is on the page, it is an extraction problem — re-photograph. If the text is right and the mark is wrong, override it.

The QR code will not scan.

Print at a larger size, or type the URL from the sheet directly. The capture session is a plain web page.

A script was photographed twice.

Duplicates are matched to the same attempt; the most recent capture wins. Nothing is double-counted.