📥 Marking Inbox

×

How AI marking works

Every student upload is graded by GPT-5.6 against the mark scheme you saved with the worksheet. You always see three numbers on each card:

• AI raw — the % score the AI gave, straight from its own mark scheme comparison.
• Self — what the student ticked as their own confidence in each answer before submitting.
• Effective % — the weighted average (70% AI + 30% self) that goes into the mark book. This is what shows up in the parent-facing progress reports.

What "AI confidence" means

The AI grader also reports how sure it is of its own mark on each question — separate from the grade itself. We collapse those per-question ratings into a single 0-100 confidence number per submission:

• 85-100 · High The handwriting was legible, the working followed a familiar pattern, the AI is confident in the mark. Rubber-stamp with one click.
• 60-84 · Medium Some ambiguity — a step was skipped, notation was unusual, or one line was hard to read. Worth a glance.
• 0-59 · Low The AI is guessing. Photo may be blurry, working may be fragmentary, or a novel solution used. Always double-check these before accepting.

Anything below 60 gets a low confidence chip so it stands out.

Other flags on a card

• low score AI marked the submission under 40%. Look for a misunderstanding you can fix in one comment.
• ai self gap The student's own self-rating is at least 25% away from the AI's mark. Big gap = probably a rubric misread by one of the two.
• very fast Fewer than 2 minutes on a 3+ question sheet. Probably clicked through without doing the work.
• late Submitted after the due date. Late-pass policy is per class.

Tip: the 4pm digest email includes the same flags so you can triage from your phone.