Teacher guide: AI marking — per-question batches, whole papers, and the review console

ScoreCoach marks step by step against marking points (M marks / A marks): mathematical equivalence is judged by a symbolic engine, method steps by AI reading the marking scheme — and low-confidence results are routed to you for review rather than force-scored.

Which mode when

Workspace → Marking has two entries:

Whole paper: one student's complete answer script (multi-page PDF or photos) in one upload; the system matches questions by the printed question numbers and marks them all. Best for collected paper scripts. Filling in year/paper makes matching more accurate.

Single question: pick one question + one class, upload each student's photo for that question, and mark the whole class in one batch. Best before a per-question debrief.

After uploading, "Recent marking jobs" at the bottom shows queued / running / done / failed in real time — no need to wait around.

The review console: what makes AI marking trustworthy

"Review console" (top right) collects everything that deserves your eyes:

• Low confidence: the AI wasn't sure about a method step;
• Pending deductions (e.g. missing units, u-1): flagged, never auto-applied — you decide;
• Student appeals: score disputes from students or parents land here too.

You see the transcribed working and the reason for every step's marks, then adjust or confirm. Every review is logged — the "AI accuracy" card on the reports page is computed against your reviews as ground truth: the share of reviewed results you kept unchanged. Honest by construction.

Cost reference

Single-question marking: 1 credit per script; whole paper: 20 credits; multiple choice: free. Credits draw from the institution's shared pool, attributed to whoever ran the job.

After marking

Scores flow into the gradebook, refresh each question's empirical difficulty (p-value) and every student's topic mastery. How to read all that: Reading class reports.