AI Test Grading Feedback: Exercise Breakdown and Grading Areas

You grade tests with photos: point a camera at a student's paper, handwriting and all, and it reads each answer the way a teacher would — works out what the question asked for, judges how well the answer met it, then hands back a grade and written feedback in plain language. What used to be an evening with a stack of papers is now a few minutes reviewing work that's already done.

Every grade and every word of feedback stays a draft until you approve it. Edit anything you'd phrase differently, and only then does it become the real record — whether you're grading one paper at a time or a whole class set in one pass. We built it that way on purpose: fast isn't worth much if you can't trust what it hands back.

That plain grade-and-feedback flow is still the default every time you grade — we didn't want to slow that down. What changed is what sits above the Grade button: the same setup panel where you paste grading instructions and an optional answer key now carries two more settings, one after another — grading areas, then a checkbox for the exercise breakdown. Both optional. Neither adds a step to grading itself.

They look at a test from two different directions, and I think that's the right way to split it. Turn on the exercise breakdown and you get a vertical read: every exercise, listed one under another, each with its own points and its own note. Add grading areas and it's horizontal instead — one pass across the whole test, scored against whatever qualities you actually care about: structure, evidence, grammar. Together they answer two things a bare grade never could: exactly where the points went, and how well the work held up against what actually mattered.

Switch on "detailed exercise-by-exercise analysis" before you grade, and each exercise becomes its own editable row: the points earned against the points possible, and a short note specific to that answer — exactly the kind of comment a student can point to and say "that's what I missed."

Every row stays editable before you save. If it reads a messy answer as fewer points than you'd give it, correct that one row — the note next to it is yours to sharpen too. It's a first draft of exercise-by-exercise feedback, not a verdict you have to accept whole. I'd rather it hand you something to edit than something you have to trust blindly.

Grading areas are closer to your own rubric than a per-exercise tally. Type a short list of qualities — structure, use of evidence, grammar, argument clarity, whatever the test is actually meant to measure — and it reads the whole paper against each one, returning a percentage and the concrete reasoning behind it: which answer, which sentence, why.

The rule that makes this worth trusting: areas describe the work, they never move the grade. A test can score high on content and lower on structure, and the grade stays exactly what it would've been with no areas listed at all. That's deliberate — the same way a good rubric keeps criteria and final score distinct — so the feedback a student reads can be specific about what to work on, without also relitigating what they scored. It's worth saying plainly, too: on an unusual or ambiguous answer, the AI's read on an area can be off. That's exactly why the reasoning is there next to the number — so you can see why it landed where it did, not just take the score on faith.

Yes, once it's saved — and that's the part I actually care most about here. Feedback that only exists as a comment on one paper is useful exactly once. Feedback saved into a student's record in Classes keeps working, because it's what gets read back the next time you generate something for that student.

The assessment generator has a checkbox, "use this student's history for context and continuity," and with it on, the generation reads that student's recent graded tests in full, not just the grade. An exercise note about a shaky proof, or a grading area that came back low on structure, becomes the signal for what this student specifically needs next — a self-assessment prompt, reflection questions, a fresh quiz that leans into what they got wrong, or a term-end summary of how they progressed. Personalized by what actually happened, not a generic template.

That's the part a one-off comment can never do. One detailed grading pass becomes the start of the next several things you make for that student, instead of a paragraph that gets read once and forgotten. It's the whole reason I built the app around one connected record instead of a pile of separate tools.