How Masterly grades
What the AI looks at, when humans should override, and why some questions don't show a grade at all.
Three flavours of grading
Different question types call for different grading. We use the simplest tool that works for each.
- Multiple choice and number answers - graded automatically. Instant. A number counts by its value, so 0.5 and 1/2 are both right - unless the question asks for a form, like "in simplest form" or "as a decimal".
- Before any question reaches your student, its answer is checked: number answers are worked out again exactly, and a question whose answer key is wrong is fixed or rewritten.
- Short answers - checked against a list of accepted phrasings the AI generated when the question was created. The AI handles obvious variations (synonyms, punctuation).
- Extended writing - graded against a rubric (a list of criteria like 'has a clear topic sentence' or 'uses correct subject-verb agreement'). The AI scores each criterion separately and writes feedback. You can preview the rubric on any approved question.
Self-checks and partial credit
For activities labelled as Teach, we sometimes ask students to self-check - they see the worked example and judge their own answer. These don't feed mastery directly; they're for learning, not assessment.
Partial credit happens when a writing answer hits some criteria but not others. We surface which criteria were met so the student knows what to work on.
What counts towards mastery
Every answer earns XP, but only some answers count as proof that your student can do a skill on their own.
- A right answer with no hint counts towards mastery.
- A wrong answer resets the run of right answers, as you'd expect.
- A partly-right answer, or a right answer given after a hint, doesn't count either way. It shows progress with help, not that they can do it alone, so we ask again another day. History marks these answers "With a hint" or "Answer shown in a hint".
- A lesson is judged on all of its Test questions together: right on at least two out of three counts as a pass, wrong on more than one in three counts as a miss.
- Where a skill can be answered by typing, at least one of the right answers has to be typed rather than picked from a list. If the run is all multiple choice, the next practice question asks for a typed answer.
When the AI gets it wrong
We're not perfect. If you spot a misgrade, you can change the mark yourself. Open the session in History, expand the question, and choose Change mark. Pick "Mark as right" or "Mark as not right" and, if you like, add a short reason.
The original mark stays visible beside yours, with your name and the date, so the history always shows what happened. Your student sees that you changed it too.
- Progress only changes if it was their most recent answer for that skill. If they've answered that skill again since, your change is recorded but doesn't rewrite what came after.
- A right answer given with a hint still doesn't count either way, so marking one right won't move their progress.
- XP doesn't change.
Limits on the Free plan
Free accounts get 200 AI-graded responses per month. Multiple choice and numeric questions don't count toward this - they're free.
"This looks wrong"
Under every marked answer, and on each lesson page, your student can tap "This looks wrong" to flag a question that doesn't make sense, an answer key that looks off, or a mark they think is wrong. You can do the same from History.
A report doesn't change the mark by itself. It comes to us and we review it. When your student sends one, you get a notification that takes you to that session, so you can look at it and change the mark if you agree.
What to read next
Mastery, spaced review, and regression
Why we keep bringing back skills your student has 'finished', and what regression means.
Staying in the loop
The Sunday email, the nudge when sessions stop, and what happens when your student flags something.
Your data and privacy
What we store, what we share, and what happens if you delete your account.