← Documentation

Rubrics and grading

Updated on August 5, 2026

The rubric tells the AI what to expect from a good answer. Without one, the AI evaluates with general judgment; with one, it applies the exact criteria you define.

What a rubric is

A rubric is a set of criteria that describes, point by point, what a correct answer should include. The AI reads them alongside the statement and the student's answer, then assigns a score and writes feedback.

Rubrics apply to the Free text and Step by step answer types.

How to add criteria

In the editor, press Add criterion in the Grading criteria section. Each criterion has three fields:

  • Criterion — the name of the criterion, for example "Identification of variables" or "Accuracy of the calculation".
  • Importance — a number from 0 to 100 indicating how much this criterion weighs in the final score. The values do not need to add up to 100; the AI uses them relatively.
  • What a good answer must include — a description of what you expect for this criterion. The more specific, the more accurate the grading.

You can add as many criteria as you need. To remove one, press the delete button on its card.

How it translates into a score

After receiving the student's answer, the AI assigns each criterion a score from 0 to 100 and weights it by the Importance you defined. The final result is an overall score from 0 to 100.

The overall score determines the outcome:

  • Passed — high score, the answer meets the criteria.
  • Partial — mid-range score, the answer meets some criteria.
  • Failed — low score, the answer does not reach the criteria.

The exact thresholds between these three levels are fixed in the system.

How to write criteria that produce useful feedback

The quality of the grading depends on the clarity of the criteria. A few tips:

  • Describe behaviors, not outcomes. Instead of "the answer is correct", write "the student identifies the net force and applies Newton's second law".
  • One criterion per idea. Do not mix two requirements into one criterion; that way the feedback can pinpoint exactly which one fails.
  • Use the description field. The criterion name is short; the description is where you give the AI context.
  • Set the importance to reflect the real weight. If reasoning matters more than the numeric result, give the reasoning criterion a higher importance.

The rubric is optional. If you add no criteria, the AI still evaluates the answer using its general understanding of the statement. Adding even one criterion noticeably improves grading accuracy.

Frequently asked questions

Can I reuse the same rubric across multiple problems? There is no saved-rubric feature. You must configure criteria in each problem separately.

Is the feedback always in the same language as the statement? Yes. The AI writes feedback in the language configured for the problem.

What happens if I add no criteria? The AI still grades, but without specific guidance the feedback may be more generic. For complex open-ended questions, it is better to add at least one criterion.

Next step