A rubric tells the AI what earns points, but not how strictly to apply it. You can now show it, by uploading answers your teachers have already graded. POST /exams/{id}/questions/{questionId}/calibration-answers takes up to 1,000 answers per call, each with the pointsAwarded and, ideally, a short reasoning.
When grading, Examplary picks the most representative of these examples and uses them to decide on partial credit, so feel free to send as many as you have. Graded on a different scale? Set pointsToEarn and the score is rescaled. Calibration answers can't be changed or removed once added.
The new calibration guide covers choosing good examples.