AI grading for computer science exams

More than one right answer

Two students can solve the same problem with a loop and with recursion, and both deserve full marks. When the class's written answers are grouped, each correct approach becomes its own group marked correct, so a working solution is not graded down for differing from the key.

Rubric items describe what earns the points, such as a correct base case, a recursive call on the rest of the list, or the running time stated, and the grader applies them one by one with a reason.

Every question type on a CS exam

• Code and pseudocode written by hand, read from the box for each answer.
• Short explanations, such as why an algorithm runs in O(n log n) time.
• Multiple choice and select-all-that-apply questions, scored from the key, with a count of how the class answered each option.
• True or false, fill in the blanks ("a ____ is last in, first out"), and matching questions.
• Sections where students answer any two of three questions, with the best answers counted.

Partial credit you control

Partial credit comes from your rubric items, not from a guess. Give points for the idea and the details separately, add a point adjustment with its reason for a case the rubric did not foresee, and reuse a comment you already wrote for the same mistake.

Change an item after seeing the class's answers and every affected answer is graded again by itself.

Large courses and their staff

Introductory courses are large. Invite your teaching assistants as graders by email; each mark records who gave it. Grade a question across the whole class, look at the score distribution and per-question results on Statistics, and publish approved grades to students, who can see their graded exam and ask for a regrade.

See how GradeOptics grades handwritten exams for the whole process from scan to published grade.