Diff Grading

2026-11-20. Filed in teaching.

How students have been graded has changed a lot over the past two centuries. Early on students recited memorized knowledge as the instructor scowled at them. Later on students produced written artifacts that the instructor sorted into stacks of varying quality. More recently, students have submitted digital artifacts that the instructor marks up for points on several dimensions.

Now we have to change grading again, because artifacts that can be generated by AI are an unreliable measure of learning. Many instructors are regressing to oral exams only to rediscover why they were abandoned: they take up so much time. With artifacts, I can at least scowl in private.

I had an idea for a new grading technique that would save time and still measure learning: diff student-before student-after. Let's call this diff grading. The more edits to the student over the semester, the higher the grade.

Even if I could diff a human, there's a problem here. Some students don't change all that much in a class but still do amazing work. That's because they come in with a lot of prior knowledge. My own children bring home a report every year that measures them in two ways on standardized tests: their absolute scores and their growth scores. They don't do as well on the growth scores because they do well on the absolute scores.

Maybe we need two assessment tracks. At the beginning of the semester we evaluate students for their prior knowledge. If they have a lot, we assess their artifacts. If they have a little, we diff them.