Chapter 4/9 chapters
Building tests, grading, and seeing where students slip
Tests are separate from assignments. Auto-grading, and accuracy by label.
Tests are not assignments. An assignment is work to go away and do; a test measures what has actually stuck, so it carries correct answers and marks.
Writing questions
- Each question has a type, a correct answer, marks and an explanation
- Questions can carry a record label (prepositions, listening, quadratics)
- Correct answers stay hidden from students until they've taken the test
Labels are what make sticking points visible
Accuracy is broken down by label, so “grammar is fine, listening isn't” becomes a number rather than a hunch. When there aren't enough answers to say anything, we say we don't know yet rather than showing a misleading figure.
Grading and releasing
Anything a machine can judge — multiple choice and the like — is graded automatically. Written answers you grade yourself. Release the test once you're done and the student sees their result and the explanations.
AI can draft a test from assignments you've already set and from what a student got wrong. Whatever comes back, read it and edit it before you use it.