exam

name	exam
description	Build an exam or quiz from the course's own learning outcomes and check the coverage: every outcome measured by at least one question, no question serving an outcome the syllabus does not have, points summing to the declared total, and the Bloom distribution reported so an 'apply' course does not get a recall paper. Produces the paper and the answer key. Use when setting a midterm, final or quiz; /syllabus defines the outcomes it measures and /rubric grades open-ended work.
disable-model-invocation	true
argument-hint	[midterm|final|quiz] [--check]

/exam — questions bound to outcomes, and the coverage checked

An exam is where a syllabus finds out whether it meant what it said. The two failures are symmetrical and both are graph problems: an outcome the paper never measures, and a question testing something the course never promised.

Gate

teaching/syllabus.json must exist with its outcomes. An exam written before the outcomes is an exam the outcomes will be reverse-engineered from, which is the thing accreditation exists to catch.

Step 1: teaching/exam.json

{
  "title": "EEM402 Vize",
  "date": "2026-11-12",
  "duration_minutes": 90,
  "total_points": 100,
  "questions": [
    {"id": "Q1", "points": 30, "outcomes": ["DÇ1"], "bloom": "apply",
     "text": "Verilen veri kümesi için k-NN sınıflandırıcısını …",
     "answer": "…", "rubric": "Doğru mesafe ölçütü 10p, …"},
    {"id": "Q2", "points": 40, "outcomes": ["DÇ2"], "bloom": "evaluate",
     "text": "…", "answer": "…"}
  ]
}

Each question declares which outcome it measures. That single field is what makes the rest checkable.

Step 2: write questions at the outcome's own level

The Bloom level of the outcome sets the level of the question. An outcome that says "uygular" and a question that says "tanımlayınız" have not met:

Outcome verb	The question asks for
tanımlar / defines	a definition, a list, a name
açıklar / explains	a mechanism in the student's own words
uygular / applies	a computation, a procedure carried out on new data
analiz eder / analyses	a comparison, a decomposition, a diagnosis
değerlendirir / evaluates	a judgement with its criterion stated
tasarlar / designs	something that did not exist in the question

The reported bloom_spread is the honest version of "was this a fair paper": three recall questions and one design question is a distribution, not an accident.

Step 3: check

python scripts/syllabus_check.py --exam teaching/exam.json

questions            2
computed_points      100  (declared 100)
unmeasured_outcomes  []
bloom_spread         {"apply": 1, "evaluate": 1}

Finding	Meaning
points ≠ declared total	the arithmetic students will do first
unmeasured_outcomes	the syllabus promised it and the paper does not ask
unknown outcome on a question	an exam wider than the course — students are being tested on something never taught
a question measuring no outcome	marks with no declared purpose

Step 4: the paper and the key

Generated in this session from the JSON: the question paper (with points per question and the instructions), and a separate answer key with the per-question mark scheme. Keeping the mark scheme in the same file as the question is what makes a re-grade defensible three weeks later.

For open-ended questions and projects, the mark scheme belongs in a proper rubric: /rubric.

What this does not do

• It does not write the questions for you unasked, and it does not judge whether a question is fair, clear or at the right difficulty. Those are teaching judgements; the coverage is not.
• It does not detect that two questions test the same thing. Both would be bound to the same outcome, which is visible in the report but is not a finding — sometimes that is deliberate.
• It does not set a passing mark.

Reference

• Script: python scripts/syllabus_check.py --exam <file>
• Reads: teaching/exam.json, teaching/syllabus.json
• Related: /syllabus (the outcomes), /rubric (open-ended marking), /accreditation (where these marks become evidence)

---

A question nobody can trace to an outcome is a question nobody can defend at an objection meeting.