rubric

name	rubric
description	Build an analytic rubric for a project, presentation, report or thesis assessment — criteria × performance levels, weights summing to 100 — and check the thing that makes a rubric a rubric: that the levels differ by an observable behaviour rather than by an adverb. 'Mükemmel / iyi / orta / zayıf' is a grade with extra steps, and it is reported as one. Use when marking open-ended work, or when a rubric has to be defensible at an objection meeting.
disable-model-invocation	true
argument-hint	[<what is being assessed>] [--check]

/rubric — levels that differ by behaviour, not by adverb

Most rubrics fail the same way. Four columns headed excellent / good / adequate / poor, and four descriptors that are the same sentence with the adverb swapped:

Mükemmel: Yöntem çok iyi seçilmiş. İyi: Yöntem iyi seçilmiş. Zayıf: Yöntem zayıf seçilmiş.

That is not a rubric. It is a grade with extra steps, it gives a student nothing to act on, and at an objection meeting it cannot be defended — because "iyi" and "çok iyi" were never distinguished anywhere.

The check strips the intensity adverbs and looks at what is left. If two levels say the same thing, or if a level says nothing at all, that is the finding.

Step 1: teaching/rubric.json

{
  "title": "EEM402 dönem projesi",
  "criteria": [
    {"name": "Yöntem seçimi", "weight": 40, "outcomes": ["DÇ1"],
     "levels": [
       {"label": "3", "points": 40,
        "descriptor": "Yöntem seçimi verinin yapısıyla ve en az bir alternatifin reddiyle gerekçelendirilmiş"},
       {"label": "2", "points": 25,
        "descriptor": "Yöntem gerekçelendirilmiş ama alternatif tartışılmamış"},
       {"label": "1", "points": 10,
        "descriptor": "Yöntem gerekçesiz seçilmiş"}
     ]},
    {"name": "Sonuçların sunumu", "weight": 30, "outcomes": ["DÇ2"],
     "levels": [
       {"label": "3", "descriptor": "Şekiller eksen etiketli, ölçek birimli ve metinde yorumlanmış"},
       {"label": "1", "descriptor": "Şekillerde eksen etiketi veya birim eksik"}
     ]}
  ]
}

Step 2: write a descriptor a second marker could apply

The test is not literary. It is: could someone else, marking the same work without talking to you, land on the same level?

Instead of	Write
"iyi bir literatür taraması"	"en az iki kaynağı birbiriyle konuşturur; hiçbiri yalnızca özetlenmemiş"
"figures are clear"	"every axis is labelled with its unit, and each figure is referred to in the text"
"yeterli analiz"	"varsayımlar sınanmış ve sınanmayanlar açıkça belirtilmiş"

Each descriptor names something present or absent. That is what makes it observable, and it is also what makes the rubric useful to the student before they submit.

Step 3: check

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

Finding	Meaning
weights ≠ 100	the arithmetic
a criterion with fewer than two levels	nothing to distinguish
a level whose descriptor is only an adverb	"Mükemmel" describes no behaviour
two levels identical once adverbs are stripped	a student cannot tell which one they are in

Step 4: bind it to outcomes

Give each criterion its outcomes. A rubric criterion that serves no learning outcome is marks with no declared purpose — the same finding /exam makes, and the same one an accreditation visit makes when it asks what evidence supports PÇ3.

Notes on thesis and presentation rubrics

• Thesis defence: the criteria that survive an objection are the ones written as artefacts — "the method chapter states the sampling frame" beats "adequate methodology". paper-types skill's rubrics carry the same shape for reviewing.
• Presentation: timing and slide count are observable; "engaging" is not. Pick the observable proxy you actually mean.
• Group work: a rubric marking a group product cannot mark individual contribution. If that is the intent, it needs its own criterion with its own evidence, declared in advance.

What this does not do

• It does not judge whether the difference it finds is the right one. Two levels that genuinely differ by an observable behaviour pass, even if that behaviour is the wrong thing to reward.
• It does not set point values or convert levels to marks.
• It does not mark anything.

Reference

• Script: python scripts/syllabus_check.py --rubric <file>
• Reads/writes: teaching/rubric.json
• Related: /syllabus (the outcomes), /exam, /accreditation, paper-types (review rubrics, same discipline for a different reader)

---

A rubric is defensible exactly to the extent that a second marker, given only the rubric, would grade the same work the same way.