How to write a rubric before you write the items
How to write a rubric in five steps — criteria, levels, descriptors, weights, a test run — and why the rubric should exist before the first item is written.
How to write a rubric comes down to four decisions — criteria, levels, descriptors and weights — tested on sample responses and made before a single item is written. Most programmes work the other way round: questions first, rubric the week before marking. This guide, for heads of assessment and programme directors, gives the five steps and then makes the case for the order. On NUADU the rubric an item is written against is the rubric the marker applies and the learner's feedback cites, so the order decides what every later record can say.
What should a rubric include?
A rubric is a scoring guide that names what a response is judged on and describes what performance looks like at each level. Whether you call it an assessment rubric, a marking scheme or a scoring guide, a usable one has four parts.
- Criteria: the qualities being judged, such as argument, use of evidence or accuracy.
- Levels: the bands of performance each criterion is divided into.
- Descriptors: what a response at each level actually does.
- Weights: how much each criterion counts towards the result.
Rubric design guides separate analytic rubrics, which score each criterion on its own, from holistic rubrics, which give one overall judgement. Where a mark may have to be explained later, the analytic form is easier to defend, because a change can point at the criterion it affected.
None of this is specific to any software. What a platform can do is stop the four decisions drifting apart once they are made. That is the job of rubric-anchored authoring, one of the four features of exam authoring on the Create pillar.
How to write a rubric in five steps
Call it a marking rubric or a scoring guide; either way, write it in this order and finish all five steps before drafting the first item.
- Name the criteria: list the qualities the task must show — analysis rather than recall, coherence rather than length — and cut any a marker could not observe in a response.
- Choose the levels: decide how many performance bands each criterion needs, and keep the scale consistent across criteria unless one genuinely needs more.
- Write the descriptors: for each level, describe what a response does, in words a marker and a learner can both apply, and say what is present rather than only what is missing.
- Set the weights: decide how much each criterion counts, and record who agreed it, because a weight is a judgement about what the programme values.
- Test it on sample responses: mark a few real or mock answers with a colleague and rewrite every descriptor the two of you read differently.
Descriptors are where most rubrics go wrong, so here is an illustrative example. Take a criterion called "use of evidence", divided into three levels, and write each level as what the response does:
- The response states its claims without support.
- The response supports some of its claims with relevant evidence.
- The response supports each main claim with relevant evidence and explains how that evidence bears on the argument.
Each line describes something a marker can point to in the script. None of them uses "good", "adequate" or "excellent", words two markers will read differently by the second batch.
One test is worth running while you write. If a criterion cannot be described at more than one level, it is not a criterion yet. It is a checkbox, and it belongs in an objective item.
Many guides add a sixth step — share the rubric with learners before they start — and it is good advice. The step most guides leave out comes before all of them.
Why write the rubric before the items?
Because an item written first tends to test what is easy to ask, while a rubric written first states what the programme wants to see. Write the criteria, then write each item to elicit them. When marking starts, the marker applies a standard that existed before any response did, which is much harder to argue with on appeal.
"We have a stack of items and a marking scheme that doesn't quite match them." Every head of assessment has heard a version of that sentence, usually from a marker, usually too late.
NUADU's Create pillar keeps the rubric attached to the item from authoring onwards. Three places show it:
- In the editor, items are written against rubric criteria, and items imported from an existing bank can bring their rubric structure with them.
- With AI drafting switched on — opt-in per institution, per programme and per assessment — the Content Scanner proposes a draft rubric alongside each candidate item. An AI-drafted rubric is a starting point; the educator finalises it.
- At review, the educator adjusts rubric criteria and weights before approving anything, so every weight is a decision with a name on it, not a default accepted by omission.
What else that review checks is set out in how AI-drafted exam questions are reviewed.
Rubric-anchored from the start. Same rubric, end to end.
One rubric from the item to the decision record
The rubric is not re-keyed for marking or summarised for a dashboard; it is the same object at every stage. With AI assistance on, essays and spoken responses receive draft scores and rubric-anchored explanations against it, and the marker reviews, overrides and signs off — the workflow behind rubric marking on the Grade pillar, and the subject of our post on AI-assisted marking. The learner then receives feedback tied to the same criteria with the marker's name attached, as rubric-based feedback describes. In Report, any figure can be followed back to the response, the marker, the item and the criterion it rests on.
The decision record is where the order pays off. The home page shows an illustrative decision record — a piece of interface, not a customer's data — with a result marked signed, an approver, a timestamp, and an override line naming a criterion, the old score, the new score and the fact that a reason was recorded. That line means something only if the criterion was defined before the item was written and reached the marker unchanged. An appeal panel told that a score moved up a band wants to know on which criterion, against which descriptor, and by whom. A rubric written first makes the answer short: approver named · timestamp written · override recorded.
Frequently asked questions
How do you write a good rubric?
Keep each criterion observable, give every level a descriptor that says what a response does, and set the weights deliberately. Then test the rubric on sample responses with a colleague and rewrite whatever the two of you read differently. A good rubric is one that two markers apply the same way.
What is a rubric and how is it used?
A rubric is a scoring guide made of criteria, levels, descriptors and weights. Markers use it to judge responses consistently and learners use it to see what is expected. On NUADU one rubric serves authoring, marking and the feedback the learner reads.
What is the best way to create a rubric?
Start from the outcome you want evidence of, write the rubric before the items, and draft it with the people who will mark against it. When you create a rubric for assessment with AI assistance on, treat the draft as a starting point: the educator finalises the criteria and weights at review.