How do you set up a marking rubric in JeddAI?

What makes a criterion markable?

A criterion is markable when it names one observable thing in the work and nothing else. “Analysis” is not markable. “Analysis of how language choices shape meaning in the text” is, because it tells a marker where to look. Susan Brookhart makes the same point about substance: criteria should describe the learning rather than the task, so “states a compelling thesis” beats “has five paragraphs”. Surface features are easy to count and tell you almost nothing about quality.

The test worth applying is whether two teachers in the same faculty, marking the same script independently, would point to the same sentences as evidence for that criterion. If they would not, it is a mood rather than a criterion. Bundled criteria fail this test most often. “Structure and expression” is two judgements sharing one mark, and the mark ends up averaging them silently. Split it into two rows.

Four to six criteria is usually enough for an extended response in Years 7 to 12. Past that you are slicing the same judgement thinner without adding information, and the rubric takes longer to apply than the marking it was meant to speed up. What you are writing down is the judgement an experienced colleague brings to expert essay marking: the specific, statable reasons one script sits above another.

• One idea per criterion, named in the language of the work rather than the task instructions.
• Every criterion carries a mark or a weighting, and those add up to the task total.
• No two criteria overlap. If they do, a strong response gets rewarded twice for the same thing.

How many performance levels should a rubric have?

Three to five, and four is the common default. Jay McTighe, Susan Brookhart and Thomas Guskey describe a rubric as three things together: criteria aligned to learning goals, a performance scale of typically three or four levels, and descriptors that differentiate those levels. More levels feel more precise and usually are not. Once the gap between adjacent levels is narrower than the difference two markers would argue about, the extra levels only add noise.

The scales you already report against are a sensible guide. In New South Wales, the Common Grade Scale runs A to E, five levels, and teachers make a holistic on-balance judgement against those grade descriptions. HSC results are reported in six performance bands, with four bands for extension courses. In Queensland the choice is largely made for you: teachers mark internal assessments with the instrument-specific marking guide provided in the syllabus, and that guide already fixes the performance levels.

Anchor the top and bottom levels first. They are the easiest to write because you have seen plenty of both, and the middle levels can then be defined by what they lack or nearly reach. Check the arithmetic too: if the criterion marks do not add to the task total, you end up reconciling it by hand.

How do you phrase descriptors so two markers agree?

Write each level as a description of the work, not a verdict on it. “Excellent response” is a verdict. “Sustains a coherent argument with precise, well-chosen textual evidence at every stage” is a description. The same authors make the useful point that a good rubric turns judging into matching: the marker holds the script against the descriptions and finds the closest fit, instead of forming an impression and then justifying it.

Keep the descriptors parallel across levels. Each level should use the same nouns in the same order, and change only the words carrying quality. Sustained, then mostly sustained, then uneven. Precise, then generally apt, then loosely relevant. When the vocabulary shifts between levels, markers start comparing different things at each level and the boundaries blur, which is exactly where disagreement and inconsistency creep in.

Avoid marking by quantity alone. “Uses three quotations” can be satisfied by a weak response, so tie quantity to function instead: “integrates well-chosen quotations to support each point”. A rubric written this specifically also converts straight into student-facing study guides, because the descriptors already say what a strong response does rather than what mark it earns.

How do you align a rubric to the syllabus you report against?

Start from the standard you will report against, not from the task. Across Foundation to Year 10, the Australian Curriculum’s achievement standards describe the expected quality of learning students should typically demonstrate by the end of each year or two-year band. Lifting the verbs and nouns from the relevant standard into your criteria keeps the rubric defensible when a mark is questioned.

The senior years differ by state, and the difference changes how much you write. The VCAA publishes performance descriptors for VCE outcomes. They are advisory, written to support teacher judgement, and teachers may customise them in line with the VCE Administrative Handbook and the VCE assessment principles. Queensland works the other way for General subjects: the syllabus supplies the instrument-specific marking guide teachers mark the internal assessment with, built on the assessment objectives and performance levels of that instrument, and comparability across schools is then checked through QCAA’s confirmation process.

So the first question is which of three situations you are in. In Queensland the marking guide for an internal assessment already exists and the work is applying it consistently. In Victoria you are adapting published descriptors to your task. In New South Wales you are building school-based criteria that eventually roll up into an A to E grade or an HSC band. Knowing which one you are in tells you how much freedom you have before you type a word.

How do you set up your rubric in JeddAI?

Setting up a rubric in JeddAI means entering the criteria and descriptors your faculty already agreed on rather than picking a template. Paste your own wording, including the mark allocation each criterion carries, so the drafted marking runs against your standards instead of a generic model of what an essay should be. If several classes sit the same task, enter the shared version once so every class is judged the same way.

Give each criterion its levels and a descriptor at each level, then set the total the task is marked out of. Where a criterion comes from a syllabus outcome or an achievement standard, say so in the descriptor. JeddAI drafts a mark and a comment against each criterion using that wording, and every draft goes to you to review, edit or reject before a student sees it.

Rubrics, success criteria and comment banks do different jobs and are worth keeping separate in your head. The rubric decides the judgement. Success criteria explain it in student-facing language. A comment bank keeps the phrasing consistent across a class. Build the rubric first, because it does most of the work, then add the other two as you notice the sentences you type again and again.

• Enter criteria in your own words, one idea per criterion.
• Set the levels for each criterion and write the descriptor for each level.
• Allocate marks per criterion and check they sum to the task total.
• Run it on one assignment first, then compare the JeddAI drafts against how you would have marked those scripts.

How do you refine the rubric after the first marking run?

The edits you make are the data. If you nudged the same criterion’s mark on a third of the class, the descriptors at that boundary are not separating the work. If you rewrote the same comment over and over, the criterion is asking for something its descriptor never named. Fix the rubric rather than the individual scripts, and the next run needs less correction.

Two habits are worth the time either way. Trial a new rubric on two scripts you have already marked before running a whole class: if the descriptors explain why one sat above the other, they are specific enough, and if you are falling back on a gut sense the wording does not capture, that is the missing detail. Then moderate a handful of borderline scripts with a colleague once a term. Whether you are marking by hand or reviewing drafts from JeddAI, agreement between markers comes from the criteria, not from the automation.

Once a rubric works, reuse it. A set of criteria that separates the work well on one persuasive essay will do the same job on the next one, and reusing it is what makes feedback consistent across a year rather than only within a single task.