How do you mark NCEA consistently across a class?

Why do Achieved, Merit and Excellence judgements drift?

Judgements drift because the words that separate the grades are qualitative. In depth. Perceptive. Well justified. Those words are easy to read slightly differently on script twelve than on script two. An NCEA grade is a best-fit judgement against an achievement standard, not a mark out of ten, so a small shift in how strictly you read a descriptor can move a student across a boundary.

Three things cause most of the drift. Fatigue, which makes late scripts hard to read with the same eye as early ones. Comparison, where you start marking a script against the script before it instead of against the standard. And unwritten expectations, where half your criteria live in your head and never make it onto the marking sheet. That last one is why two teachers can mark the same script and land on different grades.

The fix is not to mark faster or to care more. Most of how to mark NCEA consistently is settled before you read the first script, by moving as much of the judgement as possible out of your head and onto the page. What separates expert essay marking from tired essay marking is mostly whether the reference point was written down in advance.

What does your rubric need to spell out?

A rubric that says “explains in depth” for Merit is a restatement of the standard, not a marking tool. It sends you back to the same qualitative word you were trying to pin down. Write what a Merit explanation actually contains in your subject: how many links between ideas, what use of evidence, what kind of reasoning.

Then name the boundaries. Most disagreements happen at the line between two grades, not in the middle of one, so the sentence that matters most is the one describing what lifts a script from Achieved to Merit. Write that sentence for both boundaries and you have removed most of the room for drift.

Keep it short enough to use. A rubric nobody reads by script forty is worse than three tight descriptors you actually apply to every script.

• Write separate descriptors for Achieved, Merit and Excellence, not one blurred paragraph.
• Define what sufficient evidence looks like at each grade level.
• Name the boundary: what lifts a script from Achieved to Merit, and Merit to Excellence.
• Translate command words like evaluate or justify into what they look like in your subject.
• Attach a comment bank entry to each criterion so feedback stays tied to the grade.
• Re-check the rubric against the current achievement standard whenever it is revised.

How do exemplars and a calibration pass anchor judgement?

Exemplars do for a marker what worked examples do for a student. The wording of an achievement standard, like the summary in most study guides, describes the destination without showing the route. An annotated Merit script shows what the descriptor means in practice, in the vocabulary of the students actually in front of you.

One annotated benchmark per grade level is enough. Mark it yourself, write in the margin why it sits where it sits, and keep it at the top of the pile. When a borderline script arrives you are then comparing it against a fixed point rather than against whatever you happened to read ten minutes ago.

Calibrate before you commit. Mark five scripts cold, check them against the rubric and the exemplars, and adjust the descriptors wherever your instinct and your written criteria disagreed. That is the same logic behind NZQA moderation of internal assessment: shared reference points keep judgements aligned to the national standard rather than to one marker’s memory on the day.

How do you catch drift while you are still marking?

Drift is invisible from the inside, so build a check that does not rely on noticing it. The simplest one is to re-read your Achieved exemplar partway through the pile. If it still reads as Achieved, you have held the line. If it suddenly looks like a weak Merit, you have loosened, and the scripts you just marked need a second look.

Then sample. Pull four or five scripts from across the grade range once you are finished and check that the grade, the feedback and the descriptor all say the same thing. Look hardest at anything sitting on a boundary, because that is where an inconsistent judgement is both most likely and most consequential for a student.

For internally assessed standards this is also your own moderation work. The grade needs to be defensible against the standard and against the evidence in the script, not against how the rest of the class happened to do.

• Set a fixed checkpoint, say every tenth script, and re-test yourself against the benchmark there.
• Check that the evidence your feedback cites actually appears in the student's work.
• Sample scripts from across the grade range, not just the top and the bottom.
• Give every borderline script a second pass before the grade is released.
• Keep the grade tied to the achievement standard, not to the other scripts in the pile.

How do you keep judgements consistent across a department?

Cross-marking is the cheapest consistency check a department has. Two markers, the same three scripts, grades written down privately before anyone compares. Where you differ, the useful question is not who was right. It is which sentence in the rubric was ambiguous enough to allow both readings.

So fix the sentence, not the person. A disagreement is evidence that the written criteria left room, and that room will open again next term with a different marker. Keep a short log of the boundary decisions you settle, because those become the specific wording your rubric was missing.

Agree on the exemplars as a department too. If everyone anchors to the same annotated Merit script, individual differences shrink to the gap between the exemplar and the standard, which is far narrower than the gap between two teachers’ instincts.

Where does an AI marking tool fit into this?

Everything above is tool-agnostic. A specific rubric, a set of annotated exemplars and a calibration pass will steady your marking whether you work on paper, in a shared spreadsheet or in software. What software changes is how much of that discipline survives script number thirty.

A tool that marks against your uploaded rubric applies the same descriptors to the first script and the last one. It does not get tired and it has no memory of the essay it read ten minutes ago, so the mechanical half of consistency is handled. The judgement half is still yours, and it should stay there.

That is the shape JeddAI takes. You supply the rubric, the success criteria and the comment bank; it drafts a grade and feedback for each piece of work against those criteria; you review, adjust and confirm before anything reaches a student. It carries no separate idea of what Merit means, which is the point. The standard being applied is the one you wrote.

The trade-off is real: a vague rubric produces vague drafts. If grades come back more generous or harsher than you would give them, that is nearly always the descriptors rather than the software. The fix is the calibration habit a department already uses on itself. Tighten the wording, re-run your benchmark script, and check the drafted grade against your annotated exemplar. Teachers who get the most out of JeddAI are the ones who spent an hour on the rubric first.