How can NZ teachers reduce time spent marking NCEA assessments?

Where does your marking time actually go?

Time yourself on a set of scripts and the pattern is always the same. Reading each response is quick. Deciding the grade is quicker still, because after the first ten you have calibrated. What eats the afternoon is composing feedback. You find the words for the same weakness for the eleventh time, soften them, tie them back to the criteria, then type a variation of the whole thing again for the twelfth student.

The three stages have very different value. Reading is where you learn what your class actually understood. The grade decision is the professional judgement nobody else can make for you. Composing has the worst ratio of teacher effort to student benefit, because most of what you write is a variation on something you wrote an hour ago. Any honest reduction in marking time has to come out of that middle stage. A shortcut that trims the reading instead will cost you at moderation.

Break a single script into its parts and the imbalance is obvious.

• Reading and understanding the response.
• Matching the evidence in it against the criteria.
• Composing a specific strength and a next step.
• Deciding and recording the grade.
• Keeping wording and standards consistent from the first script to the last.

How do you cut marking time without cutting feedback quality?

Do the criteria work before you mark, not during. Slow marking is usually slow because the marker is still deciding what Merit looks like while judging whether this particular script reaches it. Write out the evidence statements first, in your own words, for each grade. Ten minutes on that before you open the pile removes a decision you would otherwise re-make thirty times.

Then batch. Mark one criterion across the whole class before you move to the next, rather than taking each script end to end. You hold one standard in your head at a time, your judgements stay closer together, and you stop paying the cost of switching frame every few minutes. The same discipline applies whether you are doing expert essay marking for a Level 3 English standard or ticking short-answer evidence in a junior science test.

Build a comment bank while you go. The first five scripts will surface most of the feedback the whole class is going to need. Write those comments properly once, save them somewhere you can find them, then paste and adapt for the rest. The adaptation is where the specificity comes from. Keep the shared sentence and change the evidence you point at, so the student sees their own work in the comment rather than a stock phrase.

Cap what you write per student. One strength tied to something they actually did, plus one next step they can act on, is enough for most tasks. Anything beyond that is rarely read. Push the remainder into whole-class feedback. A five minute walkthrough of the two errors half the class made is worth more than thirty paragraphs saying the same thing in slightly different words.

What does criterion-referenced marking change about the job?

NCEA grades are criterion-referenced. Achieved, Merit and Excellence describe increasing quality of evidence against a fixed standard, not a rank order inside your class. That has a practical consequence for speed. The question you answer on each script is whether this evidence meets the descriptor, which is a far faster question than working out how this student compares with the others in the set.

So write feedback that names the criterion. Something like “you have explained the process but not evaluated its significance, which is what Merit asks for” is quicker to write than a general appraisal, more useful to the student, and much easier to defend at moderation. When every comment traces to a named criterion, a colleague checking your marking can see the reasoning without reconstructing it from scratch. Point students at the standard’s own exemplars and at study guides that spell out what each grade looks like, so they can do some of that mapping themselves before they hand anything in.

Be careful when standards are revised. Every revision changes what the evidence has to show, and marking from last year’s mental model is slow, because you second-guess yourself on every borderline script, and it tends to fail moderation. Re-read the current descriptors and rewrite your evidence statements before you mark the first response of a revised standard, not after the first disagreement.

Which marking shortcuts are safe, and which cost you later?

Not every time saving is real. Some move the cost downstream into re-marking, into moderation queries, or into students who cannot act on what you wrote and hand in the same work again next term.

The safe shortcuts share one property. They remove repetition without removing judgement. Reusable comment banks, batching by criterion, whole-class feedback on shared errors, marking codes tied to your own criteria, and returning drafts with a single next step all qualify. So does designing the task to be markable, with clear question stems and an answer structure that puts the evidence where you can find it.

The costly ones remove judgement instead. Grade-only returns save an hour and teach nothing, because a number carries no information a student can use. Skim-marking the middle band because the grades feel obvious is where most moderation disagreements start. Generic praise that could be pasted onto any script in the pile is worse than silence, since it signals nobody read the work closely.

How do you make marking faster every term, not just this week?

Treat your marking materials as an asset rather than task admin. Rubrics, evidence statements and comment banks are the only part of marking that compounds. Written once and stored properly, they make the second run of a standard faster than the first and the third faster again. Keep them in a shared department folder, named by standard, so a new colleague inherits your calibration instead of rebuilding it.

Calibrate with someone else once per standard. Swap five scripts, mark them blind against the same evidence statements, then compare. Half an hour spent that way removes the drift that otherwise appears in the last third of every pile. It also surfaces the descriptors you and a colleague read differently, which is exactly the ambiguity that slows you down when you are marking alone at nine at night.

Finally, be deliberate about which tasks get full feedback. Not every piece needs a written comment on every criterion. A term with three lightly marked practice tasks and one deeply marked assessment usually produces more learning than four medium-effort ones.

How does an AI marking assistant fit into this?

Everything above works with a pen, a printed rubric and a shared folder. AI marking tools have appeared in New Zealand staffrooms because they attack the same stage the manual techniques attack: the repetitive composing between reading and judging. This section is where the product talk lives, so skip it if you only wanted the craft.

A tool of this kind takes the criteria and success statements you supply, reads a student response, and drafts criterion-linked comments plus a suggested grade. You review, edit and approve. That is the whole mechanism. It does not know your standard unless you tell it, it does not make the final decision, and the quality of what it drafts is bounded by the quality of the rubric you feed it. The front-loaded criteria work described earlier is not optional with these tools. It is the input they run on.

Jeddle is the platform we build and JeddAI is its marking engine, so read this as the vendor’s own description rather than an independent review. It works the way described above. You build the rubric, success criteria and comment bank once for a task, JeddAI drafts against them for each response in the set, and nothing is recorded or returned to a student until you have approved it. Your school’s recording and internal moderation process is unchanged.

The limits are worth stating plainly. The saving scales with class size and with how structured your criteria are, so it is largest on a big class set marked against tight descriptors and smallest on a handful of open-ended pieces. You still read every script, and you still make the call on borderline work. If your rubric is vague, a tool like this will draft vague feedback faster than you could, which is not an improvement on anything.