How can teachers cut marking time on NCEA internals?

Where does the marking time actually go on an NCEA internal?

Two costs dominate, and neither of them is the judgement itself. The first is writing comments that tie each decision to an achievement standard. Internal assessment asks you to show why a piece of work meets Achieved, Merit or Excellence against explicit criteria, so every comment has to carry evidence. Across a full class set, that careful and repetitive writing is where the hours go.

The second is drift. The script you mark at seven in the evening and the one you mark at eleven are not being held to quite the same standard, and you know it, so you go back and re-read the earlier ones. That re-reading is honest work, but it is expensive, and it is caused by how the marking session was organised rather than by anything in the students’ writing.

Formative rounds add a third cost that most teachers underestimate. Before an internal is finally assessed, students usually need feedback on drafts and practice attempts. That is the highest-volume and lowest-stakes marking you do. Fixing the final assessment session is worth less than fixing the three rounds of feedback that lead up to it.

What should you set up before you mark the first script?

Write the assessment schedule before students submit, not once you have started marking. Put the evidence statements for Achieved, Merit and Excellence into your own words, tied to the specific task your class actually did. Every judgement then becomes a lookup instead of a fresh argument with yourself, and that is the single largest saving available on an internal.

Then read before you mark. Take five scripts, read them without writing anything, and pick out a clear Achieved, a clear Merit and a clear Excellence. Those three become your benchmarks and everything else is compared against them. This habit sits underneath expert essay marking in any subject: you are matching evidence to a descriptor rather than reacting to how the prose sounds.

It is also worth checking what NZQA already publishes for your standard. Subject pages carry exemplars, assessment schedules and reports, and there is a process for requesting clarification on a standard. Reading the moderator report for a standard before you mark it is much cheaper than meeting the same point in a moderation outcome six months later.

Why does marking criterion by criterion beat marking script by script?

Take the whole class through the first evidence statement, then the whole class through the second. You hold one descriptor in mind at a time instead of the whole standard, so your judgement drifts less and you stop re-reading the criteria twenty-eight times.

It also shows you the spread. Once you have watched the whole cohort attempt the same evidence statement, you know what the middle of your class looks like, and the borderline calls that used to take five minutes each get quicker because you finally have something to compare them against.

The pattern it surfaces is worth as much as the time it saves. If half the class misses the same descriptor, that is a teaching problem rather than fourteen separate comments. Reteach it in one lesson, or point the class at the exemplars and study guides for that standard, and write the comment once.

The cost is that you handle each script two or three times. That is a good trade on a task with two or three clear criteria. On a long portfolio with mixed evidence it is not, so mark those the usual way.

How do you build a comment bank that actually saves time?

A useful bank is not a list of encouraging phrases. Each entry should name what the evidence shows, which descriptor it meets or misses, and one specific thing to do next. Comments that do only the first part read well and change nothing, and you end up writing the next step by hand anyway.

Build it by capture rather than by planning. Mark one class set the way you normally would and keep any comment you find yourself writing twice. You will finish with fifteen to twenty entries that cover most of what you say on that standard, and they will sound like you, because you wrote them.

Keep it small. A bank of eighty comments costs more to search than to retype, so prune anything you have not used in a year and group what is left under the standard’s evidence statements rather than under topics.

Personalise with one clause. A banked sentence followed by a short quotation from the student’s own work is what stops feedback feeling processed. That clause takes about ten seconds and does most of the work of making a comment feel written for that student.

• Name the evidence, the descriptor and one next step in every entry.
• Capture comments you have written twice instead of planning a bank up front.
• Group entries under the standard's criteria, not under topics.
• Add one clause quoting the student's own work to every comment you reuse.

What keeps internal marking ready for NZQA moderation?

Internal moderation exists to verify grade judgements, and external moderation exists to confirm that assessment decisions are consistent nationally. Both are looking at the same thing: whether the evidence in the student’s work supports the grade you gave against the standard. So the artefact worth protecting is the link between the evidence and the descriptor.

Annotate as you mark rather than reconstructing afterwards. A short note on the script naming the descriptor and pointing to where in the work it is met takes seconds while the work is in front of you, and several minutes once it is not. It is also precisely what a verifier or a moderator needs to see.

Have a colleague verify a sample before results are reported, and choose that sample deliberately: the grade boundaries, anything you hesitated over, and one script from each grade. Record who verified it and when. Store the benchmark scripts and the assessment schedule alongside the samples so the whole judgement can be reconstructed months later.

Where does an AI marking tool fit into all this?

This section is about our own product, so read it as disclosure rather than advice. Everything above works with a pen and a printed schedule. A tool changes which part of the work you do, not whether the work is needed.

What a tool like JeddAI does is produce a first-pass comment set against criteria you have entered, so you edit instead of composing. That is the same shift a comment bank gives you, applied to a whole script at once. Jeddle stores those banks and reuses your recurring phrasing across a class set, which is where the saving compounds on thirty responses rather than three.

It can also suggest where a response sits against Achieved, Merit or Excellence. Treat that as one opinion. You check it against the evidence and make the call, because a reported NCEA result has to be your judgement and moderation examines your judgement. In practice the clearest value is on the formative rounds, where volume is high and nothing is being reported yet.

The limits are worth naming plainly. A model can misread a subtle argument, over-credit a confident but thin point, and it has none of the context you hold on a particular student. Run one script fully before the rest of the set, exactly as you would benchmark by hand, and tighten any descriptor that produced a draft you disagreed with.

What should you change on your next internal?

Pick one lower-stakes class set rather than your largest internal. Write the assessment schedule first, benchmark three scripts before marking anything, then work through the class one criterion at a time and keep every comment you write twice.

Then measure it. Note the time you spend on that class set and on the next one done the same way. Most of the gain shows up on the second and third sets, once the schedule exists and the bank has filled out, so judging the method on its first outing will undersell it.