How do you mark HSC-style extended responses consistently?

What does a NESA marking guideline actually ask you to do?

In HSC marking, the guideline for each extended-response question sets out mark ranges rather than a checklist of points. Each range carries a holistic description of what a response at that level demonstrates. NESA’s published principles say the guidelines must indicate the quality of response required to gain a mark or a sub-range of marks. You read the whole answer, decide which description fits best, then settle on a mark inside that range.

The descriptors move through a graded vocabulary. Top-range wording tends towards skilful or sophisticated, the next towards effective, then sound, then limited or elementary. Because the judgement is holistic, one strong paragraph does not carry a script and one weak paragraph does not sink it. Two other principles matter on the page: the guidelines use language consistent with the subject’s outcomes and band descriptions, and high achievement is not defined solely in terms of the quantity of information provided. A long answer is not automatically a strong one.

The principles also make room for responses that do not follow the expected shape, including flair, originality and creativity, or alternative solutions where the question allows them. Worth remembering when a student writes something good that your criteria did not anticipate. The guideline describes quality; it is not a template.

It helps to separate two ideas that both use band language. Question-level marking guidelines are what you apply to an individual script while you mark. Performance bands, Band 1 through Band 6, are how course results are later reported, each holistically describing what students at that level typically know and can do. Staffroom talk slides between the two constantly, but only the marking guideline belongs on the script in front of you.

How do you turn a mark range into criteria you can mark against?

The published descriptor is a starting point, not a finished marking tool. Before you mark the first script, write out what a response has to show to sit in each range, in your subject, for this question. For an English essay that might be a sustained thesis, textual references integrated into the argument, analysis of how the composer makes meaning rather than what happens, and control of language. For a History extended response it might be specific content, a clear line of causation, use of sources, and a judgement that answers the question asked.

Keep the evidence observable. “Sophisticated analysis” is not something you can point to on the page. “Explains the effect of a technique rather than naming it” is. Vague descriptors are where marker disagreement lives, and where students lose faith in the mark. A useful test: if you cannot show a student the sentences that put their response in one range rather than the next one up, your criteria are not concrete enough yet.

• Write the top range first, then the bottom range, then fill in the middle. The extremes are much easier to describe.
• For each range, name two or three things you could underline on the page as evidence.
• Separate the ranges by quality of handling, not by how many features are present.
• Record question-specific requirements separately, such as the number of prescribed texts or the use of the stimulus.
• Keep the wording close to the syllabus outcomes so the mark and the report comment tell the same story.

How do professional markers stay consistent across hundreds of scripts?

The HSC marking operation is worth copying at faculty scale. Extended responses are marked by two or more markers working independently, and when their marks differ considerably the response goes to a senior marker to resolve. Before that starts, markers are briefed on applying the guidelines consistently and practise on a range of responses.

The checks continue while marking runs. A sample of each marker’s work is check-marked. The whole team marks a common script to expose disagreement. Statistical reports flag markers whose patterns look unusual, and anyone drifting is re-briefed. None of this assumes markers are careless. It assumes a human standard moves over a long pile of papers, which it does.

No faculty can run a marking centre, but a light version of this takes about an hour and catches most of the same problems.

• Choose three benchmark scripts before you start: one clearly top range, one solidly middle, one weak. Mark them as a faculty and keep them on the desk.
• Have every teacher mark the same two scripts cold and compare marks and reasons before anyone touches their own class.
• Mark question by question across the whole class rather than student by student, so one standard stays in your head.
• Re-read your first five scripts after you finish the pile. Early marks drift, usually upwards.
• Second-mark the borderline scripts blind, with the first mark hidden.

Should you mark holistically or with an analytic rubric?

The HSC marking guidelines are holistic, so the mark you enter should reflect the response as a whole against a range. An analytic rubric that scores traits separately is still useful, but as a feedback device rather than a way of arriving at the mark. Adding up trait scores can produce a total no marker would have awarded on a holistic read, because it lets a strong structure compensate for an argument that never arrives.

Plenty of faculties run both, which is reasonable: mark holistically to decide the mark, comment analytically so the advice is specific. The one rule is that the two must agree. If the comments say the thesis is undeveloped while the mark sits in the top range, students learn to read the number and ignore the comments.

What does useful feedback on an extended response look like?

A mark range tells a student where they landed. It does not tell them what to do on Monday. The most useful feedback on a long answer names one thing the response did well with the evidence attached, and one change that would move it towards the next range, written as an action rather than a quality. “Add the effect of the technique after each quotation” is actionable. “Deeper analysis needed” is not, and students who get it usually write the same essay again, longer.

Two comments beat ten. Students who see what expert essay marking looks like applied to their own writing improve faster than students handed a number and a paragraph of general encouragement. If you have five minutes, read the top-range benchmark aloud next to their own paragraph. The contrast teaches more than the comment does.

Then give the advice somewhere to land: the question again, one rewritten paragraph with a deadline, or the module’s study guides before the next task. Feedback that ends at the bottom of the page is feedback the student files away.

Where does an AI marking tool fit into this?

This is the part of the article about software, so here is the honest version. A tool like JeddAI reads the whole extended response, matches it against the criteria you entered, then proposes a mark range and draft comments anchored to specific lines. It applies your descriptors rather than a generic idea of good writing, which is why the criteria work described earlier decides the quality of the output. Vague descriptors in, vague marking out.

The gain is not speed on the first script. It is consistency on the twentieth, when your standard has quietly moved and you have stopped noticing. A drafted comment you edit beats an empty box, and shared criteria mean two teachers start from the same descriptors rather than two private interpretations. Because JeddAI drafts against your own rubric and comment bank, the comments come out in language you already use.

The limits matter as much as the features, and they apply to any AI marking tool. It is not the marker of record; you are. Read the script, compare it with the proposed range, and move the mark where your judgement differs. If a whole batch drifts high or low, the fix is in your descriptors. Used that way it does the job of a well-briefed second marker: a first pass you interrogate, not a verdict you accept.