Marking Feedback Drafter

Where it works well

It applies every criterion to every submission, in full, and shows the rubric wording behind each call.

Done by hand it is 20–30 minutes a script, and feedback quality drifts as fatigue sets in late in a pile.
Best for high-volume, rubric-marked written work: essays, reports, case studies, reflective journals across large first- and second-year units.
At a cohort of 150, the recaptured hours go back into the students who are actually at risk — not into typing the same comment for the fortieth time.

The slow, invisible part of marking is not deciding the grade — it is reading closely enough to write specific, rubric-anchored feedback on every criterion, for the fortieth script, at the same standard as the first.

Where it works badly

It is confidently wrong on the things a rubric can't capture — and a fluent draft mark looks more certain than it is.

Weak where a criterion needs contextual judgement (genuine insight, voice, a borderline pass/fail) — it should flag for the academic, not commit to a band.
It can read a confident, well-structured submission as a correct one; surface fluency is exactly what a marking model over-rewards.
A similarity score is a signal, not a verdict — it cannot tell legitimate quotation from contract cheating or undisclosed AI use, and must never be treated as proof.

The honest test

If you could not defend this mark to the student, or to a moderation panel, from the rubric and the script in front of you — then the tool has made your marking faster, not sounder, and you are about to own a number you didn't form.

Point it at an assessment where the marking turns on disciplinary judgement — originality of argument, whether a creative or clinical response actually works — and it produces a tidy, plausible mark that the descriptor never really justified. That is the trap.

What it doesn't do — and shouldn't

It drafts feedback for an academic to own. It never assigns the grade.

WHAT IT DOES

Quotes the rubric descriptor and the passage behind each provisional band

Flags criteria it could not judge confidently for the academic to decide

Surfaces an originality concern as a flag to check, not a finding

WHAT IT WON’T

Assign or release the final mark

Make an academic-integrity determination

Send any feedback to the student

A grade is an academic judgement the marker is accountable for, and an integrity allegation has real consequences for a student. The Higher Education Standards Framework (Threshold Standards) 2021 holds the provider responsible for assessment integrity, and TEQSA now expects providers to actively manage the risk generative AI poses to it. The accountable academic stays on the decision because the consequence — for the student and the institution — lands on them, not the tool.

What your data has to look like

A rubric with real performance-level descriptors, and submissions in a form the tool can actually read.

Typical readiness

across orgs we see, before the first job

A rubric with described performance levels

Needs shaping

Submissions as machine-readable text

Usual weak point

Exemplars of marked work at each band

Needs shaping

The originality / similarity report, if you use one

Usual weak point

A clear release and moderation step that stays human

Usually ready

The real first job

The rubric is usually the weak point — held as a marks table with no described bands, or as wording so vague that two markers read it three ways. Turning your rubric into explicit, described performance levels is usually the real first job — larger and more valuable than the drafting layer, and it improves marking consistency whether or not you ever automate it. Once the rubric is sharp, every draft after that is faster and defensible by default.

Right fit if…

You mark high volumes of written work against a structured rubric every semester

Your rubric has real described performance levels, not just a marks-out-of table

Large first- or second-year units where feedback timeliness and consistency slip under load

Sessional and casual markers who need to deliver feedback at a consistent standard fast

Walk away if…

The mark turns on disciplinary or creative judgement a rubric can't capture

Submissions are handwritten, scanned, or otherwise not machine-readable text

You want a tool that decides academic-integrity cases or assigns the final grade

Your "rubric" is a weighting table with no described bands to mark against

Open questions

The worried-buyer questions, answered straight

It can draft a wrong band — which is exactly why it drafts and never releases. For each criterion it quotes the rubric descriptor and the passage it judged, and flags anything it could not assess confidently, so the academic checks the reasoning before signing off. An originality or similarity score is surfaced as a flag to investigate, never as a finding — an academic-integrity determination under your provider’s policy is a human decision, not the tool’s. The tool shows its working; the academic owns the mark.

It is only as good as the rubric and the text it reads. If the rubric is a marks-out-of table with no described bands, the draft has nothing concrete to anchor to; if submissions are scans or images of handwriting, it can’t read them criterion by criterion. Sharpening the rubric into described performance levels and getting work in as machine-readable text is usually the first piece of work — and the piece that improves marking consistency across every script after, with or without the tool.

No. It removes the close-reading-and-typing load so the academic spends their time on the judgement: whether the argument actually holds, whether a borderline script passes, what feedback will move this student forward. The mark, the feedback and the accountability stay with the academic, who corrects and releases every draft. The capacity it frees goes back into teaching and the students most at risk.

It drafts against whatever rubric you give it, so it must be the one in force for this assessment in this teaching period. Point it at last semester’s rubric, or a version before a moderation meeting changed a band, and it will mark confidently to the wrong standard. The honest test: can you say, right now, that this is the exact rubric and weighting students were assessed against for this task?

Student submissions, marks and feedback are personal information about an identifiable student and must be handled under the privacy law that applies to your provider — the Privacy Act 1988 and the Australian Privacy Principles for private providers and ANU, or the relevant state or territory privacy law for most public universities. Any deployment runs against your own systems and data handling, not a shared pool — we scope where the data sits and who can see it as part of the build. The demo runs entirely on fabricated data; Jordan Nguyen is not a real student.

It is a marking-support tool, not a grade engine, and it is built to keep the academic accountable for the mark — which is what the Higher Education Standards Framework (Threshold Standards) 2021 and TEQSA’s assessment-integrity expectations require. It does not make integrity determinations; it surfaces a similarity flag for a human to investigate under your provider’s own policy. Using AI to assist marking should itself sit inside your institution’s gen-AI and assessment policy, and we scope the audit trail so each draft and each human sign-off is recorded.

What it takes to build

3–4 weeks · 4 phases

Reused from template~70%

Bespoke to this skin~30%

stack · Claude · structured rubric rules · review UI

What it would cost

Fixed scope, fixed price, fixed dates.

Bite-sized first piece

One contained change, low risk

Pilot build

Most builds land here

Embedded support

Scale on proof

Considering this for your unit or faculty?

The honest place to start is a bite-sized first piece — one rubric, one assessment, low risk. Tell us where the marking load hurts; we'll play it back, scope it, and show you what's possible.

Book a call How we work