Every examination office knows the shape of the problem: a year's worth of theses, dissertations and capstone projects arrives in a window of a few weeks, and the same three or four people who handle routine cases all year are suddenly expected to screen all of it — carefully, consistently, and on a deadline the graduation ceremony does not move.
The failure mode is rarely the software. It is attention. When every report gets the same manual once-over regardless of what it contains, reviewers spend most of their hours confirming that unremarkable submissions are unremarkable, and the genuinely difficult cases get whatever concentration is left at 6 p.m. The playbook below is how well-run offices invert that: decide in advance what the evidence must look like for a document to be waved through, escalated, or investigated — then let the queue sort itself so human judgment lands exactly where it is needed.
Before the season: agree your policy bands
The single highest-leverage meeting of the season happens before the first document is uploaded. Get the integrity office, the graduate school and one or two experienced examiners in a room and agree three bands — not thresholds plucked from folklore, but definitions written in terms of what the report actually shows once quotes, bibliographies and trivial matches are accounted for.
| Band | What the report shows | Who acts | Outcome |
|---|---|---|---|
| Green | Residual similarity is scattered, short, and attributable to common phrasing or properly cited material | Nobody — the system | Certificate issued automatically; document archived |
| Amber | One or more concentrated matches, an unusual source pattern, or an elevated AI-writing signal worth a human look | A trained second reader | Cleared to green with a note, or escalated to red |
| Red | Sustained overlap with a single source, collusion indicators, masking attempts, or second-reader escalation | The integrity office | A case file opened under the institution's formal procedure |
Two details matter more than the exact boundaries. First, the bands must be defined on the filtered report — after quoted material and reference lists are excluded — or you will drown in false ambers from well-referenced literature reviews. Second, write the definitions down and date them. When a student appeals in October, the question will be whether the May decision followed the published procedure, and a dated one-page band definition answers it.
Intake: everything through one door
Scattered intake is where seasons go wrong quietly. If some departments email PDFs, others upload to a shared drive, and a third group submits through the LMS, you will spend the busiest week of the year reconciling filenames instead of reading reports. Pick one intake path and make it the only one.
- Batch upload, not drip-feed. Have department coordinators submit in named batches — one per programme or defence cohort — so progress is measured in units the faculty actually recognises.
- Metadata at the door. Capture programme, supervisor and cohort at submission time. Every downstream report, certificate and council statistic inherits it for free.
- Receipts, immediately. A student who can see that their thesis was received and is in the pipeline does not email the office to ask — multiplied by a few thousand students, that is a real staffing decision.
This is the first of two places where tooling genuinely changes the shape of the season: iOriginally's Quick Submit bulk intake and student portal exist precisely so that a coordinator can drop an entire cohort in one action and every student gets a live pipeline status and a receipt without anyone answering mail.
Triage: let the report do the first read
With bands agreed and intake flowing, triage is mechanical by design. Green documents should never be opened by a human — that is the whole point of having defined the band. Amber is where trained attention goes, and the second reader's job is specific: not to re-litigate the percentage, but to read the flagged passages source by source and answer one question — does this pattern have an innocent explanation?
Equip readers to do that quickly. Filters that exclude quotes, bibliography and small matches — and recompute the score live — turn a twenty-minute skim into a two-minute targeted read. Seeing findings annotated on the original page layout, rather than as stripped text, matters too: a matched paragraph inside a properly formatted block quote reads very differently from the same paragraph presented as body text.
Treat AI-writing signals with the same discipline but a different posture. A sentence-level AI probability is a reason for a conversation about the assessment's permitted-use policy, not a similarity finding, and your amber procedure should say explicitly which policy applies to which programme.
Staffing and turnaround: do the arithmetic early
Most turnaround promises fail because nobody measured anything before making them. The fix is a pilot batch: in the quiet weeks before the season, run a representative set of last year's submissions through the exact workflow, and time it. That gives you three numbers that are yours, not a vendor's — the share of documents landing in each band, the minutes an amber read actually takes your readers, and the hours a red case consumes before it leaves the screening queue.
- Multiply expected volume by your measured amber rate to size the second-reader pool.
- Hold a calibration session: all readers band the same ten documents, then argue until the bands mean the same thing to everyone.
- Schedule readers in short daily blocks rather than marathon days — banding quality decays fast with fatigue, and consistency is the product.
- Name a single escalation owner so red cases move to the formal procedure the day they are identified, not at season's end.
Communicating timelines to departments
Departments do not need the screening to be instant; they need it to be predictable. Publish a simple service commitment before the season opens — when batches must arrive, when green certificates are issued, how long an amber review takes, and what happens (and who is told) when a document goes red. Then report progress against it in units departments care about: cohorts cleared, not documents processed. A weekly one-line status per programme ends almost all chase-up correspondence.
Be equally explicit with students. A visible pipeline status and a receipt at submission, a certificate on clearance, and a clearly described process if questions arise — that transparency is not a courtesy, it is what makes the whole exercise defensible.
Archiving: close each document with a certificate
A screening season should end with an artefact, not a spreadsheet. Issuing a certificate for every cleared document does three jobs at once: the student gets proof their work was screened, the graduate school gets a closing record for the file, and — if the certificate is QR-secured with a public verification page — any downstream institution, employer or journal can confirm its authenticity in seconds without emailing your office. This is the second place tooling earns its keep: iOriginally issues verifiable certificates automatically for green-band documents, so the archive builds itself as the queue drains.
Archive the submissions themselves into the institutional repository as they clear. Next season's collusion checking is only as good as the corpus it can compare against, and the cheapest time to build that corpus is now, one cleared cohort at a time.
After the season: what the academic council should see
The post-season report writes itself if the metadata was captured at intake. Resist the temptation to present raw similarity averages — they flatter nobody and inform nothing. Report instead:
- Band distribution by programme, and how it moved against last season.
- Time from submission to certificate, per band, against the published commitment.
- The categories of source behind amber and red findings — web, published scholarship, prior student work — because that is what tells you where teaching, not policing, is needed.
- Amber-to-red conversion rate per reader, as a calibration check, not a league table.
- One page of proposed band adjustments for next season, dated and minuted.
That last item is the quiet payoff of the whole playbook. An office that can show the council we defined the procedure, followed it, measured it, and here is how we are tightening it has turned exam season from an annual emergency into an operated process — one that treats students fairly, spends expert attention only where the evidence warrants it, and leaves a verifiable record behind.
