How to standardise and moderate IGCSE mock marking in your department
Standardise before marking, then moderate a sample after it. Before anyone marks, the department marks the same few scripts and agrees how the mark scheme applies. After marking, a lead marker second-marks a sample of each marker's scripts, weighted towards grade boundaries. Differences within an agreed tolerance stand; anything outside it triggers a re-mark of that question. If an AI tool is your first marker, treat it exactly like another member of the marking team.
This copies the exam boards' own approach to examiner quality, scaled down for a department.
Why mock grades differ between classes
Even experienced teachers apply the same mark scheme differently. Common causes:
- Generosity on borderline answers: "I know what she meant".
- Inconsistent use of follow through and ecf.
- Different readings of levels descriptors in extended writing.
- Marking towards a grade, which Cambridge's Generic Marking Principles explicitly rule out.
If Class A is marked two marks more generously per question than Class B, the "grade" differences between the classes are partly marking differences. Mock grades feed predicted grades, sets and intervention lists, so this matters.
Step 1: Standardisation (before marking)
- Choose 3–5 standardisation scripts across the ability range, including at least one borderline script.
- Everyone marks them independently, including the AI tool if you use one.
- Meet for 30–45 minutes and compare question by question. Agree:
- Which alternative answers are acceptable (add them to the mark scheme).
- How to treat contradictions, the list rule and crossed-out work.
- Level placement for extended responses, with one anchor script per level.
- Record the decisions on a shared, annotated copy of the mark scheme.
The Principal Examiner Report for Teachers for the same paper is invaluable here. It records how the real examiners handled the common borderline answers.
Step 2: Agree a tolerance
A tolerance is the size of difference you'll accept between two markers without changing a mark. Example tolerances for internal mocks:
| Question type | Example tolerance |
|---|---|
| Point-marked questions (maths, science, short answers) | 0, because these should match exactly |
| Levels-marked extended writing (e.g. 40-mark section) | ±2 marks |
| Whole script total | ±3% of the paper total |
These are examples, not official values. Set your own and write them down.
Step 3: Sampling (after first marking)
- How many? For internal mocks, many departments sample around 10% of each marker's scripts. Add more for new staff, new specifications, or any marker whose first sample falls outside tolerance.
- Which scripts? Weight the sample towards students near grade boundaries, where a few marks change the outcome, and include the full ability range.
- How? The second marker marks without seeing the first marks where possible. Blind marking reveals more.
Step 4: Act on what moderation finds
| Finding | Action |
|---|---|
| Within tolerance | Marks stand. |
| One marker consistently generous or harsh on a specific question | Re-mark that question across all of that marker's scripts. |
| One marker outside tolerance across the board | Re-mark a larger sample, then decide whether to re-mark all of their scripts. |
| Disagreement caused by an unclear marking point | Agree the rule, update the annotated mark scheme, re-mark that question across the cohort. |
Re-marking by question, not by script, is what keeps moderation manageable.
How to moderate AI-marked scripts
If an AI tool did the first marking, the process is the same, and in two ways easier:
- The AI applies the same interpretation to every script. If it's wrong on Question 4(b), it's wrong in the same way for everyone. You find that in the sample and fix it across the cohort in one pass.
- Its reasoning is visible. A good tool shows why it awarded each mark, so a moderator can see why a mark differs, not just that it differs.
One tip: weight your sample towards diagrams and levels-marked answers, as you would for any marker, because those are the hardest marks to apply consistently.
In The Learning Mate, teachers can change any AI mark or comment before publishing. Grades can be exported to Excel for moderation spreadsheets. Each script has an annotated PDF a second marker can check.
A one-page moderation record
Keep a short record for each mock. It answers questions from senior leaders, parents and inspectors:
- Paper, series and syllabus code.
- Standardisation date, attendees and key decisions.
- Tolerance used.
- Sample size per marker, and the results.
- Re-marks carried out.
- Where AI was used.
FAQ
What is the difference between standardisation and moderation?
Standardisation happens before marking: markers agree how to apply the mark scheme using sample scripts. Moderation happens after: a sample of marked scripts is checked to confirm the marking is consistent.
How big should a moderation sample be for mocks?
There's no official figure for internal mocks. Around 10% per marker, weighted towards grade boundaries, is a common starting point. Increase it for new markers or when the first sample is out of tolerance.
What is a marking tolerance?
The largest difference between two markers that is accepted without changing the mark. It's usually zero for point-marked questions and a small range for levels-marked writing.
Can AI-marked scripts be moderated?
Yes. Treat the AI as one marker in the team: sample its marking, compare with a teacher's marks, and re-mark any question where it's consistently out.
Where can I find how examiners treated borderline answers?
In the Principal Examiner Report for Teachers for that syllabus and series, published on Cambridge's website and the School Support Hub.
Sources
- Cambridge, Marking and grading: cambridgeinternational.org
- Cambridge Generic Marking Principles, printed at the start of each IGCSE mark scheme, e.g. 0580 June 2024 Paper 41: cambridgeinternational.org
- JCQ, AI Use in Assessments: jcq.org.uk