Key takeaways
- Compare marking decisions at question level; matching totals can hide different errors.
- Include varied copies, different markers and borderline results in the moderation sample.
- Record why a score changed and revisit every answer affected by a scheme clarification.
- Treat AI marks as draft decisions that still need evidence checks and teacher approval.
Two teachers award 42/60 to the same answer sheet. Their totals agree, but one gave an extra mark on Question 4 and one withheld a mark on Question 8. A totals-only check would miss both disagreements. Answer sheet moderation looks at the decisions behind the score.
The workflow below is a practical template for internal school, college and coaching assessments. Adapt it to the paper and your institution’s requirements. It is not a substitute for an examination body’s moderation procedure, and the sample choices are suggestions rather than a formal statistical assurance method.
What should answer sheet moderation check?
Keep three checks separate: whether the answer has been read correctly, whether the scheme has been applied correctly, and whether the marks have been totalled and recorded correctly. A good review follows an answer through all three. It also confirms that no attempted question or continuation page was overlooked.
A shared rubric gives reviewers a common reference. Carnegie Mellon University’s rubric guidance identifies consistency across graders as one benefit of explicit criteria. A shared document still needs discussion: words such as “adequate explanation” can leave teachers applying different standards to the same response.
Prepare a review pack before selecting copies
Collect the question paper, the scheme version used for marking, model answers, accepted alternatives and any decisions made during checking. Keep each original answer sheet available alongside the question-level marks. A spreadsheet of totals alone does not provide enough evidence to review subjective answers.
- Identify the assessment: subject, class, paper version, date and maximum marks.
- Identify responsibility: who marked each batch and who will settle unresolved questions.
- Identify rules: partial credit, optional questions, word limits and any applicable penalties.
- Identify changes: which rubric clarifications were made and whether earlier copies were revisited.
For future batches, compare a few varied answers before dividing the stack between markers. The marking scheme guide provides examples of criteria you can use as a starting point for that discussion.
Choose a sample that can reveal different problems
Selecting only neat, middle-scoring copies gives a narrow view of marking. For an internal assessment, combine some ordinary copies selected without reference to their score with targeted copies that deserve attention. Include work from each marker and from different parts of the marking sequence.
| Include in the review | What it helps you inspect |
|---|---|
| High, middle and low scores | Whether the scheme distinguishes different levels of evidence consistently. |
| Results near a relevant pass or grade boundary | Whether a small marking error could change a student's outcome. |
| Early and late copies from each marker | Whether the standard changed during the batch. |
| Unusual correct approaches | Whether the scheme was treated too narrowly. |
| Unclear scans and continued answers | Whether a reading or page-location problem affected marks. |
| A selection of otherwise unflagged copies | Whether problems exist outside the cases already suspected. |
Compare question-level decisions and record the reason
Where practical, ask the reviewer to score a selected answer before seeing the first score. Compare the evidence credited under each criterion. A difference might come from a missed sentence, an ambiguous rule, an alternative method or a simple transfer error. Name the cause before deciding the remedy.
Do not settle disagreements by automatically averaging the two marks. Return to the criterion and the response. If the scheme does not resolve the case, the designated subject lead should record a clarification and specify which other answers need review.
| Review log field | Illustrative entry |
|---|---|
| Copy and question | Student reference S014, Question 6 |
| Original and reviewed mark | 2/4 → 3/4 |
| Reason | A valid alternative explanation met criterion 3 but was not credited. |
| Scope of follow-up | Review Question 6 in both classes for the same explanation. |
| Owner and completion | Subject lead records the clarification; each marker confirms review is complete. |
Keep unchanged decisions too when they resolve a recurring doubt. An accepted example of a 2-mark response and a contrasting 3-mark response can make the next moderation discussion more concrete than another general instruction to “be consistent”.
Apply the same discipline to AI draft marks
Start by comparing the extracted answer with the original scan. A misread negative sign is a reading error; withholding credit for a valid explanation is a grading decision. They need different corrections. Our maths step-marking guide shows how those issues can interact.
Review unusual zeros, missing answers, unclear handwriting and disagreements between the written feedback and awarded marks. Also include unflagged answers: the absence of a warning does not prove the score is correct. Record the reviewed decision against the student’s actual response, not merely the AI explanation.
A checklist before releasing exam results
- Every copy is linked to the intended student and the correct paper.
- All attempted answers, continuation pages and question-choice rules have been accounted for.
- Each question score is within its maximum, and the total matches the question scores.
- Disagreements have a recorded resolution tied to the scheme.
- Any recurring issue has been checked across all affected answers.
- Feedback matches the final marks and describes the work accurately.
- The responsible teacher has approved the reviewed results for release.
After release, retain the scheme version and review decisions with the assessment record. If a student asks why an answer received a particular mark, the explanation should point to the criterion and the evidence, without requiring the marker to reconstruct the decision from memory.
Where Evalezy fits
Evalezy returns AI marks as drafts. Teachers can see the extracted answer, feedback and criteria breakdown in teacher review, edit marks and choose when to release results. Students see results pending until release. Keep your moderation log alongside this workflow so that changes to the scheme are followed through across the batch.
Sources
- Carnegie Mellon University, Eberly Center: Rubrics (checked 10 October 2026)
Frequently asked questions
What is answer sheet moderation?+
It is a review of whether marks have been awarded consistently against the agreed scheme. It can include comparing markers, rechecking selected answers and resolving disagreements before results are released.
How many answer sheets should be moderated?+
Use your institution or examination body's requirements where they apply. For an internal test, choose a sample that covers markers, score ranges and known risks, then expand it if disagreements reveal a recurring problem. There is no universal sample size recommended by this guide.
Is moderation the same as checking totals?+
No. Total checking catches arithmetic and transfer errors. Moderation also asks whether the evidence in an answer earned the right marks under the scheme.
What should happen if the marking scheme changes during review?+
Record the clarification and identify all answers affected by it, including those already checked. Review that group consistently before releasing results.