Assessment Moderation Synthesis

Workbook-grounded reconstruction. Assessment calibration and moderation bring together authorised samples, marks, criterion judgements, annotations, rationales and assessor discussion. The process may occur before, during or after marking. Evidence is distributed and disagreement may reveal an important criterion or evidence issue rather than an error to eliminate.

Proposed. Preparing moderation records is time-consuming, and criterion-level patterns can be obscured across many sensitive records. Generated synthesis may flatten disagreement, detach reasoning from evidence, misattribute comments or create false consensus that influences consequential grading decisions.

Proposed. Construct a source-linked synthesis of authorised moderation evidence at criterion and case level. Preserve agreements, disagreements, minority reasoning, rationales and unresolved decisions; distinguish extracted observations, generated suggestions and formal assessor decisions. Return the provisional synthesis to assessors for correction and deliberation. Preserve the approved record under explicit access and retention rules, and never allow the system to average, resolve or alter marks autonomously.

• A source-linked synthesis of moderation agreements, disagreements and rationales.
• Visible criterion, task or assessor patterns requiring discussion.
• A clear list of unresolved cases and decisions for authorised assessors.
• A human-approved moderation record without autonomous mark changes.
• An authorised retention and access record for sensitive working and final moderation artefacts.

Assessors make and revise academic judgements. Moderation leads define purpose, manage deliberation and approve the formal record. Governance, privacy or integrity staff advise on authorised processing and exceptional pathways. GenAI organises, compares and summarises provisional records; it cannot determine a mark, resolve disagreement, assume consensus or authorise a change.

Reconstructed risks

• The synthesis invents consensus or gives majority judgement unwarranted authority.
• Comments are detached from evidence or attributed to the wrong case.
• Sensitive student or assessor information is disclosed or retained inappropriately.
• Generated recommendations influence marks before assessors verify them.
• Outliers are erased even when they reveal criterion ambiguity or bias.
• A provisional synthesis is mistaken for the formal moderation record or transferred directly into grade changes.
• De-identification removes context needed to interpret evidence while still leaving records re-identifiable.

• Minimum viable implementation: provide an approved assistant with a small authorised moderation table and criteria, request a criterion-level synthesis with source references, and require every assessor to correct and approve the record.
• Robust implementation: add role-based access, structured extraction, reconciliation checks, source links, dissent labels, version history, formal decision fields and enforceable retention and disposal rules.
• Low-code implementation: connect an institutional assessment store, controlled synthesis workspace and moderation register without granting the assistant write access to official marks.
• Local or privacy-preserving implementation: minimise identifiers, process within approved institutional storage and keep source evidence accessible only to authorised moderation participants.
• Speculative implementation: monitor recurring criterion ambiguities across approved de-identified moderation records, while prohibiting autonomous changes and requiring governance review before any cross-cohort reuse.