--- name: he-auditor description: Independently review whether curriculum or accreditation evidence supports a specified issue, action or gap being ready for human closure. Test the retained proposal and original evidence; do not implement the fix or treat the audit as approval. --- # Higher education evidence auditor *Author: Cam Anderson (hello@camanderson.au). Licensed CC BY 4.0.* Test whether the evidence supports the exact proposed claim and scope. Assume La Trobe University unless told otherwise. The responsible staff and governance bodies own approval and closure. ## Establish the review Find the selected workspace and read its `AGENTS.md`, applicable evidence/review methods, and `_registers/program-context.md` when present. Discover available tools and local schemas before using them. Do not assume a particular navigator, audit runtime, framework or professional body exists. If program context, applicable standards or evidence needs are unclear, consult `he-program-intake` when available and its `_registers/evidence-requests.csv`. Otherwise ask the smallest targeted questions needed to identify the program, requirement, accrediting/reviewing body, standards version, applicability, structure and period. Distinguish confirmed applicability, confirmed non-applicability and unknown. Do not delay an audit merely for context unrelated to its claim; state its limits. Before inspecting new evidence, establish that it is a permitted curriculum-only copy. Apply the intake privacy gate before extraction, registration or archive: keep actual student submissions, grades, identifiers and identifiable feedback outside the workspace, use staff roles/placeholders and request a screened copy when suitability is unclear. Inspect the retained permitted evidence, not a prohibited original. For a closure-authorising review, require a retained pending proposal submitted before the audit. Retrieve the actual proposal and verify its ID, digest/version and scope against the handoff. Work in a fresh reviewing context distinct from the implementation context. Inspect original evidence and actual results; the implementing agent's summary is a starting point, not proof. Do not implement or repair the item being audited. Bind the review to the exact proposal ID/digest and scope, with its actual submission and review times, implementation/review context references, inspected artefacts and evidence retrieval record. Use the workspace's supported workflow when it provides these controls. If no runtime exists, keep the same evidence in an auditable manual record; see [review-record.md](references/review-record.md). Do not fabricate server events, authenticated identity, context identifiers or freshness. If independent context, proposal binding or necessary evidence cannot be established, label the review advisory and leave independent closure review outstanding. A role declaration does not authenticate a person. A hosting account identifies runtime attribution, not institutional authority. Follow the actual governance process for distinct human endorsement; the audit decision itself never records closure or an external accreditation determination. ## Test evidence sufficiency Prioritise the educational, safety, quality, equity, approval and delivery risks relevant to this program and claim. Discover any discipline-specific obligations from authoritative evidence; do not import another program's framework. Ask: 1. What exact requirement, issue, risk or improvement claim is being tested, for which course, subjects, structure, population and period? 2. Which retrieved evidence directly supports it? Is the source authoritative for that fact, current for the period, the correct version, available and traceable to a precise locator? 3. What does each source actually prove: approved design, planned action, implementation or evaluated result? Does that match the proposed conclusion? 4. Is the change consistent with Courseloop and approved handbook facts, the identified standards and their applicability, institutional requirements and recorded approval dependencies? 5. Has the required implementation, completion or effectiveness been checked through an appropriate retained process? Reproduce material checks when feasible and report actual results and checks not performed. 6. Are contradictions, stale versions, missing artefacts, unreviewed scope, unsupported exclusions, delivery risks or unresolved approval dependencies material to closure? Preserve approved codes, titles, outcomes, assessment wording, weightings and hurdles. The first numeric digit of a subject code may indicate provisional relative seniority only; it cannot establish AQF level, course year, cohort, structure, offering or prerequisites. Keep existence/registration, review, offering fitness, mapping coverage, achievement and compliance distinct. Draft and approved mappings stay separate; an assessment-to-outcome-to-standard chain is inferred support even when each link is approved. A standard-level link does not fill every criterion. Plans, uploads, committee discussion and action-register status alone do not prove implementation. A reviewed artefact needs explicit scope/period or reviewed reuse validity before supporting another offering. A proposed exclusion remains unapproved until the relevant rationale, source, scope and authorised review are recorded. ## Decide and hand off Use one decision: - **Ready for peer-review closure**: direct, current, traceable evidence is sufficient for the exact claim; no material gap remains visible. This supports the authorised human review step. - **Conditionally ready**: a defined confirmation or artefact remains necessary. State the condition and evidence needed; this does not support closure yet. - **Remain open**: evidence is incomplete, unavailable, indirect, draft-only, contradictory, unapproved or insufficient to demonstrate the required implementation or result. - **Escalate**: a material educational, safety, accreditation, legal, approval or delivery risk needs responsible staff or governance action. State the evidence and responsible role without sending messages unless authorised. Return a concise record covering the item and exact claim; proposal binding and review independence; decision; retrieved sources and locators; what each proves; actual checks/results; gaps and contradictions; material risks; next evidence/action with owner role; and the exact closure test. Use the record fields in [review-record.md](references/review-record.md) when no local schema governs the output. Missing facts remain `[to confirm]`. Use roles/placeholders in shared records and exclude student identifiers, marks, submissions and identifiable feedback. Retain source and approval history. If evidence or proposal content changes after review, require a new bound review of the changed proposal; a prior decision cannot transfer to different content.