| Teacher observations in natural language | Designed for narrative classroom evidence without forcing diagnosis labels. | Possible, but usually unstructured and hard to review consistently. | Possible, but often detached from the student record and school workflow. | Usually reduced to short prompts or checklists rather than rich context. |
|---|
| Human review before final reports | Built in. Reviewers can edit behaviours before finalisation and preserve the comparison history. | Manual and inconsistent, often spread across tabs, comments, and emails. | Not workflow-native. Edited outputs are easy to lose or overwrite. | Often skipped because the model output is treated as the answer. |
|---|
| Audit trail for escalation and exports | Owned tasks, notes, download history, retention events, and export status are persisted. | Requires manual version control and separate governance logs. | Usually no durable audit trail for prompts, edits, or follow-up actions. | May log scores, but rarely the surrounding review process and follow-up decisions. |
|---|
| Non-diagnostic posture | Public copy, trust guidance, and report wording all reinforce non-diagnostic SEND support. | Depends on local practice and can drift quickly between teams. | Depends on prompting and is easy to overstate without strong controls. | Often pushes people toward labels faster than the evidence supports. |
|---|
| Operational governance for school workflows | Includes pathway routing, retention controls, calibration governance, and protected APIs. | Needs separate tools and disciplined process to approximate this. | Typically lacks owner routing, retention visibility, and release governance. | Focuses on prediction rather than accountable operational follow-through. |
|---|