- 01Delete media on submit
- Opt-in per assessment. When it is on, candidate audio and video files are deleted from storage at submission, for every answer that already has a transcript. The transcript is kept for grading; the recording is not, and the deletion is recorded per answer and cannot be undone. An answer whose transcription did not succeed keeps its recording so the response can still be assessed.
- 02Reviewer score override
- An assessor can override the score on any individual question, with a comment explaining why, and totals recompute automatically. The AI-generated score is kept alongside the override rather than replaced, so both numbers stay on the record as an audit trail.
- 03AI transparency notice
- Tell candidates clearly that AI is used to analyse their responses, and publish a custom notice so anyone not comfortable can request an alternative method from their instructor. New results stay unreleased until an assessor releases them, so a candidate sees nothing until a person acts.
- 04Audio or video candidate choice
- Owners can let candidates choose audio-only instead of video for any video response question. Lower biometric footprint, better accessibility, and identical transcription and grading either way.
- 05One-click audit pack
- Export an assessment’s whole evidence chain as a single PDF and JSON pair - the model and configuration versions in force, the override ledger, and the release chain showing who signed off and when. It is the document you hand a moderator or an auditor when they ask how a mark was arrived at, and it needs no login to read.
- 06Confidence on every mark
- Each question comes back with the model’s own certainty about that judgement, alongside a per-criterion explanation of which rubric level was awarded and why. Reviewers sort by confidence and work the borderline marks first, rather than reading every result at the same depth.
- 07Your own agreement rate
- Every organisation gets an Accuracy page measuring how often its assessors agree with the AI - overall, per question, and per assessor. It surfaces which questions attract the most overrides and how each assessor moves scores, so you can fix a rubric that is producing disagreement rather than argue about it.
- 08Segregation of duties
- The person who builds an assessment does not have to be the person who releases its results. Roles are separable across your team, and each release is recorded against the individual who made it.
- 09Feedback now, score after sign-off
- An assessment can share its written feedback with a respondent immediately while holding the numeric score until an assessor confirms it. The score is withheld on the server - it does not reach the browser, and it is kept out of the AI’s written prose as well, so it cannot leak back through the wording.