Responsible AI reasoning for universities
Make AI reasoning defensible in every grade.
Exam+ gives universities a controlled way to introduce AI reasoning into examination grading, with evidence-linked recommendations, human academic supervision, visible moderation, and query-ready decision records.
AI-supported reasoning remains under named academic authority.
Traceable decision path
- 01Script and rubricApproved material enters
- 02Evidence-linked reasoningPassages mapped to criteria
- 03Examiner reviewNamed academic decidesHuman
- 04ModerationSecond-examiner checkHuman
- 05Approved decision recordLocked and reviewable
AI cannot approve or release a grade.
From principle to practice
Give responsible AI an academic operating model.
Universities are developing principles for responsible AI. Exam+ turns those principles into a controlled workflow inside one of the institution’s most consequential activities: evaluating student learning.
AI reasoning is used to examine evidence against an approved rubric. Named academics retain decision authority, moderation remains visible, and every approved grade carries a reconstructable history.
Evidence before recommendation
Each recommendation connects the relevant script passages to the applicable rubric criteria.
Named academic authority
An authorised examiner reviews, adjusts, and approves the academic decision.
Visible disagreement
Grade-affecting differences between AI recommendations, examiner judgement, and moderation remain recorded.
Reconstructable decisions
The university can retrieve the evidence, reasoning, approvals, and decision history behind a mark.
A defensible grading workflow
Follow every decision from script to approved record.
- 01
Controlled intake
Scripts, questions, mark allocations, and approved rubrics enter the grading workspace.
- 02
Evidence identification
Relevant passages are identified and connected to the criteria they may satisfy.
- 03
AI-supported reasoning
Exam+ develops a criterion-level recommendation with supporting evidence and an explanation.
- 04
Examiner judgement
The examiner reviews the evidence, edits marks or feedback where required, and records the academic decision.
- 05
Moderation
Second-examiner checks, adjustments, and grade-affecting disagreements remain visible.
- 06
Named approval
A final authorised examiner approves the grade. AI cannot perform this action.
- 07
Locked decision record
The approved evidence, reasoning, changes, identities, and timestamps remain available for later review.
Student response
“Osmoregulation in the nephron depends on the counter-current multiplier. Sodium is actively transported out of the ascending limb, which raises medullary osmolarity and water then leaves the descending limb passively. ADH increases collecting-duct permeability, though I have not explained the receptor mechanism.”
Highlighted passages are linked to the rubric criteria opposite.
Rubric criteria
- C1 Mechanism described4 / 4Evidence linked
- C2 Reasoning explained3 / 5Partial evidence
- C3 Terminology2 / 3Evidence linked
AI-supported recommendation
9 / 12 — criterion-level reasoning with the supporting passages for C1 and C3.
Examiner decision
Adjusted to 10 / 12. C2 credited for the osmolarity link.
Dr A. Lindqvist · Senior examiner · 14:22 UTC
Decision history
- 13:58 — Recommendation generated with linked evidence
- 14:22 — Examiner adjustment recorded (+1 mark, C2)
- 14:40 — Moderation review completed, disagreement noted and resolved
- 14:44 — Grade approved and record locked
Responsible adoption
Introduce AI reasoning through controlled academic authority.
Exam+ gives the university a bounded environment in which AI reasoning can be examined, challenged, measured, and governed. Academic policy determines how the system is used. Named people remain accountable for every grade.
- Approved rubrics and marking instructions govern the assessment.
- Recommendations show their supporting evidence and reasoning.
- Examiners can adjust every grade-affecting recommendation.
- Moderation and disagreement remain visible in the decision history.
- AI cannot approve or release a grade.
- Approved records remain locked and available for authorised review.
Responsible adoption becomes credible when the institution can explain how an AI-supported decision was reached, reviewed, changed, and approved.
Decisions that can be explained
Resolve grading queries from the full record.
When a mark is questioned, Exam+ brings together the relevant script evidence, rubric criteria, AI-supported reasoning, examiner judgement, moderation history, and final approval. Academic teams can respond from the decision record instead of reconstructing the process across files, comments, and inboxes.
- Question under review
- Student query on Q3 marking of script 042
- Original recommendation
- 9 / 12 with linked evidence for C1 and C3
- Examiner adjustment
- 10 / 12 — C2 credited for the osmolarity link
- Moderation note
- Second examiner reviewed the adjustment and agreed
- Final approved outcome
- 10 / 12, released to the student record
- Named approver
- Prof. M. Okafor · 14:44 UTC · record locked
A controlled first step
Build the evidence in one live course.
An Exam+ pilot allows the institution to evaluate AI-supported grading under real academic supervision before considering broader use. The pilot should use an approved course, defined assessment material, named academic owners, and agreed measures of success.
- time from script intake to approved result
- examiner agreement and adjustment patterns
- consistency across scripts and markers
- quality and traceability of supporting evidence
- time and effort required to resolve grading queries
- examiner confidence and usability
- data-handling and governance requirements for wider adoption
The outcome is an institutional evidence base for deciding where AI reasoning belongs, how it should be governed, and what must remain under human academic authority.
Explore a responsible Exam+ pilot.
Tell us where your institution is considering AI-supported grading. We will use the conversation to define a bounded pilot, its academic controls, and the evidence required for an informed decision.
AI-supported reasoning remains under named academic authority.
Prefer email? Contact kris@krisgranger.com