Assessment assurance for higher education
Prove what was learned.
AI changes how work is produced. CoAcademic helps institutions prove what was learned.
AI proposes. Evidence explains. Educators decide.
“The committee transcript from March 1962 records three separate objections to the proposal…”4 / 5
02The problem
A finished artifact no longer proves learning.
When generative AI can produce a competent submission, the final product alone stops carrying the outcome claim. Detection does not resolve this: it is unreliable, adversarial, and it tells a course team nothing about what a student understood.
CoAcademic takes the other route — product, process, and explanation together, aligned to the outcome the assignment was written to evidence.
No detection. No authorship scoring. No automatic grading.
03The Learning Evidence Chain
Nine links, one connected system.
An outcome claim is only as strong as the chain behind it — from the outcome the assignment is meant to evidence through to the record a reviewer can inspect.
- 01
Intended learning outcome
The outcome this assignment is meant to evidence, stated before anything is designed.
- 02
Assignment requirement
What the task asks for, and which part of it carries the outcome.
- 03
Permitted AI use
Restricted, limited, disclosed, or required — declared in advance and aligned to the outcome.
- 04
Required process evidence
The planning, drafting, or annotation artifacts the task asks students to keep and submit.
- 05
Student artifact
The submitted work itself — one link in the chain, never the whole of it.
- 06
Explanation or defense
The student's account of specific choices, in writing or a short targeted defense.
- 07
Rubric judgment
A criterion reading measured against calibrated band language, with the excerpt attached.
- 08
Faculty decision
Approve, revise, or reject. The educator decides, and nothing releases itself.
- 09
Assurance record
The chain preserved end to end, so an outcome claim can be inspected rather than trusted.
Each link contributes evidence. No single link — not the artifact, not a process log, not a local prompt — proves authorship on its own.
04Lifecycle
Design → Calibrate → Review → Decide → Assure
Five stages carry the chain. See the full journey, stage by stage.
01Design
Design the evidence, not just the task
Confirm the outcome, audit where it is exposed, then set the AI-use rule and the process evidence the assignment will require.
Chain links 01–04- Resilience audit
- chain 01–02
- AI Use Protocol
- chain 03–04
02Calibrate
Agree what a band means
Prior graded examples and anchor work fix band interpretation with the faculty who will mark.
Rubric Calibration- Rubric v3
- confirmed
- Band meaning
- agreed
03Review
Read the evidence behind the reading
Each criterion arrives with the excerpt, the calibrated band language, a written rationale, and stated uncertainty.
Evidence-Linked Review- Excerpt p. 4 · ¶2
- linked
- Proposed band
- 4 / 5
04Decide
The educator decides
Approve, revise, or reject per criterion. Nothing reaches a student without an explicit release by a named educator.
Faculty decision- Approved
- 3 criteria
- Released to students
- not yet
05Assure
Preserve the assurance record
Outcome, rule, evidence, artifact, explanation, rubric version, and every faculty decision, kept together for later inspection.
Assurance Record- Chain links
- 01–09
- Faculty changes
- recorded
Illustrative Prototype
The Review stage, in the open
One criterion at a time, with the evidence in view. This sample interface runs on demo data for HIST-204, and every score is labelled proposed.
Use of evidence
Criterion C2 · weight 30% · confidence high4 / 5
“The committee transcript from March 1962 records three separate objections to the proposal, which complicates the standard reading that the reform passed without internal resistance.”
Band 4 language requires primary sources used to complicate an existing reading. The transcript is cited and interpreted, but two later claims rest on summary rather than sourced material.
Awaiting your decision. Nothing is released automatically.
06Faculty control
Faculty judgment remains final.
CoAcademic is a review assistant. It does not issue final grades, and it does not replace the educator who knows the cohort, the context, and the intent behind the assignment.
- ProposedAI drafts a band and feedback
- ExaminedEducator reads the linked evidence
- DecidedApprove, revise, or reject
- ReleasedOnly by a named educator
Proposed, never final
Every score is labelled proposed until you approve it.
Draft feedback
Feedback is a draft you can rewrite, shorten, or discard.
Disagreement is one action
Rejecting a proposal takes one action, not a workaround.
Deliberate release
Release to students is an educator-initiated step, every time.
07Trust & governance
Built for departments, not just individuals
How the product is designed to behave. We hold no certifications and claim none — the detail sits on Security & Data.
Privacy-minded processing
Handle submissions for the assessment task at hand, with scoped access.
- Faculty control
- Departmental retention settings
- Guardrail
- No secondary use
Traceable evidence
Point every proposed score back to the passage that informed it.
- Faculty control
- Excerpt link per criterion
- Guardrail
- No unsourced score
Rubric fidelity
Constrain proposals to your confirmed criteria, weights, and bands.
- Faculty control
- Locked rubric version
- Guardrail
- No invented standards
Controlled release
Keep results in draft until a named educator releases them.
- Faculty control
- Explicit release action
- Guardrail
- No automatic publish
The Founding Assessment Assurance Pilot
One course team, one assignment, one confirmed rubric — audited, calibrated, reviewed against evidence, and decided by faculty. Founding Pilot Scope: applications are opening soon, and the pilot page sets out the intended scope and what is out of scope.