Academic integrity software that builds both sides of every case from how the work was written, never from a detector score
Referrals screened on how the work was written, with both sides in one packet.
See one case, screen by screen ↓Academic Integrity Case Management is academic integrity software for integrity offices: it turns each referral, whether an academic integrity violation or an honor code violation, into a two-sided case file built from drafts, version history and rubric evidence, so panels decide on evidence, not detector scores. Seven agents assemble the allegation and the student's side at once, replay how the work was written, and check any sanction against like cases; the officer, the Director and the panel make every decision.
AI in education: a buyer's guide to agentic AI for colleges, universities and school districts →
Updated
Academic integrity software should help an integrity office decide what happened to a piece of work, yet most referrals still arrive with whatever the instructor had: a submission, a hunch, often a screenshot of an AI-writing score. Since AI cheating in college became the headline, detector scores have stopped standing as evidence of academic misconduct, and an academic integrity policy that leans on them invites complaints, because detectors flag non-native writers far more often than anyone else.
So the officer rebuilds the case by hand: the submission log from the learning management system, the instructions, the rubric, the syllabus clause, the emails. The student's side, the earlier drafts, writing-center visits and office hours that usually clear an innocent student, arrives weeks later if the student knows to ask. Sanctions for the same academic integrity violation drift between officers, and that is exactly what appeals and accreditors test. Academic Integrity Case Management puts seven agents in front of that work, so the officer starts from both sides of the case, cited, and decides.
Estimated split of the officer hours one referral takes before a screening decision, by hand and with the agents. The 2.5-hour total is the demo office's own estimate; the split and the 'after' column are our arithmetic, kept modest.
Seven agents build both sides of every case and replay how the work was written; the officer decides and the Director confirms.
Case AIC-26-0418: Seo-yeon Han, a first-year student at Halden State University, PHIL 1100 Introduction to Ethics, Essay 1, "Moral luck and blame". Here is the academic integrity process on her case, screen by screen, in the working solution.
Celeste's desk shows 55 open cases and 7 ready to screen. The agents read each referral overnight and wrote it as one sentence, with the count of items for the allegation and on the student's side, and the screening due date. One late case is already escalated to the Director.
AIC-26-0418: one paste of 1,412 words at 11:58 p.m. The instructions ask students to paste their final text, and her shared draft shows 14 sessions.
Dr. Anselm Kovacs referred the essay with a free online checker's screenshot and the line "far more polished than I'd expect from an ESL student". The Fairness agent left the screenshot out under Policy 4.20 §6, withheld its score from every reader, and flagged the phrase. At 9:31 a.m. Dr. Kovacs replaced it with what he had actually seen: the single paste.
It compares Seo-yeon with a group of writers, not with her own work, and could penalize a non-native writer (Title VI).
The notice of referral and consent request went out on Oct 7, approved by Celeste and confirmed by the Director. Seo-yeon shared one file, "PHIL 1100 Essay 1 draft", read-only, for this case only. The allegation file (5 items, 9 pages) and her side (4 items and her statement, 19 pages) reach the officer together, every item cited to its system and passage.
The writing center appointment system timed out at 10:59 a.m. Retry.
The scrubber replays her draft: 14 sessions from Mon Sep 21 to Sat Oct 3, 6 h 52 min in all, each paragraph tied to the sessions that wrote it. Two small paste notches are her own 112-word outline and a 61-word quotation cited to the course reader. The 11:52 p.m. version matches the LMS submission word for word.
Celeste asks what the writing center recorded. The answer cites the source: a 45-minute visit on Sat Sep 26 at 11:00 a.m., "Thesis on Nagel's moral luck", and the first paragraph rewritten in the next session at 12:05 p.m. The scrubber moves to session 5.
The three elements of unauthorized assistance under Policy 4.20 §3(b) each carry the evidence for and against. The assignment instructions say "Draft in any editor and paste your final text into the submission box." Of 23 like referrals since Aug 2025, 21 were dismissed at screening and 2 went to a panel, both found not responsible.
Celeste dismisses the referral before panel with reason code DS-2, "The evidence shows the student's own writing process", and edits the reason the agents drafted from the evidence. The Notice agent drafts one letter to Seo-yeon and Dr. Kovacs, and it waits for the Director.
The officer cannot confirm her own notice; Dr. Ellsworth does, at 11:26 a.m. The student and the referrer receive it at once. Seo-yeon's portal says the case is closed, nothing appears on her transcript, and access to her shared draft has ended.
Packet v3: 11 items, 31 pages, identical for the student, the referrer and the officer.
Every action on the case, by people and agents alike, is on the trail with who, what, when, why and the evidence, including every view. The appeal, records-request, accreditor and reviewer packs export from the same screen.
Aug 24 to Oct 21: 318 referrals, a median of 3.4 business days from the student's response to screening, 0 detector scores used as evidence, and 147 of 153 proposed sanctions inside the like-case range.
The same case, seen by the four people who handle it: what their week looked like, and what it looks like now.
Seven agents build both sides of every case and replay how the work was written; the officer decides and the Director confirms.
Guides the faculty referral as a conversation, attaches the assignment, rubric, submission and syllabus clause from the learning management system, and asks what was observed; it can check a reference list against the library catalog.
Assembles the neutral allegation file and the packet, citing every item to its page and linking it to the policy's elements.
Builds the student's side at the same time: earlier drafts, the revision history the student consents to share, writing-center visits, office hours and earlier submissions.
Rebuilds how the text grew from the revision history: sessions, durations, paste events and paragraphs by session, then compares the final version with the submission word for word.
Refuses AI-detector scores as evidence and withholds them from every reader, flags referral language that compares a student with a group of writers, and checks that every party received the same packet.
Compares a proposed outcome with past cases of the same type and level, de-identified, and asks for a written reason when it falls outside the range.
Drafts notices and decision letters in plain language, schedules panels, and sends a notation to the registrar only when a sanction requires one.
Decides; the Director confirms. The agents propose; a named person decides.
The officer can ask about any item, date or passage in plain words; every answer cites its source and can move the screen to it.
Why isn't the 11:58 p.m. paste evidence of copying?
Because the assignment asks for it: "Draft in any editor and paste your final text into the submission box." The pasted text matches the 11:52 p.m. version of her draft word for word, and that draft grew over 14 sessions from Mon Sep 21. Of 23 like referrals since Aug 2025, 21 were dismissed at screening.
What did the writing center record?
A 45-minute visit on Sat Sep 26 at 11:00 a.m.: "Thesis on Nagel's moral luck; we worked on making the claim arguable." Her draft's first paragraph was rewritten in the next session, 12:05 p.m. the same day.
Why was the detector screenshot left out of AIC-26-0418?
Policy 4.20 §6 says AI-detector scores are not evidence and never go in a case file. Detectors label human writing as AI most often for non-native writers: 61.3% of TOEFL essays in one study. The Fairness agent excluded it on receipt and withheld the score from every reader.
Which cases need the Director today?
Three: AIC-26-0377, a Level 2 sanction inside the range of 11 like cases; AIC-26-0421, a Level 2 agreement; and AIC-26-0446, a notice of referral. AIC-26-0443 is also escalated to her: its screening is 1 day late.
13 screens from the working solution, on its sample data. Pick one to see it large.
The agents read your academic integrity policy's elements and sanction ladder. Every setting is saved as a new version, and the Director approves policy changes.
| Setting | Default | Choose from |
|---|---|---|
| AI-detector scores | Never evidence; withheld from every reader | Locked by Policy 4.20 §6 |
| Revision history | Only with the student's consent: one file, one case, read-only | Locked; a student who declines to share: nothing is inferred |
| Screening | 5 business days after the student's response | 3, 5 or 7 business days; escalate a missed clock to the Director the next business day |
| Student response window | 10 business days from the notice of referral | 7, 10 or 15 business days |
| Panel hearing | 20 business days from the referral to a panel | 15, 20 or 30 business days |
| Sanction ladder | Level 1 grade penalty (officer) to Level 4 suspension (Provost) | Level 2 and above confirmed by the Director; notation only at Levels 3 and 4 |
| Retention | Dismissed: 3 years, then de-identified | Resolved or found responsible: 7 years; redaction of other students in packets |
| Agent thresholds | Fairness agent flags a phrase at 0.70, low on purpose | Per agent, versioned; override reason codes OV-1 to OV-4 |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
Sources: Liang, Yuksekgonul, Mao, Wu and Zou, "GPT detectors are biased against non-native English writers", Patterns 4(7), 2023 · a 2025 national newspaper records investigation of AI cheating at 131 UK universities, as reported in June 2025 · FERPA, 34 CFR Part 99 · Title VI, 34 CFR Part 100 · "target" = design goal, measured in the live solution · "demo" = seen in the working solution, on its sample data · "typical" = published figures · "estimated" = our arithmetic, kept modest. Halden State University, its Policy 4.20 and the people named on this page are fictional: characters and sample data in the working solution.
Academic Integrity Case Management is academic integrity software for integrity offices: it turns each referral, whether an academic integrity violation or an honor code violation, into a two-sided case file built from drafts, version history and rubric evidence, so panels decide on evidence, not detector scores. Seven agents assemble the allegation and the student's side at once, replay how the work was written, and check any sanction against like cases; the officer, the Director and the panel make every decision.
The Referral agent attaches the course items and asks the instructor what was observed. The Evidence and Defense-view agents build both files at once, the student receives a notice and 10 business days to respond, and the officer screens within 5 business days of that response: dismiss, resolve by agreement, or send to a hearing panel. Every notice is confirmed by the Director before it is sent.
Not here. The Fairness agent excludes detector scores from the file and withholds them from every reader, and the setting is locked. Detector false positives fall hardest on non-native writers: in a 2023 study, seven detectors labeled 61.3% of human-written TOEFL essays as AI-generated. Instead the instructor is asked what they saw in the work itself.
No. The student chooses whether to share one file, read-only, for one case, and can revoke it. If the student declines, nothing is inferred, and the officer is told so. When a draft is shared, the Writing-process agent replays how it grew and checks the final version against the submission word for word.
The Consistency agent compares every proposed outcome with de-identified past cases of the same type and level, and a sanction outside the range needs a written reason the Director sees. The sanction ladder runs from a grade penalty on the work to suspension, and anything above a grade penalty is confirmed by the Director.
The same packet the student had, with the same fingerprint, laid out as the elements of the allegation with evidence for and against each. Panelists vote element by element, and the Notice agent drafts the letter only from the recorded votes. The officer who screened the case cannot sit on its panel.
Yes. The agents assemble, replay, compare and draft; they never decide responsibility, propose or apply a sanction on their own, or contact a student without an approved notice. The officer decides with a reason code, the panel decides responsibility, and the Director confirms every notice.
The Agentic Solution Engine builds and deploys it from your requirements and documents, such as your integrity policy, sanction ladder and a sample of past referrals, and it goes live once every quality gate has passed. We will walk you through it on your own cases first.
Every course file fixed before April 2027, and heard right by a screen-reader test.
1-min demoLearning Outcomes AssessmentAssessment reports built from scored student work, where every number opens its papers.
1-min demoCourse Evaluation Comment AnalysisFixable teaching issues by course in minutes, and no biased comment in a faculty file.
1-min demoWe'll run Academic Integrity Case Management on your policy and a sample of your past referrals.