"If a machine can pass your assessment, the failure lies in the curriculum design, not the student's character."
Try the Audit Tool | Contact for Consultancy
Screenshot: The Integrity Debt Audit upload interface
The Integrity Debt Audit is an AI-powered tool for Higher Education professionals to audit assessment briefs against the 10 Categories of Integrity Debt (Illingworth, 2026). It identifies vulnerabilities to AI automation and provides analytical critiques to support curriculum redesign.
Upload an assessment brief as a PDF or Word document. This is the document you give to students that explains what they need to do for their assignment.
Your document should include:
- The assignment task or question
- Submission requirements
- Assessment criteria or marking rubric
- Any process requirements (drafts, reflections, presentations)
- Word count and deadline information
The tool works best when your brief contains enough detail about how the assessment works. A one-line assignment question will produce less useful results than a full brief with marking criteria and submission guidance.
File Upload:
- ✅ PDF (text-based, not scanned images)
- ✅ Word (.docx only)
Text Paste:
- ✅ Plain text copied from anywhere
- ✅ URLs to public web pages
What Won't Work:
- ❌ Scanned PDFs (no OCR - the tool cannot read image-based documents)
- ❌ Login-protected pages (Moodle, Canvas, Blackboard, SharePoint)
- ❌ Direct links to PDF files via URL input (use file upload instead)
- ❌ Older Word formats (.doc) or LibreOffice (.odt) - save as .docx first
- ❌ Google Docs links - export to .docx or PDF first
Two example assessment briefs are provided:
-
Vulnerable Essay Brief (scores 40-50: Critical Integrity Failure) A typical essay assignment highly vulnerable to AI automation. Download PDF
-
Resilient Assessment Brief (scores 12-18: Pedagogical Sovereignty) A portfolio and viva assessment designed to resist AI automation. Download PDF
Download either PDF and upload directly to the tool to see how the audit works.
To trial the tool with your own assessment:
-
Find an existing brief - Use an assessment you currently run, or one from a colleague (with permission)
-
Create a simple test brief - Write a basic assignment in Word:
- Assignment title and module
- The task (what students must do)
- Word count and deadline
- How it will be marked (even a simple list of criteria)
- Submission method
-
Start simple, then add detail - Try a bare-bones brief first, then add more information to see how the scores change
Tip: The tool is most revealing when you upload assessments you suspect might be vulnerable. The audit will identify specifically which of the 10 categories are causing the problem.
The audit evaluates uploaded documents across ten critical dimensions:
- Final product weighting: Proportion of credit distributed across formative vs. summative stages.
- Iterative documentation: Requirement for evidence of the messy middle of learning (drafts, rejected ideas, process logs).
- Contextual specificity: Connection to specific local, temporal, or classroom contexts that AI cannot access.
- Reflective criticality: Requirement for deep, subjective personal synthesis rather than generic reflection.
- Temporal friction: Time-based constraints that make rapid completion physically impossible.
- Multimodal evidence: Requirement for non-textual outputs (audio, physical artifacts, hand-drawn elements).
- Explicit AI interrogation: Requirement for students to critique or analyse AI-generated outputs.
- Real-time defence: Presence of live interaction (viva, presentation, unscripted Q&A).
- Social and collaborative labour: Requirement for verified group work and peer accountability.
- Data recency: Reliance on events, datasets, or information from the last fortnight to two weeks.
Each category is scored from 1 (Slow AI/Resilient) to 5 (Fast AI/Vulnerable), with a total possible score of 50.
- 10-20: Pedagogical Sovereignty - Your assessment prioritises human agency and is resilient to AI automation.
- 21-35: Structural Drift - Your assessment has integrity debt that may lead to shallow learning and automation risk.
- 36-50: Critical Integrity Failure - Your assessment is currently highly vulnerable to AI automation.
"Could not extract text from document"
- PDF: Check it's not a scanned image. Try selecting text in your PDF viewer - if you can't highlight text, the tool can't read it either.
- Word: Ensure it's .docx format (not .doc). Open in Word and use Save As to convert.
"Error fetching URL"
- The page may require login. Copy the text manually and use "Paste Text" instead.
- The page may load content with JavaScript. Copy the visible text and paste it.
"Results seem inaccurate or generic"
- Your brief may be too short. The tool needs enough detail to analyse meaningfully.
- Include assessment criteria, submission requirements, and any process elements (drafts, presentations, etc.)
- Briefs under 200 words often lack sufficient context.
"Analysis is taking a long time"
- Very long documents (50+ pages) may be slow or hit processing limits.
- Try uploading just the assessment brief section rather than an entire module handbook.
- Clone the repository.
- Install dependencies:
pip install -r requirements.txt. - Create
.streamlit/secrets.tomland add:GEMINI_API_KEY = "your_key". - Run:
streamlit run app.py.
- Platform: Streamlit Cloud.
- API: Google AI Studio (Gemini 1.5 Flash).
- Secrets: Add
GEMINI_API_KEYin the Streamlit Cloud dashboard under Advanced Settings > Secrets.
This tool is a proof-of-concept for the Integrity Debt framework. High scores indicate high vulnerability, threatening institutional reputation and pedagogical validity.
- Consultancy: Contact sam.illingworth@gmail.com for curriculum audit and redesign workshops.
- Slow AI: theslowai.substack.com - Critical AI literacy for educators and professionals.