Academic Integrity Forensics

Academic Integrity Forensics

Goal

Identify suspicious similarity clusters across student submissions while preserving fairness, explainability, and teacher control.

Inputs

• mission.json
• rubric.md
• submissions/ and/or evidence/
• Optional logs or metadata (timestamps, commit traces, test traces)

Produce

Write:

• integrity-report.json (global integrity assessment)
• integrity-summary.csv (per-student risk indicators)
• integrity-cases/<case-id>.md (cluster-level evidence sheets)

Analysis signals

• token-level similarity after normalization
• n-gram fingerprint overlap
• control-flow keyword profile similarity
• shared rare-error patterns when available
• metadata anomalies when available

Output semantics

integrity-report.json must include:

• overall_status: pass | review | fail
• pair_findings: suspicious pair list with scores and triggered signals
• clusters: grouped suspicious submissions
• method: scoring and thresholds used
• policy_note: explicit statement that no automatic grade penalty is applied

Rules

• Never auto-penalize students based on similarity signals alone.
• Always provide evidence and alternative explanations.
• Exclude obvious boilerplate/template regions when possible.
• Require human attestation before disciplinary action.