Triangulate automated behavioral flags with oral interviews, exam-performance comparison, and targeted plagiarism detection¶

Strategy · All strategies
Evidence · 1 claim (1 for) · 1 study (1 qualitative), q2 · 0 of 1 report an effect size · 1 claim rests on one study

Description¶

Because API-level telemetry captures interaction patterns but not intent, the article recommends treating anomaly flags as a screening tool rather than evidence of misconduct. Instructors should "triangulate these automated flags with additional assessment layers": follow-up oral interviews probing conceptual understanding of submitted code, comparison of exam performance against platform activity, and targeted plagiarism detection on notebook cells against known generative AI outputs.

Design Implications¶

Context¶

Requirements¶

• Instructor time to conduct follow-up oral interviews with flagged students
• Access to exam performance records and plagiarism-detection tooling for notebook cells

Constraints¶

• A student may leave an automated script running for practical reasons unrelated to dishonesty, or may be genuinely working offline and later uploading solutions

Target Learners¶

• undergraduate and master's students in monitored programming courses

Target Learning Goals¶

• corroborated evaluation of authentic coding effort
• fair follow-up review of flagged anomalous workflows

Claims¶

• Universities converge on caution toward AI-detection systems, warning against using detection scores as dispositive evidence of misconduct [+W]

Related Strategies¶

Examples¶

-

Key Sources¶

• Germán Arroyo, Luis López, and Juan Carlos Torres. (2026). Securing the Sandbox: A Rootless Containerized Framework for Process-Oriented Monitoring in Computer Graphics Education. https://arxiv.org/abs/2606.05929