Mitigate extraction errors by requiring high-resolution PDFs, equation-editor equations, and legible graph axes, units, and legends¶

Strategy · All strategies
Evidence · no claims cited

Description¶

Because extraction errors in equations, figures, tables, and graphs can generate plausible but insufficiently supported feedback or invalid assessments, the article recommends submission practices that improve automated retrieval: students submit reports as high-resolution PDF files avoiding low-quality scans or photographs; equations are written with the word processor's equation editor rather than inserted as images, since fractions and symbols embedded as images may be difficult to read automatically; and graphs include clearly legible axes, units, and legends within the figure itself.

Design Implications¶

Context¶

Requirements¶

• Instructors should pay particular attention to AI observations depending on mathematical expressions, numerical values and units, graphs and labels, data tables, experimental diagrams, and images.

Constraints¶

• When feedback is classified as doubtful or invalid, the article recommends a focused conversational second review rather than discarding the tool altogether.

Target Learners¶

• higher education students submitting laboratory reports

Target Learning Goals¶

• accurate automated retrieval of evidence from student reports

Affordances¶

• Evidence Present Versus Effectively Available Distinction

Related Strategies¶

• Use joint AI processing of report sets to identify recurring difficulties and guide collective feedback and teaching priorities
• Design rubric criteria and model instructions in observable, verifiable terms with citable evidence requirements
• Three-stage hybrid workflow: API batch review, identification of doubtful cases, focused conversational teacher review
• Strategy-detail-verify educator review of generated representation variants
• Mitigate transcription failures through image-quality supports: quality review before submission, preview transcriptions, and multiple submission modes

Examples¶

-

Key Sources¶

• Marcos Abreu, Cecilia Stari, Arturo C. Martí. (2026). AI-Assisted Assessment of Experimental Physics Laboratory Reports: Potential, Limitations, and Support for Teaching Practice. arXiv. https://arxiv.org/abs/2609.22417