LogMorph tool weights Prolog mutation tests on real beginner bugs
Analysis of 7,201 student Prolog submissions reveals that real beginner bugs cluster into predictable types, not the uniform distribution that mutation testing tools assume.
Researchers classified errors across 7,201 Prolog submissions from 265 undergraduates to map what beginners actually get wrong. Current mutation testing tools treat all faults as equally probable, generating synthetic bugs that diverge from classroom reality.
Using real-world error patterns, the team built LogMorph, with 17 mutation operators weighted by observed frequency. Realistic synthetic bugs should help automated tutors give better feedback in logic programming courses.
- 200 bug-fixing submissions
- Manual classification
- Logic programming education
- Domain
Why it mattersAutomated coding tutors trained on realistic bug patterns give students more accurate hints, which matters as CS courses rely increasingly on AI feedback tools.
Open this story in InSnip →
