Prolog Student Submissions with Versioning and Test Results

A dataset from figshare contains 7,201 Prolog program submissions from students, annotated with debugging and test result metadata. It includes 200 manually annotated entries and 16,000 synthetic instances generated via mutation. Author Ricardo Brancas published this collection for research on automated feedback and program repair in logic programming education.

Use Cases

• Train models to predict the 'correct' boolean label using features like 'tests_passed_count', 'total_tests_count', and 'syntax_error'.
• Analyze patterns in student debugging by tracking changes between 'program' and 'previous_submission' fields via the 'diff' string.
• Develop automated feedback systems using the 'failed_tests' list and 'category' labels (e.g., 'BUGFIX_CORRECT') from the 200 manually annotated programs.
• Generate synthetic training data for program repair models by applying mutations to the original 'program' field, as demonstrated in the 'synthetic.jsonl' file.
• Study the progression of student solutions using the 'sequence_id', 'time' stamps, and 'modified_predicates_count' across assignment submissions.

Strengths

• 7,201 real student Prolog program submissions provide a substantial corpus of authentic code.
• 200 programs have detailed manual annotations for ground-truth validation of automated methods.
• Includes 16,000 synthetic instances generated by mutating student submissions, expanding the dataset size for model training.

Limitations

• The dataset's total temporal coverage and geographic origin of the student population are not specified.
• Only 200 of the 7,201 real submissions have manual annotations, limiting supervised learning on the full set.
• The 232.0 MB size indicates a medium-scale dataset, which may be insufficient for training very large language models without augmentation.

Provenance

Source

Ricardo Brancas via figshare.

Collection Method

Collected from student Prolog program submissions within an educational context, with manual annotations added to a subset.

Freshness

Last updated on April 3, 2026.