{"id":"W4407771524","doi":"10.1145/3641554.3701809","title":"How Effective and Efficient are Student-Written Software Tests?","year":2025,"lang":"en","type":"article","venue":"","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ontario Tech University","funders":"Universitas Brawijaya","keywords":"Computer science; Software testing; Software; Software engineering; Programming language","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02385528,0.0006533524,0.0007540851,0.002735707,0.0002930717,0.003982977,0.001208001,0.001215639,0.001467001],"category_scores_gemma":[0.2746381,0.000297101,0.0004134533,0.002173176,0.001410984,0.005798251,0.001110867,0.0008292366,0.0009654947],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006855711,"about_ca_system_score_gemma":0.00156427,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001208891,"about_ca_topic_score_gemma":0.002004208,"domain_scores_codex":[0.9472227,0.03313598,0.003873416,0.002762429,0.01152289,0.001482507],"domain_scores_gemma":[0.645355,0.253372,0.05186038,0.01499339,0.03006324,0.004355959],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0004553663,0.0004695482,0.3723715,0.0003978307,0.0001504041,0.0002072857,0.001961275,0.002298343,0.003174368,0.0008012973,0.002238435,0.6154745],"study_design_scores_gemma":[0.0002179763,0.003577923,0.8957169,0.001803448,0.0003683101,0.002835564,0.009583189,0.03600002,0.02444247,0.00795675,0.0173221,0.0001754248],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9560221,0.003100157,0.02731186,0.003772341,0.0001258821,0.0001871664,0.0002788569,0.0006094904,0.008592112],"genre_scores_gemma":[0.9831777,0.000689541,0.01438081,0.0003680285,0.00006387648,0.00005878194,0.0002407493,0.0001075571,0.0009128643],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02385528,"threshold_uncertainty_score":0.1261603,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.008202861097243937,"score_gpt":0.2658796855053294,"score_spread":0.2576768244080854,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}