{"id":"W4400267760","doi":"10.1145/3649217.3653545","title":"Are a Static Analysis Tool Study's Findings Static? A Replication","year":2024,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Static analysis; Computer science; Replication (statistics); Programming language; Mathematics; Statistics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1653622,0.002313326,0.002101963,0.007155766,0.003338865,0.005446758,0.005765331,0.002639158,0.007016897],"category_scores_gemma":[0.5421864,0.001601964,0.004471302,0.004899585,0.006168291,0.009388487,0.006080703,0.002873804,0.002360605],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00367423,"about_ca_system_score_gemma":0.006316929,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01548656,"about_ca_topic_score_gemma":0.008456971,"domain_scores_codex":[0.8311525,0.07034694,0.02684756,0.02609896,0.0410113,0.00454268],"domain_scores_gemma":[0.4149958,0.2881475,0.03752112,0.1616217,0.09418527,0.003528688],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005003075,0.002563598,0.5487362,0.01566241,0.009351704,0.003019164,0.1528464,0.001466998,0.01002554,0.01869475,0.03305997,0.1995702],"study_design_scores_gemma":[0.001795007,0.006029123,0.7172453,0.01495637,0.005772159,0.002453212,0.09885153,0.005171149,0.01360848,0.02470041,0.1085993,0.0008179062],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8292795,0.01762615,0.06243339,0.02037574,0.0105315,0.01007519,0.008931959,0.001790142,0.03895636],"genre_scores_gemma":[0.9626486,0.001096177,0.017024,0.006021352,0.0006274354,0.007399319,0.001716724,0.0007254456,0.002740862],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8346378,"threshold_uncertainty_score":0.8745295,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02653728832208449,"score_gpt":0.3193673178622187,"score_spread":0.2928300295401343,"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."}}