{"id":"W4382363007","doi":"10.1145/3589806.3600043","title":"Fingerprinting and Building Large Reproducible Datasets","year":2023,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Computer science; Documentation; Context (archaeology); Data science; Data mining; Software; Scale (ratio); Empirical research; Software quality; Quality (philosophy); Software engineering; Software development; 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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0586818,0.001852707,0.00232318,0.01295505,0.002734164,0.008695997,0.006188073,0.003453693,0.001809641],"category_scores_gemma":[0.2243226,0.001677542,0.003359999,0.01564706,0.002661601,0.01119719,0.01084737,0.003756977,0.002677608],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001570804,"about_ca_system_score_gemma":0.00593739,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002958032,"about_ca_topic_score_gemma":0.003925174,"domain_scores_codex":[0.9324459,0.02573776,0.01025797,0.0157417,0.01412164,0.001695013],"domain_scores_gemma":[0.7235565,0.08712204,0.01467716,0.1505251,0.02229027,0.001828929],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001103006,0.001056345,0.1119997,0.005225108,0.002258118,0.001339654,0.003026529,0.09664903,0.03525405,0.0642736,0.07326298,0.6045519],"study_design_scores_gemma":[0.0005359288,0.0008725145,0.05407098,0.002137342,0.00104406,0.001619042,0.002468378,0.3293662,0.06638577,0.282423,0.2585505,0.000526236],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05158165,0.00192291,0.8921241,0.002212306,0.0007596516,0.001773276,0.0338477,0.01194741,0.003830959],"genre_scores_gemma":[0.1220186,0.001110534,0.7805802,0.0007489584,0.0004017206,0.00411457,0.08861921,0.001425605,0.0009805893],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9413182,"threshold_uncertainty_score":0.3103427,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02638882800382723,"score_gpt":0.3086968651925593,"score_spread":0.2823080371887321,"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."}}