{"id":"W2087097337","doi":"10.1145/1022494.1022539","title":"Relevant empirical testing research","year":2004,"lang":"en","type":"article","venue":"ACM SIGSOFT Software Engineering Notes","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Software testing; Empirical research; Test strategy; Computer science; Software engineering; Position paper; Software; Software performance testing; Data science; Systems engineering; Engineering; Software construction; Software development; World Wide Web","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":[0.04321586,0.001303694,0.001524726,0.008815325,0.002320776,0.007399913,0.004079916,0.003105264,0.03495628],"category_scores_gemma":[0.2884843,0.0007368627,0.0009837591,0.01592642,0.006483055,0.009159607,0.003244019,0.004432277,0.00554499],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005190294,"about_ca_system_score_gemma":0.006878586,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005437464,"about_ca_topic_score_gemma":0.00433207,"domain_scores_codex":[0.9447487,0.03022456,0.003945475,0.004944469,0.01447959,0.001657119],"domain_scores_gemma":[0.3908071,0.5199423,0.02395979,0.02446138,0.03644056,0.004389015],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005510229,0.0027528,0.141827,0.01247194,0.0007042047,0.00100176,0.00599492,0.002168892,0.0008141136,0.2744958,0.07058222,0.4866354],"study_design_scores_gemma":[0.0002948529,0.0009553932,0.1801623,0.04431692,0.0009583928,0.003023837,0.02291276,0.005794452,0.003613051,0.2716529,0.4660945,0.0002207637],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1524824,0.2589893,0.05748237,0.1231039,0.00402348,0.00140912,0.007845428,0.0004412024,0.3942229],"genre_scores_gemma":[0.8057365,0.1222632,0.02416198,0.02151207,0.003680479,0.001223841,0.0090126,0.0005748136,0.01183438],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.04321586,"threshold_uncertainty_score":0.22855,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1418802265110641,"score_gpt":0.360298768827243,"score_spread":0.2184185423161789,"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."}}