{"id":"W2159715729","doi":"10.1145/1852786.1852789","title":"Can we evaluate the quality of software engineering experiments?","year":2010,"lang":"en","type":"article","venue":"","topic":"Software Engineering Techniques and Practices","field":"Computer Science","cited_by":63,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Usability; Computer science; Checklist; Quality (philosophy); Software quality; Reliability (semiconductor); Context (archaeology); Process (computing); Quality Score; Software; Software engineering; Data science; Software development; Psychology; Human–computer interaction; Engineering; Operations management","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":["metaresearch"],"category_scores_codex":[0.6400118,0.003548121,0.007832887,0.01682989,0.002905185,0.01794936,0.00622849,0.009302222,0.003808411],"category_scores_gemma":[0.9206253,0.00190816,0.006256808,0.01117571,0.01108883,0.02466529,0.007381692,0.004339954,0.001512969],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00976582,"about_ca_system_score_gemma":0.01028832,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001824996,"about_ca_topic_score_gemma":0.001714101,"domain_scores_codex":[0.2407143,0.5645998,0.08841157,0.01918161,0.08503849,0.002054138],"domain_scores_gemma":[0.02545341,0.7185174,0.09540668,0.05850203,0.09876528,0.0033552],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.008536079,0.001584658,0.1366355,0.04292791,0.02226846,0.0004534656,0.01624271,0.01279357,0.005156122,0.02129009,0.02562965,0.7064818],"study_design_scores_gemma":[0.01191248,0.02916767,0.3113556,0.07153409,0.02380115,0.002229331,0.01620954,0.06995011,0.03756635,0.2335764,0.1880756,0.004621701],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2297918,0.1251456,0.4950383,0.07886888,0.02305264,0.01354965,0.003242346,0.003817547,0.0274933],"genre_scores_gemma":[0.642655,0.01057873,0.3181179,0.009377221,0.004247799,0.01177458,0.0009403079,0.000846996,0.00146152],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3599882,"threshold_uncertainty_score":0.4439296,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04263385987164619,"score_gpt":0.3358986261512139,"score_spread":0.2932647662795677,"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."}}