{"id":"W4389494812","doi":"10.2139/ssrn.4650826","title":"A Comprehensive Framework for Evaluating Software Engineering Technologies","year":2023,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Data Stream Mining Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"Wycliffe College","funders":"","keywords":"Software engineering; Computer science; Systems engineering; Engineering","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":[0.01374756,0.002677389,0.00246765,0.01912953,0.001627248,0.009762122,0.002454967,0.002090683,0.003010602],"category_scores_gemma":[0.03539621,0.0005968488,0.001782443,0.008017601,0.002038842,0.008459787,0.004488843,0.001780138,0.001061699],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002827658,"about_ca_system_score_gemma":0.005436962,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007381021,"about_ca_topic_score_gemma":0.009359462,"domain_scores_codex":[0.9815926,0.00344504,0.002636803,0.001186785,0.01028436,0.0008544777],"domain_scores_gemma":[0.9798995,0.007435965,0.002475129,0.001954088,0.007279029,0.000956372],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002234403,0.001004644,0.03225752,0.0009228218,0.0006374069,0.0003080796,0.0008801082,0.1142411,0.01149604,0.2437888,0.012512,0.581728],"study_design_scores_gemma":[0.00007451783,0.00132913,0.01331404,0.0006690749,0.0003574687,0.0004681791,0.001304465,0.6903233,0.008628292,0.2605279,0.02281531,0.0001883399],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02827431,0.0009193978,0.9530081,0.000747045,0.00005353032,0.0009606955,0.001184078,0.001736667,0.01311618],"genre_scores_gemma":[0.233941,0.0004888545,0.761384,0.0001124779,0.00005282881,0.0008080859,0.001258174,0.00008773014,0.001866857],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01912953,"threshold_uncertainty_score":0.07270491,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03318069035774149,"score_gpt":0.3221317097187211,"score_spread":0.2889510193609797,"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."}}