{"id":"W4296087229","doi":"10.3390/app12189017","title":"Evidence-Based Software Engineering: A Checklist-Based Approach to Assess the Abstracts of Reviews Self-Identifying as Systematic Reviews","year":2022,"lang":"en","type":"article","venue":"Applied Sciences","topic":"Software Engineering Research","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Systematic review; Checklist; Computer science; Management science; Data science; Engineering ethics; Psychology; Risk analysis (engineering); MEDLINE; Medicine; Political science; Engineering; Cognitive psychology","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":[0.4763815,0.01099466,0.01749703,0.09168075,0.007815364,0.01791826,0.01342694,0.009153719,0.01278822],"category_scores_gemma":[0.6578933,0.007607281,0.02787305,0.0668128,0.01078974,0.01639148,0.02064726,0.01347995,0.006291002],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02406823,"about_ca_system_score_gemma":0.1116432,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006698166,"about_ca_topic_score_gemma":0.01338523,"domain_scores_codex":[0.2802054,0.4329333,0.2188044,0.01118913,0.05435384,0.002513976],"domain_scores_gemma":[0.2293527,0.4289213,0.0936561,0.04030989,0.2000998,0.007660197],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.001491187,0.0004864395,0.002841213,0.4843167,0.008934155,0.0007321787,0.01658105,0.002727203,0.004291023,0.03344743,0.06423462,0.3799168],"study_design_scores_gemma":[0.005128226,0.001865847,0.01087412,0.4092826,0.01789778,0.001910989,0.01113793,0.009994393,0.007481189,0.1014004,0.4207524,0.002274054],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"empirical","genre_scores_codex":[0.003716848,0.03013266,0.4035522,0.01203692,0.003486368,0.5193005,0.01272129,0.003738004,0.01131515],"genre_scores_gemma":[0.00551114,0.005183558,0.8341346,0.0008588601,0.0001702472,0.1515672,0.001788039,0.000199391,0.0005870265],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5236185,"threshold_uncertainty_score":0.6457148,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1682083824029681,"score_gpt":0.3307259528974853,"score_spread":0.1625175704945172,"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."}}