{"id":"W4321061983","doi":"10.1109/apsec57359.2022.00071","title":"A checklist-based approach to assess the systematicity of the abstracts of reviews self-identifying as systematic reviews","year":2022,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Systematic review; Checklist; Computer science; Quality (philosophy); Field (mathematics); Management science; Data science; Engineering ethics; Psychology; MEDLINE; Political science; Epistemology; 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":[0.5151922,0.008586025,0.01267427,0.07858091,0.007690798,0.01217188,0.009687406,0.008403773,0.01405022],"category_scores_gemma":[0.7215353,0.005082108,0.0246509,0.04892991,0.009003795,0.01276466,0.01448026,0.01029623,0.005175648],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01663454,"about_ca_system_score_gemma":0.08339153,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004465867,"about_ca_topic_score_gemma":0.0109197,"domain_scores_codex":[0.2399208,0.4281262,0.2622209,0.011014,0.05657982,0.002138268],"domain_scores_gemma":[0.1469071,0.5556311,0.08381972,0.03929768,0.1694516,0.004892848],"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":[0.003354786,0.0005996617,0.007360257,0.4404133,0.0155896,0.0008218626,0.01913205,0.002390866,0.007724518,0.03149565,0.05906669,0.4120508],"study_design_scores_gemma":[0.01404564,0.00512436,0.02877285,0.2755932,0.04173308,0.002973739,0.01262946,0.01758454,0.01480892,0.1256534,0.4572587,0.003822139],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"empirical","genre_scores_codex":[0.007397077,0.02418514,0.4557863,0.009832933,0.004002624,0.4657069,0.01790001,0.004642395,0.01054669],"genre_scores_gemma":[0.01318436,0.003466809,0.7520917,0.001171067,0.0002759358,0.2261375,0.00275358,0.0001960333,0.0007229625],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4848078,"threshold_uncertainty_score":0.5978544,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1156332705048288,"score_gpt":0.3291899797183531,"score_spread":0.2135567092135244,"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."}}