{"id":"W3163329370","doi":"10.1002/bin.1793","title":"Further evaluating interobserver reliability and accuracy with and without structured visual‐inspection criteria","year":2021,"lang":"en","type":"article","venue":"Behavioral Interventions","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Brock University","funders":"","keywords":"Visual inspection; Psychology; Reliability (semiconductor); Context (archaeology); Artificial intelligence; Computer science; Geography","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003532688,0.0001939587,0.0005846497,0.0001429459,0.0002503376,0.0002500174,0.00009038426,0.0001110971,0.00111206],"category_scores_gemma":[0.001066704,0.0002198479,0.0001368072,0.0001515333,0.0001336506,0.0007266761,0.0001138828,0.0002002799,0.00006543373],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002134223,"about_ca_system_score_gemma":0.00005813669,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008307005,"about_ca_topic_score_gemma":0.001703037,"domain_scores_codex":[0.9966609,0.0003783274,0.001972675,0.0006607087,0.00007949837,0.0002478519],"domain_scores_gemma":[0.9982404,0.0001442223,0.0009108696,0.0003793279,0.0001798152,0.0001453569],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007596432,0.0006364321,0.9815466,0.0008864792,0.0001189981,0.000003976667,0.004401222,0.0000312679,0.0004671423,0.007763285,0.001372318,0.002696352],"study_design_scores_gemma":[0.001844404,0.0006175155,0.9788234,0.0006683421,0.00009867184,0.00008316535,0.004246042,0.005684852,0.0000965662,0.00625868,0.001070536,0.000507783],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9897059,0.001164245,0.003682413,0.004228557,0.0003761927,0.0004653656,0.0001461165,0.00008461047,0.0001465724],"genre_scores_gemma":[0.9906635,0.0000313317,0.007989355,0.0004651229,0.00009303564,0.0001055973,0.00006368481,0.0000327021,0.0005556587],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005653584,"threshold_uncertainty_score":0.999801,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4952346674730059,"score_gpt":0.5462249051868463,"score_spread":0.05099023771384043,"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."}}