{"id":"W2946853871","doi":"10.1126/science.364.6443.813","title":"Psychology's reproducibility solution fails first test","year":2019,"lang":"en","type":"article","venue":"Science","topic":"Health, Environment, Cognitive Aging","field":"Environmental Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"World Federation of Science Journalists","funders":"","keywords":"Reproducibility; Psychology; Test (biology); Experimental psychology; Applied psychology; Cognitive psychology; Epistemology; Psychiatry; Philosophy; Cognition","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":["insufficient_payload"],"category_scores_codex":[0.005222633,0.0001247445,0.0001162906,0.00004963068,0.0004172029,0.0000514836,0.0006073335,0.00005111581,0.003217587],"category_scores_gemma":[0.001420347,0.0001172791,0.0000315198,0.0007529973,0.001435843,0.001026744,0.0004068864,0.0001969154,0.01295736],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004216589,"about_ca_system_score_gemma":0.00003827449,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001563142,"about_ca_topic_score_gemma":0.00009354453,"domain_scores_codex":[0.9963006,0.0000537511,0.0002253783,0.002192711,0.0006029287,0.0006246417],"domain_scores_gemma":[0.997673,0.0001117004,0.00009540664,0.001920766,0.00001012528,0.0001889556],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00000384926,0.0001068111,0.836865,0.000005933881,5.695052e-7,0.00000193145,0.0002966692,0.0001130821,0.1464493,0.00001378357,0.0003859259,0.01575719],"study_design_scores_gemma":[0.0001571542,0.00008692251,0.9763829,0.000009859836,0.000002032976,0.000008066454,0.00005255207,0.0008844871,0.00634391,0.0005183574,0.01539568,0.0001581084],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9524285,0.00001734796,0.0006302908,0.00279744,0.0003389094,0.000475215,0.000002998911,0.00006093509,0.04324832],"genre_scores_gemma":[0.996535,0.00001856286,0.001369637,0.00141161,0.00003045108,0.00001498794,9.977005e-7,0.000008120675,0.0006106014],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1401054,"threshold_uncertainty_score":0.9976936,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02082810609938397,"score_gpt":0.29980705383303,"score_spread":0.278978947733646,"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."}}