{"id":"W4379794606","doi":"10.1098/rsos.221553","title":"Predicting and reasoning about replicability using structured groups","year":2023,"lang":"en","type":"article","venue":"Royal Society Open Science","topic":"Species Distribution and Climate Change","field":"Environmental Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Defense Advanced Research Projects Agency","keywords":"Replication (statistics); Heuristics; Replicate; Psychology; Protocol (science); Cognitive psychology; Reputation; Computer science; Field (mathematics); Social psychology; Artificial intelligence; Data science; Statistics; Mathematics; Social science","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":[0.3952654,0.001505803,0.001859627,0.00845072,0.002472888,0.006196036,0.004082333,0.002728253,0.002583277],"category_scores_gemma":[0.7613174,0.001203418,0.002397596,0.002904003,0.008774624,0.01044475,0.005750012,0.002841543,0.0005023815],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003196127,"about_ca_system_score_gemma":0.00238186,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001070837,"about_ca_topic_score_gemma":0.001382402,"domain_scores_codex":[0.5880994,0.3192796,0.02604747,0.02403319,0.03998221,0.002558086],"domain_scores_gemma":[0.1029723,0.747371,0.05791683,0.06974086,0.0205609,0.001438279],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00507234,0.00134005,0.2825513,0.006664647,0.004824377,0.00184171,0.2065263,0.02296205,0.02022075,0.08237253,0.004149091,0.3614748],"study_design_scores_gemma":[0.001233462,0.006013582,0.1141816,0.003630518,0.001495745,0.001896644,0.03634335,0.168986,0.02687921,0.622943,0.01539534,0.001001565],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.538128,0.0009548128,0.4461297,0.00187425,0.0004651061,0.005080358,0.0003337829,0.0005665831,0.00646755],"genre_scores_gemma":[0.7743343,0.0001452762,0.222284,0.0002771583,0.0001401834,0.002290142,0.0002168967,0.00005412259,0.0002577982],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6047347,"threshold_uncertainty_score":0.7457455,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03298372070893709,"score_gpt":0.3059827069795194,"score_spread":0.2729989862705823,"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."}}