{"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":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001775019,0.0001132191,0.0001281729,0.000009737814,0.001315375,0.0004379631,0.0008865853,0.000052996,0.004635668],"category_scores_gemma":[0.0002551838,0.0001026626,0.00005096267,0.001140799,0.001168874,0.0004953874,0.002604701,0.000122235,0.0001064083],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005858847,"about_ca_system_score_gemma":0.00003283685,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001558605,"about_ca_topic_score_gemma":0.0001332403,"domain_scores_codex":[0.9981217,0.00002568585,0.0001685012,0.0006836536,0.0005220041,0.0004784969],"domain_scores_gemma":[0.9992986,0.00003986728,0.00008241227,0.0003652738,0.00001979677,0.0001940172],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006712501,0.00002082947,0.9663914,0.00001100051,0.000004658642,0.000001897372,0.002120745,0.0006097788,0.02408828,0.0006879143,0.00346094,0.002595844],"study_design_scores_gemma":[0.0001760303,0.00001533325,0.9178463,0.00001098144,0.000005142647,0.000004704603,0.003856083,0.0759443,0.000811561,0.0001912476,0.0009912383,0.0001470545],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.987502,0.00001273969,0.00007111402,0.0001898411,0.0001144809,0.0002594049,0.00002695875,0.00009725762,0.01172614],"genre_scores_gemma":[0.9982857,0.00001491733,0.001067314,0.0002348569,0.00002210179,0.00001128919,0.000008142727,0.000007279788,0.0003484091],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07533452,"threshold_uncertainty_score":0.9999848,"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."}}