{"id":"W6958654200","doi":"10.6084/m9.figshare.c.6662674","title":"Supplementary material from \"Predicting and reasoning about replicability using structured groups\"","year":2023,"lang":"en","type":"other","venue":"Figshare","topic":"","field":"","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Replication (statistics); Heuristics; Replicate; Protocol (science); Field (mathematics); Subject (documents); Domain (mathematical analysis); Qualitative reasoning","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","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008066461,0.001614095,0.001140888,0.004308686,0.001771961,0.003685826,0.003614644,0.002220734,0.8535767],"category_scores_gemma":[0.1211007,0.001356005,0.001512575,0.006267811,0.0008090721,0.004410218,0.003985123,0.001384963,0.4148281],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002411671,"about_ca_system_score_gemma":0.004060567,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00518103,"about_ca_topic_score_gemma":0.01404418,"domain_scores_codex":[0.9954669,0.00145694,0.0005863122,0.0005018956,0.001679556,0.0003083546],"domain_scores_gemma":[0.8975208,0.07664163,0.002152667,0.0100662,0.01143847,0.00218019],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001142682,0.00005964282,0.0005183725,0.0006946066,0.00001404209,0.00006942682,0.0001803618,0.0002611066,0.000127745,0.00312625,0.9749029,0.01993134],"study_design_scores_gemma":[0.0003411327,0.00006257762,0.003145358,0.0009993374,0.00002745571,0.0002839954,0.0002111332,0.001180107,0.0007786485,0.01779644,0.9750974,0.00007654378],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"other","genre_scores_codex":[0.002025234,0.0002541051,0.03630504,0.002776809,0.001755917,0.001012697,0.8277026,0.02219349,0.105974],"genre_scores_gemma":[0.01674276,0.0007853771,0.08878858,0.00254476,0.0005467227,0.005955904,0.6854747,0.04141869,0.1577426],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9919335,"threshold_uncertainty_score":0.208855,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03727258727552094,"score_gpt":0.2941631147844553,"score_spread":0.2568905275089344,"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."}}