{"id":"W3082627313","doi":"10.31234/osf.io/gkduw","title":"Cultivating credibility with probability words and numbers","year":2019,"lang":"en","type":"article","venue":"","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"Defence Research and Development Canada","funders":"Ministère de la Défense Nationale","keywords":"Credibility; Probability distribution; Probability model; Computer science; Econometrics; Psychology; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.00273628,0.0001233897,0.0002704428,0.00006361357,0.00008619628,0.000314925,0.0003298432,0.00006201609,0.001604374],"category_scores_gemma":[0.00106705,0.00006974822,0.00004924111,0.0003326588,0.0001336045,0.0004279915,0.0001652409,0.0001251579,0.0004015931],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004204898,"about_ca_system_score_gemma":0.00005660967,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001192814,"about_ca_topic_score_gemma":0.0001869305,"domain_scores_codex":[0.9980241,0.00007669365,0.0004664011,0.000704375,0.0005330094,0.0001954261],"domain_scores_gemma":[0.9978032,0.00101137,0.0001507055,0.0007562865,0.0001664817,0.0001119439],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004220186,0.00004234806,0.7661927,0.000002045753,0.000002292831,8.605667e-7,0.0002841349,0.00006028575,0.0000715103,0.0003601186,0.0003275838,0.2326139],"study_design_scores_gemma":[0.001905436,0.0006375581,0.5559354,0.0000847406,0.00002203508,0.0000545563,0.008963185,0.006255563,0.0007199463,0.3996013,0.02489543,0.0009247992],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9631533,0.000005626062,0.0008648855,0.000227845,0.000146862,0.0002463889,0.000003732233,0.00004989331,0.03530148],"genre_scores_gemma":[0.9814572,4.597619e-7,0.01604746,0.0001128538,0.00001683775,0.000004005994,6.134798e-7,0.000006405749,0.002354195],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3992412,"threshold_uncertainty_score":0.9993083,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1096227674434059,"score_gpt":0.3752661060700869,"score_spread":0.265643338626681,"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."}}