{"id":"W2901602856","doi":"10.1016/j.infbeh.2018.09.010","title":"Should I test more babies? Solutions for transparent data peeking","year":2018,"lang":"en","type":"review","venue":"Infant Behavior and Development","topic":"Health, Environment, Cognitive Aging","field":"Environmental Science","cited_by":29,"is_retracted":false,"has_abstract":false,"ca_institutions":"Concordia University","funders":"Concordia University; Natural Sciences and Engineering Research Council of Canada; Fonds de Recherche du Québec-Société et Culture; Canada Research Chairs","keywords":"False positive paradox; Sample (material); Test (biology); Data collection; Computer science; Statistics; Artificial intelligence; Mathematics; Biology","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001204474,0.0005850826,0.0009122203,0.00009998523,0.0007715925,0.00008034997,0.000743812,0.0002958166,0.0005737238],"category_scores_gemma":[0.0001075047,0.0005094235,0.0001124079,0.0001652614,0.00035567,0.0002623719,0.00115518,0.0003622119,0.0002659997],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005554362,"about_ca_system_score_gemma":0.0002376107,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000096002,"about_ca_topic_score_gemma":0.0002206744,"domain_scores_codex":[0.9960911,0.0000710758,0.0009342621,0.001476121,0.0005219631,0.0009054428],"domain_scores_gemma":[0.998113,0.0002925494,0.0002998226,0.00096254,0.00001189836,0.0003202045],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00000180563,0.0001543525,0.002457106,0.001062626,0.00002634753,0.00001164358,0.0004548996,5.632459e-7,0.000002919824,0.000001629798,0.001395386,0.9944307],"study_design_scores_gemma":[0.0002203108,0.00004159024,0.009264248,0.001742735,0.0005036239,0.00004706853,0.00005968465,0.0000215762,0.000004634423,0.000002689449,0.9874172,0.0006746636],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0004735378,0.9822455,0.007406901,0.0001254704,0.0004954534,0.006260973,0.002095726,0.0001448298,0.0007516167],"genre_scores_gemma":[0.0002316274,0.9845194,0.01081026,0.0001978732,0.0001150256,0.001831935,0.001933991,0.00009813353,0.0002618045],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9937561,"threshold_uncertainty_score":0.9997357,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3214620153321917,"score_gpt":0.4081116767200691,"score_spread":0.08664966138787744,"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."}}