{"id":"W4206214771","doi":"10.1001/jamainternmed.2020.7817","title":"Error in Number","year":2020,"lang":"en","type":"article","venue":"JAMA Internal Medicine","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002843139,0.00004441072,0.00009308707,0.00002280124,0.00003567247,0.00001262909,0.0001689826,0.00004104765,0.009693275],"category_scores_gemma":[0.0004768542,0.00003595809,0.00001451565,0.0001482165,0.0001072584,0.000104053,0.00001615789,0.0001701542,0.0003195657],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004419202,"about_ca_system_score_gemma":0.00008806637,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008951009,"about_ca_topic_score_gemma":0.002208777,"domain_scores_codex":[0.999306,0.00007307443,0.0001388582,0.00009909132,0.0002585416,0.0001244733],"domain_scores_gemma":[0.9995879,0.0001773784,0.00003760802,0.00003459791,0.00003730053,0.0001252082],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00006545919,0.00005031314,0.677398,0.00001359177,0.00001326928,0.00003899671,0.08588692,0.000001675491,0.000135541,0.06456759,0.1592564,0.01257228],"study_design_scores_gemma":[0.0005114175,0.00006023487,0.07990891,0.00006976016,0.000004344703,0.000001345037,0.01780887,0.00004659796,0.00001047685,0.002188383,0.8993247,0.00006498927],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2988717,0.00004301842,0.00005224899,0.2781287,0.0005299333,0.00005505635,8.023486e-7,0.00001810076,0.4223005],"genre_scores_gemma":[0.9821043,0.00002967194,0.00007726876,0.009413816,0.002934513,0.000005308991,0.000003786337,0.000003659416,0.005427691],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7400683,"threshold_uncertainty_score":0.9976485,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1013137847362322,"score_gpt":0.4578451593047975,"score_spread":0.3565313745685653,"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."}}