{"id":"W4234048446","doi":"10.1093/med/9780199685219.003.0015","title":"Reporting test results","year":2014,"lang":"en","type":"book","venue":"","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Reliability (semiconductor); Test (biology); Flow chart; Scale (ratio); Chart; Medical physics; Psychology; Medicine; Medical education; Engineering; Statistics; Cartography; Geography","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":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.05189667,0.0002959364,0.001062124,0.0008592693,0.0001623057,0.0003739906,0.001461575,0.0004167197,0.0009163471],"category_scores_gemma":[0.9063559,0.0001838453,0.0003295934,0.0009110921,0.00009683804,0.00008022392,0.0004265128,0.0005394657,0.001521147],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006809203,"about_ca_system_score_gemma":0.0002965092,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001095661,"about_ca_topic_score_gemma":0.0000103234,"domain_scores_codex":[0.9902193,0.0002444506,0.005706719,0.00125962,0.002174523,0.000395377],"domain_scores_gemma":[0.4789743,0.4887901,0.02547159,0.004880922,0.001498435,0.000384611],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004059945,0.000005279483,0.0009221993,0.000003764542,0.000005358163,0.00002419429,0.00001197642,0.000003590081,0.000002729907,0.0009901166,0.8088129,0.1892139],"study_design_scores_gemma":[0.0001315637,0.00006723705,0.002246636,0.0000546469,0.00000817026,0.00004180037,0.00002005869,0.0001639729,0.000007310887,0.1578712,0.8391632,0.0002241567],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.00001296123,0.0001371412,0.01039511,0.0004982068,0.00150108,0.000135615,0.00003201588,0.0002040319,0.9870839],"genre_scores_gemma":[0.0001847353,0.000009031442,0.04903941,0.0003751764,0.0009359055,0.000002922076,0.00001849254,0.00002597463,0.9494084],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.8544592,"threshold_uncertainty_score":0.999997,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7504532026062126,"score_gpt":0.5341415242884985,"score_spread":0.2163116783177141,"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."}}