{"id":"W6904130683","doi":"10.13140/rg.2.2.35393.45923","title":"Canadian Association of Pathologists Statistics for Pathologists Primer: Focus on Diagnostic Accuracy Statistics Task 3 Measuring Agreement Workshop","year":2018,"lang":"en","type":"article","venue":"","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Focus (optics); Task (project management); Association (psychology); Diagnostic accuracy; Task force; Summary statistics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002217216,0.0002112933,0.0005297239,0.000109936,0.0001494038,0.00004286617,0.0001316794,0.000232304,0.0003731483],"category_scores_gemma":[0.08820546,0.0001802783,0.0000782047,0.0001671828,0.0001017707,0.000082634,0.0000331649,0.0002750373,0.000105512],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006222014,"about_ca_system_score_gemma":0.0006896793,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.008294419,"about_ca_topic_score_gemma":0.1350643,"domain_scores_codex":[0.9975162,0.0002586991,0.0008433126,0.000387972,0.0004899733,0.0005038008],"domain_scores_gemma":[0.9818491,0.01528122,0.0007112157,0.0004531639,0.00133736,0.0003679649],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002746089,0.002472094,0.3604893,0.0007165943,0.001022271,0.0004820838,0.001535749,0.00004937214,0.00246295,0.08827338,0.180387,0.3593631],"study_design_scores_gemma":[0.02102332,0.01485439,0.7195076,0.0009856039,0.003394653,0.00002848574,0.001148954,0.005047398,0.004587589,0.02129767,0.2061547,0.001969578],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2565453,0.000580187,0.651929,0.0463931,0.0035364,0.009177529,0.0211888,0.0002882012,0.01036146],"genre_scores_gemma":[0.9560142,0.0001375212,0.03936355,0.002539853,0.0003876212,0.00007156525,0.0002138757,0.00002843657,0.001243387],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6994689,"threshold_uncertainty_score":0.9983094,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07252068649582594,"score_gpt":0.3763816300789241,"score_spread":0.3038609435830982,"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."}}