{"id":"W2891656379","doi":"10.1002/ev.20346","title":"The Evolving Market for Systematic Evaluation in Canada","year":2018,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Accreditation; Professionalization; Context (archaeology); Supply and demand; Government (linguistics); Supply side; Business; Exploratory research; Affect (linguistics); Public relations; Marketing; Economics; Political science; Economic growth; Sociology; Social science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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"],"category_scores_codex":[0.03830241,0.0001357571,0.0002330049,0.0002540105,0.0009370116,0.0003191387,0.0003637202,0.00005014844,0.001459473],"category_scores_gemma":[0.02540108,0.00009128386,0.00009584875,0.0009488863,0.00003445672,0.0004608813,0.00002361854,0.00006540828,0.00004059229],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002197262,"about_ca_system_score_gemma":0.0076341,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.07362279,"about_ca_topic_score_gemma":0.9436849,"domain_scores_codex":[0.9941405,0.0009443959,0.001044904,0.0004001142,0.003174943,0.0002951584],"domain_scores_gemma":[0.9890132,0.005511147,0.0004750648,0.0005685717,0.004353782,0.00007822949],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002174891,0.00006305654,0.007075967,0.0002650531,0.00008390297,4.72551e-8,0.001481421,0.005686263,0.000134621,0.001710881,0.4264057,0.5568756],"study_design_scores_gemma":[0.001457052,0.00008495279,0.02742544,0.0002014257,0.000107867,0.000001343121,0.001751982,0.9301785,0.0001425885,0.02058451,0.01794586,0.0001184341],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4183674,0.00615795,0.2231886,0.03805948,0.04139125,0.1027075,0.000216339,0.0002981038,0.1696134],"genre_scores_gemma":[0.9877036,0.00001204603,0.001280432,0.0001833876,0.0004426396,0.003982652,0.00002753326,0.00001573564,0.006351939],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9244923,"threshold_uncertainty_score":0.9994533,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1834456181021273,"score_gpt":0.4959395639226441,"score_spread":0.3124939458205168,"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."}}