{"id":"W2171507080","doi":"10.1177/1356389011430371","title":"Evaluation models and evaluation use","year":2012,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":59,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Sherbrooke; Université de Montréal","funders":"Canadian Institutes of Health Research; U.S. Public Health Service","keywords":"Context (archaeology); Management science; Psychological intervention; Process (computing); Perspective (graphical); Field (mathematics); Evaluation methods; Impact evaluation; Systematic review; Affect (linguistics); Program evaluation; Computer science; Knowledge management; Process management; Psychology; Political science; Engineering; Artificial intelligence; Medicine; MEDLINE","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":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1294791,0.0002182865,0.0002443094,0.0004763856,0.0003357339,0.0005037873,0.0002409877,0.0001367602,0.007703298],"category_scores_gemma":[0.0125624,0.0001769626,0.0000790741,0.0007856099,0.00006728505,0.006292223,0.00008957772,0.000148192,0.0006465879],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005734608,"about_ca_system_score_gemma":0.0008599508,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002201438,"about_ca_topic_score_gemma":0.00008454061,"domain_scores_codex":[0.97816,0.004084202,0.0009581088,0.0005798193,0.01579661,0.0004212246],"domain_scores_gemma":[0.9907592,0.001266883,0.0004907554,0.0007430733,0.006521757,0.0002183534],"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":[0.00002554527,0.0001227148,0.01624074,0.000002877659,0.00003193146,3.588135e-8,0.00368302,0.04743944,0.0004399017,0.004128042,0.004689848,0.9231959],"study_design_scores_gemma":[0.001175243,0.00003720746,0.1623908,0.00001081563,0.0003346908,0.000003663857,0.0006582409,0.7754046,0.000173651,0.05766772,0.001963241,0.0001801079],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9668698,0.001564567,0.006812742,0.0008694028,0.001497331,0.002281789,0.000008145826,0.00005124658,0.02004499],"genre_scores_gemma":[0.9968998,0.00004582698,0.001446365,0.0002728429,0.0004208759,0.0004829321,0.000109272,0.00001768266,0.0003044425],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9230158,"threshold_uncertainty_score":0.9957552,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6626932025379391,"score_gpt":0.5990196685991209,"score_spread":0.06367353393881814,"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."}}