{"id":"W2801790340","doi":"10.1177/1356389018765487","title":"Theory-based evaluations: Framing the existence of a new theory in evaluation and the rise of the 5th generation","year":2018,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":57,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre Intégré Universitaire de Santé et de Services Sociaux du Saguenay–Lac-Saint-Jean; Université de Sherbrooke; University of Victoria","funders":"","keywords":"Epistemology; Framing (construction); Complementarity (molecular biology); Realism; Sociology; Computer science; Philosophy","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"],"category_scores_codex":[0.1229896,0.0001540918,0.0002480612,0.0002244391,0.0004303684,0.0001247286,0.0006983181,0.00008327469,0.001711332],"category_scores_gemma":[0.02141665,0.00006981396,0.0001140901,0.001373655,0.0008581619,0.0004040151,0.00009762762,0.0001672573,0.00002819772],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001306102,"about_ca_system_score_gemma":0.00177896,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000787203,"about_ca_topic_score_gemma":0.000550226,"domain_scores_codex":[0.9809361,0.01142325,0.00131312,0.0003827791,0.005752627,0.0001921572],"domain_scores_gemma":[0.9910332,0.003868874,0.001083138,0.001111032,0.002868616,0.00003514644],"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":[0.0005085781,0.0001214138,0.01236637,0.00001243496,0.00004721638,2.714402e-8,0.03323808,0.03965747,0.00416569,0.1441031,0.000825192,0.7649544],"study_design_scores_gemma":[0.002188488,0.0000692762,0.1052326,0.00004454776,0.000141889,5.495108e-7,0.002191672,0.6534495,0.003167833,0.2332838,0.0001644405,0.00006540613],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9602085,0.001297981,0.02640411,0.004957583,0.0006424254,0.003033627,0.000005543953,0.000008794125,0.003441392],"genre_scores_gemma":[0.9981953,0.00001866241,0.0006308205,0.0004261561,0.0001753886,0.0002062516,0.000007177797,0.00000878782,0.0003314902],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.764889,"threshold_uncertainty_score":0.9992012,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2883079238159595,"score_gpt":0.5359311604701663,"score_spread":0.2476232366542068,"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."}}