{"id":"W2320230425","doi":"10.3138/cjpe.29.1.104","title":"Contributing Factors to the Continued Blurring of Evaluation and Research: Strategies for Moving Forward","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Terminology; Confusion; Management science; Psychology; Control (management); Political science; Engineering ethics; Computer science; Economics; Engineering; Artificial intelligence","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1005672,0.00009515938,0.0002440403,0.0006575753,0.0005051011,0.0006387408,0.000334495,0.00005913134,0.0001509262],"category_scores_gemma":[0.02945998,0.00006113902,0.00008748136,0.000655,0.0001078343,0.0005497179,0.00002311308,0.0001792747,0.000003683107],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00024311,"about_ca_system_score_gemma":0.003372903,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004938818,"about_ca_topic_score_gemma":0.01587394,"domain_scores_codex":[0.9942142,0.001414665,0.0009070383,0.0001892224,0.00291198,0.0003629418],"domain_scores_gemma":[0.9866313,0.002037655,0.0006556973,0.0002381924,0.01013763,0.000299465],"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":[0.00003328875,0.00001484998,0.02605238,0.00001196044,0.00003312097,1.483119e-7,0.007084948,0.01301886,0.0003254178,0.003619096,0.001143136,0.9486628],"study_design_scores_gemma":[0.002167854,0.001574917,0.2103112,0.0001856047,0.0001780379,0.000008496697,0.03319602,0.6550601,0.0009971518,0.06237079,0.03376278,0.0001869562],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.97953,0.0002423961,0.01400395,0.002659184,0.0004257922,0.002451989,0.000009547482,0.000003598538,0.0006735432],"genre_scores_gemma":[0.9978887,0.000001976962,0.00173863,0.00005591743,0.00016429,0.0001144838,0.00000697566,0.000007397364,0.00002160553],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9484758,"threshold_uncertainty_score":0.9787153,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5421588467804487,"score_gpt":0.5906286609896096,"score_spread":0.04846981420916097,"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."}}