{"id":"W2996945182","doi":"10.56645/jmde.v15i33.609","title":"Big Shoes to Fill: An Evaluation Journey in the Footsteps of Daniel L. Stufflebeam","year":2019,"lang":"en","type":"article","venue":"Journal of MultiDisciplinary Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Athabasca University; Laurentian University","funders":"","keywords":"Context (archaeology); Relevance (law); Foundation (evidence); Intervention (counseling); Program evaluation; Evaluation methods; Sociology; Engineering ethics; Psychology; Engineering; History; Political science; Archaeology; Law","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":[],"category_scores_codex":[0.08369327,0.0001906544,0.0004594211,0.001063136,0.0001396053,0.0001719831,0.001141122,0.0001024207,0.001276964],"category_scores_gemma":[0.005191211,0.0001122192,0.0001755394,0.001651208,0.00004320336,0.001322057,0.0001218168,0.000333397,0.0001137059],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000289168,"about_ca_system_score_gemma":0.001612876,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001291002,"about_ca_topic_score_gemma":0.0002162184,"domain_scores_codex":[0.9843395,0.003384002,0.002264303,0.0003593171,0.00940001,0.0002528602],"domain_scores_gemma":[0.9910341,0.001041823,0.001728225,0.0007794129,0.005255705,0.0001607622],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.0008812541,0.001389428,0.1954548,0.00002980267,0.00006514984,0.000004010657,0.05311293,0.3879632,0.01738889,0.0002937601,0.007051991,0.3363648],"study_design_scores_gemma":[0.002722431,0.001813355,0.6082678,0.0001422284,0.0001068615,0.00003266642,0.02546236,0.3534339,0.0007939024,0.006065798,0.0009931654,0.0001654782],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.987825,0.0001471477,0.00087589,0.005641827,0.001403438,0.001475259,0.000007336686,0.000004071964,0.002620046],"genre_scores_gemma":[0.997468,0.00002209572,0.001705771,0.0002255411,0.0002815925,0.00003487878,0.00001440382,0.00001259593,0.000235154],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.412813,"threshold_uncertainty_score":0.999636,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3461999936431824,"score_gpt":0.5442462073245252,"score_spread":0.1980462136813428,"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."}}