{"id":"W4366449572","doi":"10.3138/cjpe.0021.004","title":"Will Evaluation Prosper in the Future?","year":2007,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Professionalization; Accountability; Audit; Public sector; Public relations; Professional standards; Business; Political science; Accounting; Public administration; Sociology; Engineering ethics; Engineering; Law","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.1171451,0.0001185937,0.0001821191,0.0009232468,0.0002026848,0.000436083,0.0006730786,0.00009429253,0.002379502],"category_scores_gemma":[0.002975161,0.00007173151,0.0001132707,0.001545637,0.00007748537,0.001246858,0.000006936251,0.0003112969,0.00006900783],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006352165,"about_ca_system_score_gemma":0.004801555,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002261265,"about_ca_topic_score_gemma":0.09114942,"domain_scores_codex":[0.9916764,0.001113771,0.001252468,0.0002046903,0.005386578,0.0003660959],"domain_scores_gemma":[0.9943455,0.0003594056,0.0006682636,0.0003563513,0.004009115,0.0002613608],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00001585346,0.00004215163,0.06645361,0.000001314982,0.000007203652,0.000005767629,0.002716804,0.001014953,0.00001349237,0.001084506,0.002187364,0.926457],"study_design_scores_gemma":[0.001409847,0.0003867812,0.7832654,0.00002904359,0.0000765287,0.0000784437,0.006737735,0.03460921,0.00003567822,0.02604055,0.1471912,0.0001395996],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9687702,0.001673226,0.001443879,0.01127858,0.002329064,0.003142959,0.000002553174,0.000006370106,0.01135314],"genre_scores_gemma":[0.9971578,0.000009196498,0.001399246,0.000567627,0.0007210475,0.00007129558,0.000009730231,0.000006992642,0.00005704399],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9263174,"threshold_uncertainty_score":0.9985325,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2834746849839425,"score_gpt":0.5446628427125925,"score_spread":0.26118815772865,"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."}}