{"id":"W2995471644","doi":"10.3138/cjpe.53365","title":"“Deliverology” and Evaluation: A Tale of Two Worlds","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Corporate governance; Politics; Context (archaeology); Value (mathematics); Public value; Public management; Critical reflection; Political science; Public relations; Reflection (computer programming); Complement (music); Public administration; Sociology; Economics; Management; Law; Computer science; History","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"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":true,"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":true,"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":true,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2749453,0.00161982,0.002767086,0.0125422,0.01388811,0.0454921,0.003683038,0.01096349,0.006414883],"category_scores_gemma":[0.2062234,0.00113255,0.001535997,0.007275216,0.1629716,0.03843717,0.02206307,0.02164122,0.0008812744],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0417209,"about_ca_system_score_gemma":0.04709981,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01615343,"about_ca_topic_score_gemma":0.01416118,"domain_scores_codex":[0.6286276,0.3196268,0.007641132,0.007214174,0.03215472,0.004735502],"domain_scores_gemma":[0.7007428,0.2306902,0.007041307,0.0218274,0.03031097,0.009387316],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00005701758,0.00005213449,0.0005776202,0.0005653782,0.0000598825,0.0001101167,0.02393775,0.0003188284,0.0001291287,0.9088526,0.01846356,0.04687611],"study_design_scores_gemma":[0.00007256641,0.0001354614,0.0006762002,0.004633483,0.00005834758,0.000160442,0.03050996,0.000787959,0.0004754619,0.6694869,0.2928455,0.0001577347],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.006102795,0.03432305,0.118187,0.6961049,0.007767021,0.00075632,0.00008925844,0.0003436368,0.1363261],"genre_scores_gemma":[0.6515962,0.02214062,0.154225,0.1361746,0.00685117,0.002642073,0.0001109131,0.0008676141,0.02539171],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7250547,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2895418597302168,"score_gpt":0.5351081862029722,"score_spread":0.2455663264727554,"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."}}