{"id":"W3205038101","doi":"10.3138/cjpe.69949","title":"Evaluation Advisory Groups: Considerations for Design and Management","year":2021,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Advisory committee; Resource (disambiguation); Scale (ratio); Political science; Computer science; Public administration; Geography","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.03593303,0.0001123648,0.000203115,0.0005149912,0.0003739121,0.0006305628,0.0001423244,0.00006511903,0.002127168],"category_scores_gemma":[0.007407591,0.00009757996,0.00009672814,0.000458097,0.00006805038,0.0006114302,0.00001198167,0.00009910088,0.00001868562],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003761099,"about_ca_system_score_gemma":0.006648301,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002279899,"about_ca_topic_score_gemma":0.004161042,"domain_scores_codex":[0.9948416,0.001319551,0.0008631463,0.0002588322,0.002492053,0.0002248592],"domain_scores_gemma":[0.9889579,0.0008493058,0.0004868468,0.0002554452,0.009117982,0.0003324876],"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":[0.00001222341,0.00004494325,0.0008849272,0.00001004549,0.00008611879,0.000007702852,0.0008316045,0.01322418,0.00004840296,0.004986677,0.009357181,0.970506],"study_design_scores_gemma":[0.003357148,0.0005638952,0.02992739,0.00008055454,0.000809556,0.0001575859,0.003219378,0.6007414,0.0002818728,0.3216194,0.03900947,0.0002323403],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3019251,0.01429758,0.5973284,0.03727256,0.00683541,0.02481628,0.00005905089,0.00005927576,0.0174064],"genre_scores_gemma":[0.8612571,0.00007151374,0.13734,0.0004540054,0.0001532918,0.0005161923,0.00002016115,0.00001140141,0.0001763193],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9702737,"threshold_uncertainty_score":0.9989831,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5679625182669052,"score_gpt":0.5376190604386261,"score_spread":0.03034345782827919,"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."}}