{"id":"W2139159602","doi":"10.1002/ev.92","title":"PIE à la Mode: Mainstreaming Evaluation and Accountability in Each Program in Every County of a Statewide School Readiness Initiative","year":2003,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Accountability; Mandate; Mainstreaming; Legislature; Program evaluation; Plan (archaeology); Mainstream; Benchmarking; Quality (philosophy); Public administration; Political science; Public relations; Medical education; Psychology; Business; Pedagogy; Special education; Medicine","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"],"consensus_categories":[],"category_scores_codex":[0.02727976,0.0001656699,0.0003093257,0.0007689873,0.0001309035,0.0001445582,0.0001321472,0.0001261081,0.0004491635],"category_scores_gemma":[0.01350881,0.0001506323,0.00006181662,0.001714985,0.00007781114,0.001254649,0.00002220656,0.0001723772,0.000006877861],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007648396,"about_ca_system_score_gemma":0.002174522,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005654849,"about_ca_topic_score_gemma":0.006039717,"domain_scores_codex":[0.9943261,0.001667312,0.001070939,0.0005440626,0.002153022,0.0002385515],"domain_scores_gemma":[0.9955288,0.001736531,0.0004163414,0.0003542086,0.001878742,0.00008536489],"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":[0.0002101608,0.0005914194,0.2649134,0.00002782792,0.00002262746,1.572709e-7,0.003182675,0.0242478,0.0003186727,0.0008954399,0.0004752357,0.7051146],"study_design_scores_gemma":[0.004210941,0.0001491338,0.5022712,0.00007911863,0.00007120272,0.000002677038,0.003822383,0.4356048,0.0006198055,0.05085579,0.002130345,0.0001826665],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9821094,0.0001883176,0.002773273,0.000192576,0.0002747822,0.003444088,0.00003292804,0.0000276957,0.01095698],"genre_scores_gemma":[0.9945275,0.0000551865,0.003779954,0.00003887956,0.00002675565,0.001329425,0.00007134672,0.00001165238,0.0001593403],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7049319,"threshold_uncertainty_score":0.9948008,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2119641125322008,"score_gpt":0.5393907107995318,"score_spread":0.327426598267331,"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."}}