{"id":"W4390063624","doi":"10.3138/cjpe-2023-0024","title":"Arts-Based Evaluation","year":2023,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"Lawson Foundation","keywords":"The arts; Recreation; Arts in education; Painting; Narrative; Psychology; Sociology; Visual arts; Political science; Art","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"qualitative","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":true,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"qualitative","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":["metaresearch","insufficient_payload"],"category_scores_codex":[0.05407947,0.0001257924,0.0002236151,0.001693303,0.0002766247,0.0004946691,0.0004985392,0.0000885945,0.007026814],"category_scores_gemma":[0.008443646,0.00009941217,0.0001690646,0.002411179,0.00007957761,0.0006323497,0.000009040995,0.0001871835,0.0009740839],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005670149,"about_ca_system_score_gemma":0.01254243,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000179541,"about_ca_topic_score_gemma":0.009371943,"domain_scores_codex":[0.9919808,0.0008629073,0.001069845,0.0002333202,0.005507024,0.0003460841],"domain_scores_gemma":[0.9909841,0.0004068927,0.0007507324,0.0003800765,0.006973712,0.0005045261],"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":[0.00001057377,0.00002067141,0.01625752,0.000002482729,0.0000147597,0.000004843385,0.0003984239,0.03873305,0.00004931666,0.0002624975,0.01618091,0.9280649],"study_design_scores_gemma":[0.001285175,0.0003476467,0.1500162,0.00003552467,0.0001025097,0.00001122434,0.0007622966,0.7266079,0.0001501007,0.01522135,0.1053256,0.0001344755],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9653127,0.0004169555,0.002100941,0.01202583,0.00338854,0.002729179,0.00001366444,0.00005717042,0.01395504],"genre_scores_gemma":[0.997476,0.000004034918,0.001545298,0.0003005405,0.0002524913,0.000148363,0.00005000073,0.0000121747,0.0002110792],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9279305,"threshold_uncertainty_score":0.9999086,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6412074581239966,"score_gpt":0.6406298731733805,"score_spread":0.0005775849506161057,"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."}}