{"id":"W4413128915","doi":"10.1002/ev.70009","title":"Community‐Based Research on Evaluation: Beyond Academic and Evaluator Perspectives","year":2025,"lang":"en","type":"article","venue":"New Directions for Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa; University of Waterloo","funders":"Canadian Mental Health Association","keywords":"Evaluation methods; Program evaluation; Computer science; Sociology; Engineering ethics; Management science; Psychology; Political science; Public administration; Engineering","routes":{"ca_aff":true,"ca_fund":true,"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","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0765892,0.0002157374,0.0002927718,0.00202876,0.002381888,0.0003802412,0.0005549098,0.0002210901,0.001877893],"category_scores_gemma":[0.01941723,0.0001873534,0.0001341559,0.002989948,0.0002292463,0.000567909,0.00009967574,0.0008815387,0.0001144528],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008032143,"about_ca_system_score_gemma":0.00365051,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001056643,"about_ca_topic_score_gemma":0.0002932491,"domain_scores_codex":[0.9865605,0.005657405,0.0008275741,0.0007289115,0.005852534,0.0003730962],"domain_scores_gemma":[0.9860942,0.006909046,0.0002421281,0.0009942732,0.005604618,0.0001557542],"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":[0.0003698355,0.0003571015,0.002483261,0.00002064481,0.00007967331,5.287163e-8,0.004655154,0.0116708,0.001128491,0.02691375,0.1146754,0.8376458],"study_design_scores_gemma":[0.005863603,0.0006464811,0.05036207,0.0001222498,0.0002852748,0.000001397898,0.01575377,0.5815353,0.0025586,0.2541974,0.0883907,0.0002831476],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.674174,0.004150063,0.01688242,0.06072876,0.003273525,0.01057693,0.0000677313,0.0002529663,0.2298937],"genre_scores_gemma":[0.987997,0.00007788343,0.001215269,0.0008346651,0.0002624102,0.001514025,0.00008553801,0.00001791455,0.007995293],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8373627,"threshold_uncertainty_score":0.9990345,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6424948000151569,"score_gpt":0.6631852882198059,"score_spread":0.02069048820464903,"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."}}