{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2723688,0.0008286124,0.00174415,0.01064376,0.01069573,0.02262862,0.003644809,0.003877523,0.005984748],"category_scores_gemma":[0.298512,0.0006422098,0.0006688263,0.00734354,0.04109269,0.02500782,0.01881148,0.006183357,0.0004727539],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01412619,"about_ca_system_score_gemma":0.02318132,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003629763,"about_ca_topic_score_gemma":0.005341526,"domain_scores_codex":[0.5651576,0.4029247,0.005797733,0.005340577,0.01711977,0.003659564],"domain_scores_gemma":[0.3778877,0.541665,0.01818694,0.02167646,0.03177176,0.008812181],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002056109,0.00101159,0.01993892,0.003612632,0.0002085523,0.0003882669,0.3828173,0.001213372,0.0006079891,0.3754376,0.00661499,0.2079432],"study_design_scores_gemma":[0.000175479,0.0007060266,0.0142436,0.01248741,0.0001392624,0.0004024376,0.5558316,0.003391275,0.002112803,0.2893775,0.1209622,0.0001704535],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.3638635,0.03850247,0.1410915,0.1570833,0.001700894,0.002379614,0.0002077074,0.0003457305,0.2948253],"genre_scores_gemma":[0.9780473,0.002966447,0.0129854,0.003061397,0.0001860391,0.0007511856,0.00003142274,0.00007437327,0.001896414],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7276312,"threshold_uncertainty_score":0.8972988,"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."}}