{"id":"W2616845762","doi":"10.3138/cjpe.328","title":"A Case Study of the Guiding Principles for Collaborative Approaches to Evaluation in a Developmental Evaluation Context","year":2017,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Context (archaeology); Set (abstract data type); Reflection (computer programming); Engineering ethics; Sociology; Psychology; Knowledge management; Management science; Computer science; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":["metaresearch"],"domain":"methods","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1447383,0.0007205839,0.0007409055,0.003101965,0.0205853,0.01003691,0.003785076,0.005783069,0.002136397],"category_scores_gemma":[0.1250324,0.0008281135,0.00117007,0.003679082,0.01136266,0.008499083,0.01234934,0.008813492,0.0003891419],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0136269,"about_ca_system_score_gemma":0.02092484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008347312,"about_ca_topic_score_gemma":0.02076016,"domain_scores_codex":[0.7520316,0.2261308,0.00505473,0.002682207,0.009460728,0.00463991],"domain_scores_gemma":[0.8177432,0.140926,0.006975984,0.008471324,0.01831669,0.007566898],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0005688821,0.004710488,0.0330016,0.001702823,0.000112701,0.02653833,0.5716545,0.003947338,0.002006196,0.1582539,0.01064825,0.186855],"study_design_scores_gemma":[0.0004543808,0.002167401,0.01405163,0.005834496,0.0001810368,0.01473355,0.714609,0.01151316,0.006668979,0.05889986,0.1706149,0.000271656],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6528193,0.004745585,0.2041906,0.04494192,0.000576153,0.007495424,0.0001645612,0.0001539862,0.0849124],"genre_scores_gemma":[0.8871142,0.001186786,0.1058203,0.001432035,0.00004175797,0.001907466,0.00003927751,0.000034227,0.002423826],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1447383,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9106579329392426,"score_gpt":0.5628987801781327,"score_spread":0.3477591527611099,"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."}}