{"id":"W4366449610","doi":"10.3138/cjpe.22.001","title":"Methodological and Conceptual Challenges in Studying Evaluation Process Use","year":2007,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Operationalization; Process (computing); Construct (python library); Context (archaeology); Exploratory research; Process management; Quality (philosophy); Knowledge management; Management science; Computer science; Qualitative research; Conceptual model; Scale (ratio); Psychology; Sociology; Epistemology; Business; Engineering","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1153398,0.000127164,0.0003065356,0.000967785,0.0001429884,0.0002896587,0.0002718073,0.000117434,0.000612001],"category_scores_gemma":[0.02073213,0.00009567966,0.00005864655,0.0007021486,0.0001660719,0.001099658,0.00001240333,0.0002734936,0.000008953082],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004017851,"about_ca_system_score_gemma":0.00268061,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001749158,"about_ca_topic_score_gemma":0.03652841,"domain_scores_codex":[0.9929923,0.002061782,0.001197673,0.000296294,0.00311334,0.0003385754],"domain_scores_gemma":[0.9933177,0.001679734,0.0006731537,0.0001866097,0.003743317,0.0003995029],"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":[0.00002607035,0.00003696245,0.1253528,0.00000308702,0.00001077067,0.000005326743,0.00569647,0.001491698,0.00001921432,0.000533612,0.00003567023,0.8667883],"study_design_scores_gemma":[0.001685011,0.0006230227,0.915161,0.00006250143,0.00007230428,0.00004261631,0.02489732,0.03975852,0.00006410205,0.01413625,0.003341377,0.0001559915],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9922889,0.002361224,0.001144001,0.001301488,0.0004129914,0.001346307,0.00000130808,0.000005888917,0.001137926],"genre_scores_gemma":[0.9945027,0.00006971697,0.005143095,0.00009718464,0.0001151282,0.00004966801,0.000003686429,0.000006676995,0.00001216458],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8666323,"threshold_uncertainty_score":0.9875166,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9321012710267086,"score_gpt":0.6602703593788898,"score_spread":0.2718309116478188,"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."}}