{"id":"W4366384728","doi":"10.3138/cjpe.0025.009","title":"Navigating Expectations, Values and Context: A Canadian Evaluator Abroad","year":2011,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Context (archaeology); Politics; Political science; Psychology; Public relations; Sociology; History; Law; Archaeology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":true},"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":[],"domain":null,"study_design":"design_other","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01302475,0.0001644122,0.0002777023,0.0005661823,0.0004965338,0.0004451798,0.0004304566,0.0001028617,0.004495015],"category_scores_gemma":[0.004216396,0.0001353984,0.000107617,0.0006988454,0.0001705245,0.0009318901,0.00001117441,0.0003290223,0.0001205124],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005527466,"about_ca_system_score_gemma":0.009426222,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.07461027,"about_ca_topic_score_gemma":0.6390299,"domain_scores_codex":[0.995653,0.0007100424,0.001090609,0.0002669739,0.001871736,0.0004076709],"domain_scores_gemma":[0.9936472,0.0002205304,0.0007116297,0.0002950145,0.003817981,0.001307599],"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":[0.000008716142,0.00001692221,0.055688,0.000002627353,0.00002897654,0.000009616746,0.01741064,0.00003589058,0.000009383557,0.0003553092,0.001770074,0.9246638],"study_design_scores_gemma":[0.004083289,0.002200952,0.6452364,0.0004620416,0.0004960721,0.0004536601,0.1087494,0.1124897,0.0005133256,0.08099003,0.04338913,0.0009359333],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9874447,0.002860579,0.0002278511,0.001668119,0.0009523461,0.001218949,0.00001562005,0.0000128169,0.005599028],"genre_scores_gemma":[0.9932708,0.00001467131,0.006039664,0.0003456643,0.0001428493,0.00008025703,0.000009846634,0.00001403988,0.00008226998],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9237279,"threshold_uncertainty_score":0.996415,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3205900033941147,"score_gpt":0.530989583284331,"score_spread":0.2103995798902163,"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."}}