{"id":"W2054415360","doi":"10.1016/j.stueduc.2009.12.005","title":"Evaluation for learning: A cross-case analysis of evaluator strategies","year":2009,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta; Queen's University","funders":"","keywords":"Process (computing); Citizen journalism; Participatory evaluation; Reflection (computer programming); Program evaluation; Knowledge management; Computer science; Process management; Psychology; Sociology; Political science; Engineering","routes":{"ca_aff":true,"ca_fund":false,"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","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05653324,0.0001989881,0.0005395718,0.001579867,0.000328773,0.0001655678,0.0003002257,0.00008562938,0.00214874],"category_scores_gemma":[0.03172893,0.0001668821,0.0002547673,0.003418691,0.0001882419,0.0009230479,0.00003422657,0.0001304046,0.000026486],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007306662,"about_ca_system_score_gemma":0.002908047,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001773224,"about_ca_topic_score_gemma":0.0003115428,"domain_scores_codex":[0.9903953,0.001414824,0.001568703,0.000627549,0.005737365,0.0002562412],"domain_scores_gemma":[0.9808043,0.003431085,0.0008838341,0.0004614738,0.01436255,0.00005672436],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001131356,0.0004764412,0.04781553,0.00001673139,0.0007925123,5.181113e-7,0.01177567,0.7467484,0.0002182783,0.01626085,0.001162219,0.1746197],"study_design_scores_gemma":[0.0009599401,0.0001968297,0.3201584,0.00001596046,0.001175001,0.000004327582,0.01445136,0.5031605,0.00003943631,0.1594791,0.0002264353,0.000132804],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9862316,0.001712077,0.001455247,0.002458384,0.0008283548,0.001668887,0.00002058414,0.00001381707,0.005611094],"genre_scores_gemma":[0.9965371,0.00005417718,0.001888819,0.0001228622,0.0001705878,0.0007142198,0.0001598917,0.000006811762,0.0003455035],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2723428,"threshold_uncertainty_score":0.9987634,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5328970254040822,"score_gpt":0.6892803130688502,"score_spread":0.1563832876647679,"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."}}