{"id":"W4392681814","doi":"10.22318/icls2023.105241","title":"Social Context Challenges in Administering Authentic Assessments of Citizenship Competency","year":2023,"lang":"en","type":"article","venue":"Proceedings.","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vancouver Island University","funders":"","keywords":"Citizenship; Situated; Context (archaeology); Dimension (graph theory); Competency assessment; Vocational education; Pedagogy; Authentic assessment; Medical education; Psychology; Sociology; Computer science; Medicine; Political science; Artificial intelligence; Curriculum; Politics","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":[],"consensus_categories":[],"category_scores_codex":[0.0008820615,0.0001070259,0.000237129,0.0001859931,0.0002163031,0.00006384616,0.0003010793,0.00008968905,0.00008483026],"category_scores_gemma":[0.0001298367,0.0001151645,0.00006529114,0.00056312,0.0001462033,0.0001494342,0.0001062512,0.0001190209,0.00005095892],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008105631,"about_ca_system_score_gemma":0.00008713944,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000572474,"about_ca_topic_score_gemma":0.0001648565,"domain_scores_codex":[0.9985896,0.00002624183,0.0002685314,0.0002240638,0.0005066463,0.0003849045],"domain_scores_gemma":[0.9995889,0.00007265144,0.0001367265,0.00003749349,0.0001074555,0.00005675603],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00004532234,0.0001437143,0.2994901,0.0002945867,0.00006946216,0.000009169996,0.1629786,1.133353e-7,0.0008062138,0.5109993,0.00186213,0.0233013],"study_design_scores_gemma":[0.001030994,0.0001548103,0.6989754,0.0002888298,0.00003461689,5.681067e-7,0.2734408,0.000122165,0.0001287209,0.01796244,0.007456285,0.0004044178],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7268223,0.0001313116,7.53364e-7,0.0018613,0.0002220486,0.0002235365,0.00000166884,0.0001237747,0.2706133],"genre_scores_gemma":[0.9981492,0.0004177766,0.0000650792,0.00002320295,0.0001959074,0.00003398121,0.000001773425,0.00001328781,0.001099846],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4930368,"threshold_uncertainty_score":0.469627,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1870756728074141,"score_gpt":0.4202267850047601,"score_spread":0.233151112197346,"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."}}