{"id":"W2318621088","doi":"10.1515/cercles-2014-0009","title":"Investigating language assessment literacy: Collaboration between assessment specialists and Canadian university admissions officers","year":2014,"lang":"en","type":"article","venue":"Language Learning in Higher Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Test (biology); Psychology; Language assessment; Literacy; Medical education; Diversity (politics); Language proficiency; Pedagogy; Applied psychology; Sociology; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007579179,0.0001120635,0.0001430573,0.0002483808,0.0005663913,0.000247501,0.0001522125,0.0001103694,0.000604775],"category_scores_gemma":[0.0001949489,0.0001304997,0.00002118485,0.0005816133,0.00006449571,0.000373299,0.00003271094,0.0003051005,0.000008909798],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008614584,"about_ca_system_score_gemma":0.002234581,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1105537,"about_ca_topic_score_gemma":0.08602412,"domain_scores_codex":[0.9983748,0.0006389482,0.0001612017,0.000256642,0.000291958,0.0002764379],"domain_scores_gemma":[0.9991,0.0002210207,0.0001328293,0.0001186715,0.00008488597,0.0003425567],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000001403361,0.00003082765,0.8917998,0.00001425615,0.00001000603,0.000002326742,0.07616906,0.00002885815,0.0002807098,0.01545312,0.0004749319,0.01573471],"study_design_scores_gemma":[0.0001979963,0.0000204717,0.8444734,0.00005619885,0.00001819299,9.367626e-8,0.05362011,0.00006836411,0.000004943682,0.00003739857,0.1013584,0.0001444169],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.899168,0.00004941629,0.00001582563,0.007787144,0.0004651335,0.000241478,0.00000421076,0.0000682373,0.09220058],"genre_scores_gemma":[0.9792876,0.00001357978,0.002928527,0.0003346135,0.0008916664,0.000005978269,0.0001422322,0.00001244304,0.01638336],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1008834,"threshold_uncertainty_score":0.9306536,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01794507060251066,"score_gpt":0.3650585029343927,"score_spread":0.347113432331882,"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."}}