{"id":"W4313552311","doi":"10.1145/3579632.3558395","title":"Closing the Assessment Excellence Gap: Why digital assessments should go beyond recall and be more inclusive","year":2023,"lang":"en","type":"article","venue":"eLearn","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Quest University Canada","funders":"","keywords":"Closing (real estate); Excellence; Recall; Implementation; Computer science; Action (physics); Public relations; Psychology; Political science; Cognitive psychology; Software 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":["sts"],"consensus_categories":[],"category_scores_codex":[0.001329034,0.0001243581,0.0001246142,0.00006239041,0.001392888,0.0007552521,0.0003607381,0.00009439114,0.0002744735],"category_scores_gemma":[0.000669532,0.0001020925,0.00003867576,0.0006243739,0.0003583257,0.0008837884,0.0002412185,0.0004692054,0.0001569019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001458976,"about_ca_system_score_gemma":0.0003618229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000585671,"about_ca_topic_score_gemma":0.000135145,"domain_scores_codex":[0.9979134,0.0004285617,0.0001807798,0.0003244454,0.0007766799,0.0003761777],"domain_scores_gemma":[0.9983788,0.0009749919,0.0001758585,0.0002262137,0.0001104587,0.000133743],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000324163,0.0003336111,0.3793203,0.00005679958,0.0001903278,0.0000558334,0.2283395,0.0003679684,0.0004425949,0.02020853,0.2732784,0.09737378],"study_design_scores_gemma":[0.0001067121,0.00003801365,0.03331339,0.00002014651,0.00001747325,0.000001548263,0.02870512,0.0001594069,0.000005464008,0.0009600587,0.9365278,0.0001448479],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6021587,0.0001588852,0.0001748852,0.2592679,0.0009626337,0.000464447,0.00001058682,0.0003220367,0.13648],"genre_scores_gemma":[0.96729,0.0002801422,0.0005613273,0.00254046,0.000352855,0.00003465121,0.0000194918,0.00002421476,0.02889686],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6632494,"threshold_uncertainty_score":0.9999071,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08205664865131557,"score_gpt":0.4645999048396862,"score_spread":0.3825432561883706,"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."}}