{"id":"W4402746259","doi":"10.18357/otessac.2023.3.1.227","title":"Online Collaborative Testing: Design and Implementation in a Large First-Year Undergraduate Course","year":2023,"lang":"en","type":"article","venue":"The Open/Technology in Education Society and Scholarship Association Conference","topic":"Reflective Practices in Education","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Victoria","funders":"","keywords":"Course (navigation); Online course; Medical education; Computer science; Mathematics education; Psychology; Software engineering; Engineering; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.006892704,0.000125657,0.0001784768,0.0001614173,0.0009771779,0.000429352,0.000421228,0.0002912005,0.00002711931],"category_scores_gemma":[0.00412846,0.0001260456,0.00001647527,0.003236942,0.0002122675,0.001154967,0.0001683567,0.0006145964,0.00001422715],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007274326,"about_ca_system_score_gemma":0.001937452,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006448579,"about_ca_topic_score_gemma":0.008826472,"domain_scores_codex":[0.9976214,0.001058436,0.0003131636,0.0003537494,0.000282327,0.0003709193],"domain_scores_gemma":[0.99707,0.001671651,0.0004348653,0.0001713662,0.000600209,0.00005196004],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001057966,0.0002570046,0.8535029,0.000007731196,0.00002606237,2.410525e-7,0.03631821,0.00001432597,0.00007229178,0.09925748,0.001009921,0.009523248],"study_design_scores_gemma":[0.0006831372,0.00004203554,0.5475625,0.00005604908,0.0000235976,6.21427e-7,0.2406387,0.0003814581,0.00004942371,0.2057443,0.004654226,0.0001639525],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8403689,0.0002419274,0.0001826083,0.1562025,0.0003139873,0.001467396,0.00001528565,0.00009292147,0.001114448],"genre_scores_gemma":[0.9895277,0.003188433,0.005888552,0.0004381828,0.00004835203,0.0003821863,0.00003167114,0.00001068879,0.0004842422],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3059404,"threshold_uncertainty_score":0.7515761,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07646078215281538,"score_gpt":0.4520532303481831,"score_spread":0.3755924481953677,"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."}}