{"id":"W2490078899","doi":"","title":"SYMPOSIUM: The Validity of Testing: Perceptions of Stakeholders","year":2016,"lang":"en","type":"article","venue":"ITC 2016 Conference","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Perception; Psychology; Computer science","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005381899,0.00006789997,0.0001243705,0.00002835274,0.0001738458,0.00002088115,0.0004181351,0.00004697196,0.001124568],"category_scores_gemma":[0.0002634643,0.00003654579,0.00005258949,0.0001512467,0.0006907726,0.0001330193,0.00005232443,0.00004369314,0.00005589889],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002488582,"about_ca_system_score_gemma":0.0003055483,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009855132,"about_ca_topic_score_gemma":0.0003486647,"domain_scores_codex":[0.9990144,0.0001513041,0.0001836607,0.0001232649,0.0003492462,0.0001780978],"domain_scores_gemma":[0.9986442,0.0006396754,0.0001519098,0.0002053334,0.0003114074,0.00004742827],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001107526,0.0001328638,0.7741662,0.00001480191,0.0000330005,5.383382e-7,0.01198048,6.834846e-7,0.1389858,0.05503018,0.007247318,0.01239703],"study_design_scores_gemma":[0.001144905,0.0003679672,0.924099,0.0004862216,0.0001249947,7.959905e-7,0.03797141,0.00002766125,0.004674027,0.01008857,0.02055263,0.000461807],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8165427,0.0000125807,0.001427666,0.004427387,0.0002592625,0.0002388211,0.00002465544,0.00003184946,0.1770351],"genre_scores_gemma":[0.9935519,0.0001492319,0.0001799606,0.00001934237,0.00007338949,0.000007342303,4.040955e-7,0.000003753814,0.006014707],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1770092,"threshold_uncertainty_score":0.9997885,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.240144715236295,"score_gpt":0.3671960779230274,"score_spread":0.1270513626867324,"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."}}