{"id":"W2943350688","doi":"10.5539/ies.v12n5p109","title":"The Impact of Electronic Tests on Students’ Performance Assessment","year":2019,"lang":"en","type":"article","venue":"International Education Studies","topic":"Online and Blended Learning","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Handwriting; Mathematics education; Reading comprehension; Test (biology); Psychology; Reading (process); Achievement test; Comprehension; Standardized test; Computer science; Linguistics; Artificial intelligence","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004891617,0.00005033878,0.00006736274,0.00004901565,0.0002100882,0.00003986459,0.0002557129,0.00001456816,0.00008905346],"category_scores_gemma":[0.0002524324,0.00003218517,0.00004792662,0.00009922555,0.00006247236,0.0001098407,0.00003266199,0.0001022264,0.00005820163],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004229531,"about_ca_system_score_gemma":0.001038949,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001467972,"about_ca_topic_score_gemma":0.00008269039,"domain_scores_codex":[0.9991705,0.0000549241,0.0001240841,0.00008694838,0.0004375787,0.0001259989],"domain_scores_gemma":[0.998865,0.0002576855,0.0001144184,0.00007832782,0.0006688142,0.00001575618],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001678026,0.0002391401,0.9145688,0.000003486949,0.0002577826,2.850317e-8,0.002050425,0.0003036571,0.00003611931,0.02281481,0.002884679,0.05682431],"study_design_scores_gemma":[0.0001487045,0.0002127719,0.8202832,0.00005201541,0.000007004867,2.767547e-7,0.01459521,0.00006296147,0.00002417781,0.001180332,0.1633637,0.00006963763],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9699926,0.0002599927,0.000001352903,0.002423279,0.001237793,0.0001044815,8.289376e-7,0.00001045331,0.02596925],"genre_scores_gemma":[0.9870119,0.001094853,0.00005582737,0.00004072513,0.0002101718,0.00001900238,0.000002711277,0.000003142625,0.01156164],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1604791,"threshold_uncertainty_score":0.184305,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03183750560199882,"score_gpt":0.5092079431401298,"score_spread":0.477370437538131,"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."}}