{"id":"W4382752350","doi":"10.3389/feduc.2023.1192754","title":"Assessment purposes and methods used by EFL teachers in secondary schools in Jordan","year":2023,"lang":"en","type":"article","venue":"Frontiers in Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Descriptive statistics; Medical education; Psychology; Mathematics education; English as a foreign language; Pedagogy; Medicine; Mathematics","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":[],"consensus_categories":[],"category_scores_codex":[0.002112867,0.00009482175,0.0001796842,0.000457255,0.00009211901,0.0001056604,0.0001735102,0.0001007858,0.00005172755],"category_scores_gemma":[0.0001445585,0.000108384,0.00002088783,0.001014128,0.00007705919,0.0003872331,0.00003454088,0.0003145041,0.000004859744],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005778105,"about_ca_system_score_gemma":0.000699732,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002936834,"about_ca_topic_score_gemma":0.002005736,"domain_scores_codex":[0.9984564,0.0005491067,0.0002338397,0.0002563566,0.0002177331,0.0002866293],"domain_scores_gemma":[0.9996541,0.00009615492,0.00006585582,0.0001017688,0.00001597222,0.00006621984],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000002886645,0.0001038456,0.8213439,0.000006562202,0.000005202272,5.438465e-7,0.02343708,0.000002473709,0.0001161466,0.0003769153,0.04082816,0.1137763],"study_design_scores_gemma":[0.0003633917,0.00001258215,0.7551817,0.00003329777,0.000003909133,4.589889e-8,0.2149429,0.00007328708,0.00001800196,0.003832517,0.02541892,0.0001193733],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9748075,0.0007083091,0.0002022311,0.002706116,0.002180435,0.0005315251,0.000002108038,0.00003560191,0.01882616],"genre_scores_gemma":[0.9465792,0.0008731888,0.04679674,0.0001498018,0.0001409789,0.0002650307,0.00005918496,0.00001730248,0.005118616],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1915059,"threshold_uncertainty_score":0.4439635,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01925520963117444,"score_gpt":0.4148986868049889,"score_spread":0.3956434771738145,"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."}}