{"id":"W2923578044","doi":"10.1186/s40468-019-0078-7","title":"The language assessment literacy needs of Iranian EFL teachers with a focus on reformed assessment policies","year":2019,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":59,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Rubric; Psychology; Literacy; Curriculum; Pedagogy; Active listening; Mathematics education; Language assessment; Alternative assessment; Perception; Focus group; Sociology","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":[0.001297913,0.0002024818,0.0002800356,0.0001848834,0.0002733534,0.0002027784,0.0005256163,0.00008174123,0.0000902723],"category_scores_gemma":[0.0002155173,0.0001315704,0.00006249021,0.0009569668,0.0001742015,0.0002002915,0.00008140898,0.0003746251,0.00001823696],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003471719,"about_ca_system_score_gemma":0.0002686807,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008235031,"about_ca_topic_score_gemma":0.00268891,"domain_scores_codex":[0.9978777,0.0002786916,0.0003367868,0.0002079076,0.0007577382,0.0005412422],"domain_scores_gemma":[0.9982225,0.0009541163,0.0002653372,0.000401505,0.00007523277,0.00008127846],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002767608,0.0001522454,0.7561647,0.00002200665,0.00004108395,0.00002292083,0.205721,0.00002304503,0.002387247,0.0082842,0.00005940245,0.02709447],"study_design_scores_gemma":[0.001109891,0.0003988157,0.6625733,0.0003172536,0.00002464109,0.000002201009,0.3342867,0.0001649392,0.0001402042,0.0001161922,0.0006004857,0.0002653157],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7385437,0.00008623664,0.00001111572,0.0008582077,0.000106944,0.0004782941,0.00000327636,0.00008218337,0.25983],"genre_scores_gemma":[0.9922814,0.000008688894,0.00398759,0.0001025739,0.0001238617,0.00003919059,0.000005908945,0.00002603912,0.003424744],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2564053,"threshold_uncertainty_score":0.9983692,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01648693977985285,"score_gpt":0.3597537479867195,"score_spread":0.3432668082068666,"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."}}