{"id":"W2055314676","doi":"10.1080/02602930601122555","title":"Assessment purposes and procedures in ESL/EFL classrooms","year":2007,"lang":"en","type":"article","venue":"Assessment & Evaluation in Higher Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":61,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; Queen's University","funders":"Beijing Foreign Studies University","keywords":"Psychology; Mathematics education; Pedagogy; Teaching method; Linguistics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.006380016,0.0002066542,0.0002329562,0.0005510771,0.0002664238,0.0002343697,0.0002334461,0.0001758988,0.001826685],"category_scores_gemma":[0.0001191503,0.0002212384,0.00004040911,0.0009773329,0.0001323535,0.0007661438,0.00005259586,0.0002741425,0.00001449221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001666128,"about_ca_system_score_gemma":0.002593879,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001158552,"about_ca_topic_score_gemma":0.006333088,"domain_scores_codex":[0.9964021,0.0005053334,0.0006263299,0.0005043023,0.001464096,0.0004978437],"domain_scores_gemma":[0.9987444,0.0003447536,0.0002540711,0.0002132239,0.000313643,0.0001299681],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001061162,0.0006764329,0.9030169,0.0000252937,0.00001268116,8.216687e-7,0.004026949,0.00003370775,0.0002119078,0.07217859,0.0007105648,0.0190955],"study_design_scores_gemma":[0.0008387013,0.00005248858,0.9664642,0.00008957901,0.0000364352,3.73312e-7,0.01348338,0.0001843037,0.000007552866,0.00654562,0.01205415,0.0002432212],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8048983,0.0004378584,0.0000853541,0.004960719,0.001898643,0.001785525,0.000001002162,0.00005542497,0.1858771],"genre_scores_gemma":[0.9912133,0.0002233521,0.002844332,0.0003614158,0.0004887061,0.0005173077,0.00004835707,0.00001928427,0.004283909],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.186315,"threshold_uncertainty_score":0.9990858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07152919239186825,"score_gpt":0.4767516717717072,"score_spread":0.4052224793798389,"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."}}