{"id":"W3095967847","doi":"10.37213/cjal.2020.30458","title":"Comments from the Chalkface Margins: Teachers’ Experiences with a Language Standard, Canadian Language Benchmarks","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Applied Linguistics","topic":"Second Language Learning and Teaching","field":"Arts and Humanities","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Milestone; Portfolio; Language assessment; Protocol (science); Medical education; Mathematics education; Psychology; Computer science; Pedagogy; Medicine; Business","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0118575,0.0004776922,0.0006044651,0.001011597,0.02049852,0.004905802,0.001558681,0.003586563,0.003405738],"category_scores_gemma":[0.06622523,0.0005860379,0.0003033185,0.001311472,0.007479774,0.003123056,0.005635634,0.005702997,0.0007490773],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0300083,"about_ca_system_score_gemma":0.03051452,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.5855812,"about_ca_topic_score_gemma":0.6999627,"domain_scores_codex":[0.9827713,0.008327438,0.0007126228,0.001154219,0.005208554,0.001825894],"domain_scores_gemma":[0.9537908,0.01926765,0.002685347,0.001169533,0.01477409,0.008312646],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00006905131,0.00002439764,0.004536034,0.00009652602,0.000002996822,0.001110793,0.9729425,0.00009454961,0.001471373,0.0007526931,0.01153775,0.007361375],"study_design_scores_gemma":[0.00000578195,0.00007094,0.003661989,0.0001787767,0.000004217915,0.0002155901,0.9023808,0.0001167906,0.0005854754,0.0001497344,0.09256536,0.00006437588],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9239556,0.001619885,0.002479451,0.03896771,0.0009786609,0.0002136949,0.0004302354,0.0001673279,0.03118736],"genre_scores_gemma":[0.9737415,0.0008381077,0.00128306,0.005562794,0.00007946324,0.000143636,0.0001385959,0.0001807791,0.0180321],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4144188,"threshold_uncertainty_score":0.8337187,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01455364548171062,"score_gpt":0.2035850631241514,"score_spread":0.1890314176424408,"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."}}