{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002425939,0.0001860817,0.0002566208,0.00009307412,0.0006046724,0.0003998704,0.0004541024,0.00005504723,0.004917641],"category_scores_gemma":[0.000488829,0.0001273147,0.0000584767,0.00005842226,0.0002248547,0.00003494092,0.00001174911,0.0007081973,0.00001606367],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001415241,"about_ca_system_score_gemma":0.0009948016,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.5438957,"about_ca_topic_score_gemma":0.8933942,"domain_scores_codex":[0.9989067,0.0000486216,0.0002721746,0.0001568338,0.0002231844,0.0003924718],"domain_scores_gemma":[0.9985276,0.0001296809,0.0002050646,0.0001745847,0.0001146977,0.0008483655],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002734168,0.000003528342,0.0003968856,0.000005657093,0.00009264734,0.0005080843,0.9782016,0.00003650696,0.000004612928,0.005939275,0.01191669,0.002867154],"study_design_scores_gemma":[0.0003015455,0.00007725032,0.00004769245,0.00004066253,0.00004290308,0.00000428835,0.5710645,0.00002748236,0.00002121809,0.0000210988,0.4282118,0.000139473],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8444871,0.00204834,0.00003503449,0.001807691,0.000775427,0.0001628159,0.0003439078,0.00001902346,0.1503206],"genre_scores_gemma":[0.9904575,0.000002625578,0.0002976845,0.005792372,0.003206818,0.000003795792,0.00006213691,0.00003059853,0.0001464851],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4162951,"threshold_uncertainty_score":0.995992,"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."}}