{"id":"W7096583644","doi":"","title":"Washback in language assessment","year":2013,"lang":"en","type":"article","venue":"","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Language assessment; Quarter (Canadian coin); Language proficiency; English language; Criterion-referenced test","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0002770924,0.00005002234,0.00007372296,0.00004612811,0.00008141379,0.0001005482,0.0001652478,0.00003950918,0.01278497],"category_scores_gemma":[0.00001619792,0.00004257583,0.00002514199,0.0001897961,0.00004841805,0.000240341,0.00003427668,0.00006908585,0.000975728],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008093517,"about_ca_system_score_gemma":0.00005508772,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01140916,"about_ca_topic_score_gemma":0.004944234,"domain_scores_codex":[0.9992412,0.00006385092,0.0001006522,0.0001074068,0.0002561377,0.0002307384],"domain_scores_gemma":[0.9997649,0.0000533363,0.00001910406,0.00008273641,0.00002315424,0.00005679519],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000001168139,0.0001709866,0.8247873,0.000003723534,0.00001101675,0.00000654099,0.02210432,0.000001417105,0.001411228,0.1068329,0.0306038,0.01406561],"study_design_scores_gemma":[0.0003972167,0.00001791878,0.8935949,0.000007238287,0.000002664667,7.513235e-8,0.08715221,0.0001105097,0.00003625093,0.00113966,0.01737901,0.0001623837],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.4830998,0.00001318113,0.00002350966,0.001656134,0.00009329527,0.0001650627,1.059654e-7,0.00003665627,0.5149122],"genre_scores_gemma":[0.955054,0.00002500527,0.001468514,0.0002938797,0.00009364822,0.00003227515,0.00000148461,0.000003837631,0.04302736],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4719542,"threshold_uncertainty_score":0.9998021,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01766615138195563,"score_gpt":0.3784169880295634,"score_spread":0.3607508366476077,"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."}}