{"id":"W2472075195","doi":"10.1017/s0261444815000233","title":"Review of washback research literature within Kane's argument-based validation framework","year":2015,"lang":"en","type":"article","venue":"Language Teaching","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":64,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Empirical research; Argument (complex analysis); Maturity (psychological); Systematic review; English language; Psychology; Political science; Epistemology; Mathematics education; Medicine; Philosophy; Law; Developmental psychology","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":[0.008405294,0.00009677374,0.0001930687,0.0001076777,0.0002755386,0.0001413566,0.0003822704,0.0001173853,0.0001927444],"category_scores_gemma":[0.002783663,0.00008355034,0.00006624773,0.0005155813,0.0001002474,0.000237498,0.00006173008,0.0007020785,0.00005114832],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001297221,"about_ca_system_score_gemma":0.0001785059,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007269448,"about_ca_topic_score_gemma":0.00006574538,"domain_scores_codex":[0.9968656,0.001354676,0.0002524979,0.0002029523,0.001032305,0.0002919937],"domain_scores_gemma":[0.9989012,0.0003555972,0.0001327331,0.0002900443,0.0001928764,0.0001275712],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00006070068,0.0004102897,0.02273194,0.002547823,0.00007092238,0.0000511237,0.8282193,0.00002511056,0.001215261,0.08848168,0.0392401,0.01694577],"study_design_scores_gemma":[0.003763122,0.0009055446,0.002045689,0.1229825,0.0003066562,0.000005826112,0.5872716,0.0005816926,0.005315928,0.01855049,0.2561239,0.002147038],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6932727,0.09987199,0.001739795,0.008966687,0.001415091,0.001836243,0.00002071576,0.0003516127,0.1925252],"genre_scores_gemma":[0.9798496,0.0007700308,0.0151538,0.001036793,0.0006753333,0.00002309469,0.00007431699,0.00002017043,0.002396829],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.286577,"threshold_uncertainty_score":0.3407084,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08897567591661655,"score_gpt":0.4676807480738409,"score_spread":0.3787050721572244,"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."}}