{"id":"W4251721468","doi":"10.1177/0265532220925448","title":"Understanding writing quality change: A longitudinal study of repeaters of a high-stakes standardized English proficiency test","year":2020,"lang":"en","type":"article","venue":"Language Testing","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Language proficiency; Psychology; Test (biology); Sophistication; Linguistics; Mathematics education","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002241288,0.0001224476,0.0003421897,0.00008010985,0.000408273,0.00007068421,0.000206812,0.00005387907,0.00002839875],"category_scores_gemma":[0.01870777,0.0001279636,0.00004833745,0.0008254434,0.0001545475,0.0001931282,0.00005227977,0.0001639019,5.450574e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001197702,"about_ca_system_score_gemma":0.0001413651,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01034827,"about_ca_topic_score_gemma":0.0008358911,"domain_scores_codex":[0.9978936,0.0003301819,0.0005530925,0.0003204525,0.0005815259,0.0003211011],"domain_scores_gemma":[0.9974143,0.001463405,0.0004942503,0.0001634391,0.0003573686,0.0001072413],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001970686,0.0002170055,0.5127835,0.00020176,0.00001398079,0.000009674598,0.4819844,0.000007215635,0.002245616,0.001125266,0.00001413296,0.001377782],"study_design_scores_gemma":[0.0008441821,0.0004019697,0.06764398,0.0003562781,0.00004923287,6.725659e-7,0.929633,0.00007016296,0.0007148705,0.00009175786,0.000003507175,0.0001904475],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9934057,0.00009116431,0.000314361,0.000275711,0.00009609381,0.0004807782,0.00002620062,0.0002294369,0.005080551],"genre_scores_gemma":[0.997425,0.000001650696,0.00186095,0.00003498076,0.0006196122,0.0000208457,0.0000046631,0.00001642345,0.0000158506],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4476486,"threshold_uncertainty_score":0.9962419,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4041793967710166,"score_gpt":0.4111835522633015,"score_spread":0.007004155492284891,"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."}}