{"id":"W4406043859","doi":"10.1080/15427587.2024.2448815","title":"The impact of TOEFL iBT preparation on Chinese test-takers’ perceptions of integrated speaking and writing design","year":2025,"lang":"en","type":"article","venue":"Critical Inquiry in Language Studies","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Test (biology); Test of English as a Foreign Language; Psychology; Perception; Context (archaeology); Construct validity; Language assessment; Test design; Applied psychology; Social psychology; Medical education; Mathematics education; Developmental psychology; Psychometrics; Test method; Medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01937195,0.0005874946,0.000464477,0.001271847,0.001929791,0.003097897,0.000702403,0.0005856817,0.00212305],"category_scores_gemma":[0.05903351,0.0002891769,0.0005919235,0.0007114265,0.002533244,0.001147689,0.002690534,0.001257772,0.0002495069],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002313279,"about_ca_system_score_gemma":0.002796663,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004850521,"about_ca_topic_score_gemma":0.005439557,"domain_scores_codex":[0.9872146,0.006867002,0.0009176361,0.00058946,0.003174961,0.001236464],"domain_scores_gemma":[0.9533818,0.02480493,0.009803039,0.003158247,0.005065951,0.003786107],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000865098,0.001225612,0.3781837,0.0005142752,0.00009266222,0.001253381,0.4764106,0.0004691865,0.02274393,0.0006002795,0.0008072121,0.116834],"study_design_scores_gemma":[0.00008229495,0.004694515,0.6305172,0.0002764078,0.0001007439,0.0007088767,0.3468953,0.001613033,0.008507777,0.0006619964,0.005732791,0.0002090635],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9986149,0.00003469054,0.0003759087,0.0001002638,0.000006044875,0.00005442728,0.000008065981,0.00001017746,0.0007956141],"genre_scores_gemma":[0.9987589,0.00004619053,0.0006860471,0.0000729919,0.000006683963,0.00006587041,0.00001697562,0.000005636847,0.0003406657],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01937195,"threshold_uncertainty_score":0.1024499,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07430071452784202,"score_gpt":0.5133190650981386,"score_spread":0.4390183505702966,"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."}}