{"id":"W7099706504","doi":"","title":"TOEFL iBTTM Research Insight Reliability and Comparability of TOEFL iBT ™ Scores","year":2015,"lang":"en","type":"article","venue":"","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test of English as a Foreign Language; Comparability; Reliability (semiconductor); Test (biology); Language proficiency","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03322014,0.0004437759,0.0007967679,0.005488204,0.0008350139,0.002136203,0.00133661,0.001007179,0.01498677],"category_scores_gemma":[0.230708,0.0003937508,0.001626314,0.003134022,0.001624204,0.002698358,0.002481368,0.001334098,0.004206182],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00157628,"about_ca_system_score_gemma":0.001433901,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007891451,"about_ca_topic_score_gemma":0.008706408,"domain_scores_codex":[0.966696,0.01052913,0.004439134,0.003906302,0.01302827,0.001401248],"domain_scores_gemma":[0.855997,0.0654351,0.010539,0.01490681,0.05103801,0.002084082],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005148946,0.0003989453,0.5275596,0.0008829144,0.00145605,0.0002553617,0.005821805,0.002421762,0.002113146,0.01664332,0.1234388,0.3138593],"study_design_scores_gemma":[0.000481779,0.001857322,0.8888548,0.0006650573,0.000499108,0.0007443928,0.003051639,0.008709679,0.003801202,0.01466431,0.07643396,0.0002368304],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7184758,0.001695655,0.09476104,0.002897936,0.00255178,0.00303621,0.03599663,0.00302563,0.1375594],"genre_scores_gemma":[0.9532644,0.0001962906,0.01902568,0.0003630874,0.0002177676,0.003717314,0.01202822,0.0008937325,0.0102935],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03322014,"threshold_uncertainty_score":0.175687,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1394942875517064,"score_gpt":0.3431285007982764,"score_spread":0.20363421324657,"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."}}