{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":12,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":12,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"ec1f2ca963e3","filters":{"venue":"Language Testing in Asia"}},"results":[{"id":"W2886638441","doi":"10.1186/s40468-018-0065-4","title":"How does anxiety influence language performance? From the perspectives of foreign language classroom anxiety and cognitive test anxiety","year":2018,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":175,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Anxiety; Psychology; Test anxiety; Foreign language; Test (biology); Foreign language anxiety; Cognition; Developmental psychology; Language assessment; Clinical psychology; Consistency (knowledge bases); Mathematics education; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.01252989119022532,"gpt":0.2365920863830132,"spread":0.2240621951927879,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005953672,0.0002993873,0.0003162064,0.00008493106,0.0004940272,0.0002973357,0.0003020028,0.00009487068,0.0002161168],"category_scores_gemma":[0.002356213,0.0001935421,0.00006983305,0.0001608188,0.0007944602,0.0004058352,0.0001450709,0.0006604454,0.00001217678],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003954934,"about_ca_system_score_gemma":0.00004167727,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004401876,"about_ca_topic_score_gemma":0.003528976,"domain_scores_codex":[0.9983842,0.0001817979,0.0002800013,0.0004660888,0.0002668684,0.0004210918],"domain_scores_gemma":[0.9980743,0.001080456,0.0002530688,0.0003466776,0.0001778765,0.0000676016],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002099709,0.0000580681,0.4244426,0.00004903699,0.00002942727,0.000009619511,0.5406032,0.00000255438,0.005143813,0.0003828808,0.00004656493,0.02921125],"study_design_scores_gemma":[0.0006721328,0.0001915657,0.1680066,0.0006251332,0.00006940213,0.00002266621,0.825694,0.0009319549,0.002693045,0.0001006557,0.0006195605,0.0003732985],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9707579,0.001771053,0.00001398903,0.0001464094,0.00007345172,0.0002607606,0.0001399537,0.0001939972,0.02664248],"genre_scores_gemma":[0.9967192,0.00004089029,0.0008657102,0.0001806378,0.0009854918,0.00002039047,0.00003309478,0.00004444307,0.001110094],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2850908,"threshold_uncertainty_score":0.7892417,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2923578044","doi":"10.1186/s40468-019-0078-7","title":"The language assessment literacy needs of Iranian EFL teachers with a focus on reformed assessment policies","year":2019,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":59,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Rubric; Psychology; Literacy; Curriculum; Pedagogy; Active listening; Mathematics education; Language assessment; Alternative assessment; Perception; Focus group; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.01648693977985285,"gpt":0.3597537479867195,"spread":0.3432668082068666,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001297913,0.0002024818,0.0002800356,0.0001848834,0.0002733534,0.0002027784,0.0005256163,0.00008174123,0.0000902723],"category_scores_gemma":[0.0002155173,0.0001315704,0.00006249021,0.0009569668,0.0001742015,0.0002002915,0.00008140898,0.0003746251,0.00001823696],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003471719,"about_ca_system_score_gemma":0.0002686807,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008235031,"about_ca_topic_score_gemma":0.00268891,"domain_scores_codex":[0.9978777,0.0002786916,0.0003367868,0.0002079076,0.0007577382,0.0005412422],"domain_scores_gemma":[0.9982225,0.0009541163,0.0002653372,0.000401505,0.00007523277,0.00008127846],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002767608,0.0001522454,0.7561647,0.00002200665,0.00004108395,0.00002292083,0.205721,0.00002304503,0.002387247,0.0082842,0.00005940245,0.02709447],"study_design_scores_gemma":[0.001109891,0.0003988157,0.6625733,0.0003172536,0.00002464109,0.000002201009,0.3342867,0.0001649392,0.0001402042,0.0001161922,0.0006004857,0.0002653157],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7385437,0.00008623664,0.00001111572,0.0008582077,0.000106944,0.0004782941,0.00000327636,0.00008218337,0.25983],"genre_scores_gemma":[0.9922814,0.000008688894,0.00398759,0.0001025739,0.0001238617,0.00003919059,0.000005908945,0.00002603912,0.003424744],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2564053,"threshold_uncertainty_score":0.9983692,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2967888896","doi":"10.1186/s40468-019-0089-4","title":"Critical review of validation models and practices in language testing: their limitations and future directions for validation research","year":2019,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":46,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Argument (complex analysis); Empirical research; Construct (python library); Computer science; Test (biology); Construct validity; Language assessment; Management science; Psychology; Data science; Psychometrics; Statistics; Mathematics education; Mathematics; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.2871225134574545,"gpt":0.4973541369068704,"spread":0.2102316234494159,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002735364,0.00006949196,0.0001502008,0.0001527026,0.0001499818,0.00008018706,0.00009263175,0.00006174558,0.0000160549],"category_scores_gemma":[0.01415882,0.00006526084,0.00001480205,0.0007933145,0.00008805448,0.000451443,0.00003817369,0.0001685078,0.000001404285],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004227183,"about_ca_system_score_gemma":0.00008187256,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001162787,"about_ca_topic_score_gemma":0.0006360557,"domain_scores_codex":[0.9987629,0.0003636256,0.0002224989,0.0002051008,0.0002461404,0.0001997506],"domain_scores_gemma":[0.9934998,0.005968047,0.0001359382,0.0001049516,0.0002546,0.00003669989],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.00003628138,0.0004542522,0.5374438,0.004326604,0.00002693351,0.00001091504,0.2725309,0.00004047256,0.0113778,0.03678451,0.0002439781,0.1367236],"study_design_scores_gemma":[0.001971348,0.0006180754,0.3427038,0.01084864,0.0001250269,0.00001389455,0.6132694,0.01257004,0.0007303106,0.01276918,0.003591396,0.0007888534],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.950391,0.009614783,0.00003289545,0.002890097,0.000101035,0.001037774,0.000009458484,0.00004831866,0.03587464],"genre_scores_gemma":[0.9844843,0.001522727,0.01356802,0.00003832878,0.0001552732,0.00007539793,0.00001775307,0.00001002022,0.0001281946],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3407385,"threshold_uncertainty_score":0.9941453,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4309467076","doi":"10.1186/s40468-022-00201-5","title":"Lessons from the Chinese imperial examination system","year":2022,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Global Educational Reforms and Inequalities","field":"Social Sciences","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Imperial examination; Language assessment; Linguistics; Set (abstract data type); Psychology; Field (mathematics); History; Pedagogy; Computer science; Philosophy; Ancient history","retraction":null,"screen_n_in":null,"score":{"opus":0.03157084234721545,"gpt":0.3443788321747615,"spread":0.312807989827546,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000991193,0.00004922734,0.00006389169,0.00002140548,0.0005765511,0.0000580021,0.0002249937,0.00001800108,0.0002365939],"category_scores_gemma":[0.00178924,0.00003400508,0.00001865283,0.0003975803,0.00004095038,0.000077365,0.00005975056,0.0001113706,0.000008436563],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002781647,"about_ca_system_score_gemma":0.0001406745,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.07144972,"about_ca_topic_score_gemma":0.003574894,"domain_scores_codex":[0.9990451,0.000287446,0.0001193804,0.0001021094,0.0003011112,0.0001447975],"domain_scores_gemma":[0.9992653,0.0005210587,0.00006343453,0.00009752299,0.00002875753,0.00002396533],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00001100065,0.0001044083,0.1310181,0.00001558686,0.00001165539,0.00002975802,0.6172789,0.0002693631,0.0009669657,0.2040165,0.000951018,0.0453268],"study_design_scores_gemma":[0.0001495954,0.00001663882,0.5286991,0.00002232146,0.000003536617,0.000002585506,0.4665159,0.0002990234,0.000005308489,0.001393366,0.002800515,0.00009217271],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9350086,0.000179904,0.00000311374,0.001083583,0.00052946,0.0001048386,0.0000390982,0.0000624903,0.06298887],"genre_scores_gemma":[0.9986337,7.020661e-7,0.0002034275,0.0001117463,0.0005432155,0.00005118388,0.00002863673,0.000004743163,0.0004225748],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.397681,"threshold_uncertainty_score":0.9347336,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1569132542","doi":"10.1186/s40468-015-0020-6","title":"Chinese university students’ perceptions of assessment tasks and classroom assessment environment","year":2015,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Psychology; Learning environment; Context (archaeology); Perception; Mathematics education; Formative assessment; Scale (ratio); Pedagogy","retraction":null,"screen_n_in":null,"score":{"opus":0.02810909251904739,"gpt":0.3653477989662329,"spread":0.3372387064471855,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001036943,0.0001250062,0.0002032653,0.0001117371,0.0001617467,0.00005206881,0.0003022655,0.00007039295,0.0001334023],"category_scores_gemma":[0.0001281235,0.0001241186,0.00003318454,0.0003240037,0.0001847441,0.0001664971,0.0002150146,0.0001870794,0.000007998821],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000412319,"about_ca_system_score_gemma":0.000162351,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002242851,"about_ca_topic_score_gemma":0.0009390415,"domain_scores_codex":[0.9983996,0.0002604751,0.000193727,0.0002383659,0.0006561989,0.0002516025],"domain_scores_gemma":[0.9993121,0.0002093647,0.0001227598,0.0001701777,0.0000404026,0.0001452128],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000003935604,0.0002067115,0.9788175,0.000006567232,0.00001354316,0.00002710288,0.01832952,0.00003672598,0.0004779334,0.0006787766,0.00006474539,0.001336982],"study_design_scores_gemma":[0.0007492424,0.00007007319,0.8975031,0.00002637691,0.00001945099,8.846732e-7,0.1008537,0.0002398718,0.000001031032,0.0001082074,0.0003066649,0.0001213934],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9419419,0.00005137185,0.0002485311,0.0002058121,0.00008667508,0.0002722003,0.00000679497,0.00004980525,0.0571369],"genre_scores_gemma":[0.9782459,0.00002711561,0.02057732,0.00001642559,0.00006138612,0.000004238641,0.000008887506,0.00000886621,0.00104992],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08252421,"threshold_uncertainty_score":0.5061409,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2984367764","doi":"10.1186/s40468-019-0094-7","title":"Assessing peer review pattern and the effect of face-to-face and mobile-mediated modes on students’ academic writing development","year":2019,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Psychology; Academic writing; English for academic purposes; Peer feedback; Mathematics education; Cohesion (chemistry); Task (project management); Second language writing; Face-to-face; Pedagogy; Medical education; Second language; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.03272756180375445,"gpt":0.399730670787874,"spread":0.3670031089841195,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003838398,0.0001330916,0.0003103133,0.00005554549,0.0001331091,0.0001071996,0.0002673445,0.00006301434,0.00001686165],"category_scores_gemma":[0.00173841,0.0000917752,0.00001942071,0.00029218,0.00008494952,0.0001520018,0.0001636997,0.0002678819,0.00001440599],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000516445,"about_ca_system_score_gemma":0.00003002464,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002210361,"about_ca_topic_score_gemma":0.00003098051,"domain_scores_codex":[0.9981391,0.0003768807,0.0002883372,0.0002286602,0.0007193938,0.0002475729],"domain_scores_gemma":[0.9975908,0.002056705,0.0001460705,0.0001069056,0.000046,0.00005349484],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000008676761,0.00000994297,0.9367851,0.0002485946,0.00001166153,0.000003662041,0.02714923,0.00001497534,0.0005578483,0.00001279793,0.00001870741,0.03517878],"study_design_scores_gemma":[0.002465648,0.0001645185,0.9365973,0.007880027,0.00005373359,0.000002261747,0.05132966,0.0005392309,0.0003291846,0.000006138485,0.0002789835,0.000353278],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9930333,0.002334074,0.00001476415,0.0004469808,0.00004423836,0.001127324,9.702906e-7,0.00003908738,0.002959202],"genre_scores_gemma":[0.9989048,0.0001295381,0.0003268345,0.0002641664,0.00003159541,0.00005697989,0.000003483803,0.00001129506,0.0002712542],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03482551,"threshold_uncertainty_score":0.3742484,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1921505225","doi":"10.1186/s40468-015-0018-0","title":"Raising the bar: language testing experience and second language motivation among South Korean young adolescents","year":2015,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University; University of British Columbia","funders":"","keywords":"Psychology; Language assessment; Test (biology); Exploratory factor analysis; Salient; Context (archaeology); Language proficiency; Language education; Developmental psychology; Mathematics education; Pedagogy; Social psychology; Psychometrics","retraction":null,"screen_n_in":null,"score":{"opus":0.1066673638797676,"gpt":0.4019436657537405,"spread":0.295276301873973,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001196652,0.0001727293,0.0001635759,0.0001165519,0.0003790079,0.0002336757,0.0002965634,0.0000952475,0.00007031806],"category_scores_gemma":[0.01178221,0.0001466103,0.00002505532,0.0006690885,0.000281807,0.0002718141,0.00008347486,0.0003070033,0.0000187586],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001225597,"about_ca_system_score_gemma":0.0001959123,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02358272,"about_ca_topic_score_gemma":0.004344651,"domain_scores_codex":[0.9982811,0.0003105792,0.000292235,0.0003105471,0.0003516006,0.0004539703],"domain_scores_gemma":[0.998784,0.0004489564,0.0001965715,0.0002605709,0.00009759096,0.0002122783],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000002386804,0.00001581847,0.4479496,0.00001283272,0.000001440253,0.00001631798,0.5419576,0.000007893404,0.0008383516,0.00004456532,0.0000213882,0.009131786],"study_design_scores_gemma":[0.0004066693,0.00001566158,0.4924166,0.0002642742,0.000007156562,0.00001066884,0.5053908,0.0007270044,0.0004317423,0.00003614063,0.00006318431,0.0002300983],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9562259,0.0002515588,0.00001701482,0.00009724402,0.0002272971,0.0002991774,0.000005636289,0.0001872283,0.04268894],"genre_scores_gemma":[0.9968168,6.188339e-7,0.001143609,0.0002691511,0.0005653978,0.00001761427,0.000007544709,0.00002469487,0.001154632],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04446694,"threshold_uncertainty_score":0.996542,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4387821100","doi":"10.1186/s40468-023-00261-1","title":"Instructional practices and students’ reading performance: a comparative study of 10 top performing regions in PISA 2018","year":2023,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Enthusiasm; Psychology; Reading (process); Multilevel model; Mathematics education; China; Sample (material); Teacher education; Pedagogy; Geography; Social psychology; Political science; Chemistry","retraction":null,"screen_n_in":null,"score":{"opus":0.1189039487434486,"gpt":0.4392416959406387,"spread":0.3203377471971901,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001106272,0.00008586227,0.0001803531,0.0002450246,0.0002254995,0.0000631839,0.0001942953,0.00004042251,0.00004181255],"category_scores_gemma":[0.0006629465,0.00008531625,0.00001134704,0.001151276,0.00009610339,0.0003471971,0.0001140482,0.0001708647,0.00000954461],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007613523,"about_ca_system_score_gemma":0.00005888296,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003298593,"about_ca_topic_score_gemma":0.00445518,"domain_scores_codex":[0.9987656,0.0001394709,0.0002451937,0.0001963367,0.0004269023,0.0002264976],"domain_scores_gemma":[0.9990399,0.000550924,0.0002436645,0.00008858718,0.00004299718,0.00003388579],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001214405,0.0001168239,0.8157675,0.00001227443,0.000008748239,0.00001349425,0.1827108,0.00002725254,0.0001591224,0.0001071485,0.00003846141,0.00102626],"study_design_scores_gemma":[0.0004513881,0.00009097797,0.6555846,0.0000816633,0.000006686676,0.000001434494,0.3431694,0.0004788203,0.00001064495,0.00001496348,0.00004333949,0.00006609773],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9801283,0.00002914688,2.665783e-7,0.00005662151,0.00009448681,0.0003288048,7.081243e-7,0.00005447915,0.01930722],"genre_scores_gemma":[0.9984033,0.00001589653,0.0006440891,0.000005131015,0.00007744503,0.00002700377,0.000002561579,0.000005682444,0.0008189497],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1604586,"threshold_uncertainty_score":0.4986508,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2112346409","doi":"10.1186/s40468-015-0016-2","title":"How language proficiency contributes to Chinese students’ academic success in Korean universities","year":2015,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Grit, Self-Efficacy, and Motivation","field":"Psychology","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Psychology; Language proficiency; Test of English as a Foreign Language; Academic achievement; Medical education; Graduate students; Mathematics education; English language; Pedagogy; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.03571550564783218,"gpt":0.3631143499755787,"spread":0.3273988443277466,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006558978,0.0001876034,0.0002471975,0.0004071264,0.00004053965,0.00006347328,0.0004871137,0.0001531512,0.00001650024],"category_scores_gemma":[0.001137608,0.0001730474,0.0000252972,0.0008938827,0.00004606253,0.0001673615,0.0001219971,0.0004107679,0.0000339168],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002157792,"about_ca_system_score_gemma":0.00006246549,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003261622,"about_ca_topic_score_gemma":0.001028545,"domain_scores_codex":[0.9985335,0.0001854125,0.0002352835,0.000353833,0.0002566888,0.0004352307],"domain_scores_gemma":[0.9991309,0.0003096101,0.0001003799,0.0002760911,0.00007182575,0.0001111786],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004406338,0.0001881591,0.9131663,0.00001611891,0.000009500326,0.000226078,0.08126364,0.00003956577,0.001881333,0.0003128351,0.0004734241,0.002378961],"study_design_scores_gemma":[0.001926845,0.00009083447,0.9062352,0.0001132534,0.000007756833,0.00001135536,0.0910452,0.00005630235,0.000190618,0.000058382,0.00007698336,0.0001873254],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9902109,0.0006793257,0.0001090796,0.0007084857,0.0002817216,0.0005232987,0.00001441335,0.0001813721,0.007291435],"genre_scores_gemma":[0.9963127,7.97734e-7,0.0003909403,0.0002315679,0.0001416545,0.00002676031,0.000040175,0.00002355493,0.002831849],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009781558,"threshold_uncertainty_score":0.7056669,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4400514459","doi":"10.1186/s40468-024-00294-0","title":"Adaptation and norm determination of the Boston Naming Test for healthy Lebanese adults aged between 50 and 88 years","year":2024,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Neurobiology of Language and Bilingualism","field":"Neuroscience","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Normative; Boston Naming Test; Psychology; Norm (philosophy); Test (biology); Cognition; Clinical psychology; Developmental psychology; Gerontology; Neuropsychology; Medicine; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.04089131030687088,"gpt":0.3074530149037648,"spread":0.2665617045968939,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000272353,0.00008520471,0.0001144175,0.000086022,0.00006757446,0.00002992796,0.00008941167,0.00005520617,0.000001234253],"category_scores_gemma":[0.003079358,0.00006662204,0.0000193448,0.0002153447,0.0000899544,0.00008575224,0.00004672453,0.0001417966,5.274068e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001483916,"about_ca_system_score_gemma":0.00002616151,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001753165,"about_ca_topic_score_gemma":0.0001933547,"domain_scores_codex":[0.9992387,0.00007353339,0.0001903686,0.0002679345,0.00008017769,0.000149301],"domain_scores_gemma":[0.9974929,0.002284358,0.00007921534,0.0001057396,0.00001332554,0.00002443487],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.00003210846,0.00003173219,0.0310995,0.0005265856,0.000001775055,0.0002234805,0.0323463,0.000003599621,0.6024109,0.00002959834,0.000008499842,0.3332859],"study_design_scores_gemma":[0.003517745,0.001956578,0.7021337,0.003408092,0.0001271266,0.00138225,0.009717518,0.03671434,0.2387107,0.001125819,0.0002641152,0.0009420536],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9987556,0.000443649,0.00005500437,0.0001651875,0.00009472301,0.0003299053,0.00002986388,0.00005742707,0.00006867422],"genre_scores_gemma":[0.9980042,0.000006286512,0.001696177,0.0001416276,0.00007418334,0.00001197367,0.000004499774,0.00001365296,0.00004737492],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6710342,"threshold_uncertainty_score":0.3686502,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4415055623","doi":"10.1186/s40468-025-00390-9","title":"Shaping written corrective feedback perspectives and practices: comparing novice and experienced instructors of English for academic purposes in Bangladesh","year":2025,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba; Red River College","funders":"","keywords":"Corrective feedback; Professional development; English for academic purposes; Grammar; Faculty development; Relevance (law); Focus group; Qualitative research; Higher education","retraction":null,"screen_n_in":null,"score":{"opus":0.06050911231811046,"gpt":0.3330383500299867,"spread":0.2725292377118762,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004441626,0.0001268291,0.0002527679,0.0002395462,0.0001517895,0.00009495523,0.0001027142,0.00005701318,0.00001128497],"category_scores_gemma":[0.007142053,0.0001254863,0.00001676443,0.0001327321,0.0001819447,0.0002957906,0.00006818872,0.0004703496,1.516548e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004867193,"about_ca_system_score_gemma":0.00003288026,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001957641,"about_ca_topic_score_gemma":0.00153859,"domain_scores_codex":[0.999107,0.00009461372,0.0002542592,0.0002740081,0.0000740574,0.0001960462],"domain_scores_gemma":[0.9983083,0.001234022,0.0002426784,0.00007986353,0.0001105858,0.00002454755],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00003153592,0.0000230568,0.2210036,0.0001602658,0.00001098769,0.000003405008,0.7662357,0.00002726711,0.0006677266,0.005352811,0.00000766528,0.006475984],"study_design_scores_gemma":[0.0008005681,0.00006386975,0.06674029,0.001178633,0.00001671367,0.000003560749,0.9269313,0.003379063,0.0001753089,0.0001412198,0.0003982996,0.0001712162],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9757022,0.001950664,0.00001139327,0.00003730824,0.0001312834,0.0002357359,0.0000030465,0.00007004987,0.02185837],"genre_scores_gemma":[0.9983167,0.000009140178,0.001275464,0.00003765396,0.0001630937,0.00002987947,0.000002909116,0.00001263933,0.0001525056],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1606956,"threshold_uncertainty_score":0.8550221,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4407613621","doi":"10.1186/s40468-025-00341-4","title":"Correction: Lessons from the Chinese imperial examination system","year":2025,"lang":"en","type":"article","venue":"Language Testing in Asia","topic":"Legal Education and Practice Innovations","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Imperial examination; Linguistics; Psychology; Language education; Mathematics education; History; Philosophy; Ancient history","retraction":null,"screen_n_in":null,"score":{"opus":0.02787893634611101,"gpt":0.3862787152161436,"spread":0.3583997788700326,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0009194731,0.00005303391,0.00006303699,0.00005747834,0.0004553492,0.0001935383,0.0001607166,0.0000525225,0.0001115084],"category_scores_gemma":[0.009805205,0.00004109513,0.00001432347,0.001483834,0.00005170594,0.0002086711,0.0000189829,0.0002368111,0.00002629095],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001626809,"about_ca_system_score_gemma":0.00036551,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04688612,"about_ca_topic_score_gemma":0.01466422,"domain_scores_codex":[0.9991472,0.0003443221,0.0001502557,0.0001174597,0.0001298369,0.000110912],"domain_scores_gemma":[0.9974899,0.00218081,0.00008131265,0.0001375348,0.00009364911,0.00001677144],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00001663118,0.0001727676,0.2101386,0.00002506434,0.00002990049,0.00001767051,0.3773506,0.00007536494,0.001995421,0.1468868,0.02233243,0.2409588],"study_design_scores_gemma":[0.0002794149,0.000007134662,0.8164561,0.0001842922,0.00001725918,0.00000226604,0.162214,0.001471556,0.00005531991,0.0005464772,0.01864121,0.0001249638],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4997166,0.0001146974,0.0002146032,0.01642594,0.004508721,0.0001917295,0.000004821427,0.0001835642,0.4786393],"genre_scores_gemma":[0.9950226,0.00000114152,0.0004434508,0.0003025394,0.0006510383,0.00002839086,0.0000128314,0.000003710808,0.00353428],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6063176,"threshold_uncertainty_score":0.9985356,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}