{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":57,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":57,"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":"6e475cc8b99a","filters":{"venue":"Language Testing"}},"results":[{"id":"W2155682078","doi":"10.1191/0265532206lt337oa","title":"How assessing reading comprehension with multiple-choice questions shapes the construct: a cognitive processing perspective","year":2006,"lang":"en","type":"article","venue":"Language Testing","topic":"Reading and Literacy Development","field":"Psychology","cited_by":243,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa","funders":"","keywords":"Construct (python library); Reading comprehension; Psychology; Variety (cybernetics); Comprehension; Cognitive psychology; Cognition; Multiple choice; Perspective (graphical); Test (biology); Reading (process); Selection (genetic algorithm); Task (project management); Computer science; Linguistics; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.02646085463337262,"gpt":0.3216621472035859,"spread":0.2952012925702133,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002283761,0.0002434074,0.0002156308,0.0001108192,0.0007084723,0.000548546,0.0001376334,0.00006991483,0.0000238625],"category_scores_gemma":[0.00131618,0.0001635571,0.00003497804,0.0004912741,0.0002089865,0.0003127529,0.00004413012,0.0003785491,0.00001102192],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001057574,"about_ca_system_score_gemma":0.00007505581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003125566,"about_ca_topic_score_gemma":0.0002443612,"domain_scores_codex":[0.9984963,0.0001797195,0.000218231,0.0004625499,0.0002217411,0.0004214444],"domain_scores_gemma":[0.9963829,0.00279183,0.0002582659,0.0001552328,0.0003582245,0.00005355005],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007025608,0.0002364359,0.6751266,0.0000966199,0.0001497225,0.0006212133,0.1068639,0.00009739653,0.01403943,0.003832681,0.0001840682,0.1986817],"study_design_scores_gemma":[0.001487748,0.0001023093,0.8128787,0.00221474,0.0001468411,0.001063438,0.1757799,0.004787926,0.0003559259,0.0002948716,0.0003223631,0.0005651739],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9665036,0.00098653,0.004748041,0.0004993891,0.0001062971,0.0003523188,0.000009116025,0.0004097635,0.02638495],"genre_scores_gemma":[0.983562,1.685733e-7,0.01454108,0.0002099199,0.000345125,0.00005791464,0.00001782621,0.0000418845,0.001224067],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1981165,"threshold_uncertainty_score":0.6669666,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2134278381","doi":"10.1177/0265532207083743","title":"The key to success: English language testing in China","year":2008,"lang":"en","type":"article","venue":"Language Testing","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":242,"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":"Ministry of Education, India; Ministry of Earth Sciences","keywords":"Language assessment; China; Context (archaeology); Psychology; English language; Test (biology); Linguistics; Test of English as a Foreign Language; Language proficiency; Chinese language; Mathematics education; History","retraction":null,"screen_n_in":null,"score":{"opus":0.0648421376261999,"gpt":0.4101903691881555,"spread":0.3453482315619556,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0009655377,0.0001110473,0.0001188323,0.0000888387,0.0009532226,0.0001115393,0.0003842894,0.00005532264,0.00007885817],"category_scores_gemma":[0.03555979,0.00009147055,0.00002704573,0.001128732,0.0001181554,0.0001222962,0.00006277621,0.000233851,0.00005939662],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009257929,"about_ca_system_score_gemma":0.0002977969,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0610561,"about_ca_topic_score_gemma":0.01353228,"domain_scores_codex":[0.9986054,0.0001898932,0.0002345989,0.0002051478,0.0002678884,0.0004971278],"domain_scores_gemma":[0.9978673,0.001551022,0.00008655296,0.0002198012,0.000106766,0.0001684935],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000002479984,0.00002764066,0.08376308,0.000006508226,0.000001947626,0.00004572174,0.8534475,0.00003453374,0.0004868277,0.0004217965,0.0002865836,0.06147538],"study_design_scores_gemma":[0.0008019304,0.00008541813,0.4806236,0.0003289129,0.00001345924,0.00002905023,0.4647095,0.001137606,0.0009347451,0.000149304,0.05025275,0.0009337381],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8010755,0.0002265084,0.000002933243,0.0005168193,0.0002680528,0.0001837448,0.000002374435,0.0002081385,0.1975159],"genre_scores_gemma":[0.9903392,0.000005412271,0.003070153,0.0005809875,0.00137789,0.00002179936,0.000002244072,0.00001677681,0.004585589],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3968606,"threshold_uncertainty_score":0.9725641,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2042225150","doi":"10.1191/0265532204lt273oa","title":"Evaluation of an in-depth vocabulary knowledge measure for assessing reading performance","year":2004,"lang":"en","type":"article","venue":"Language Testing","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":237,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Test of English as a Foreign Language; Vocabulary; Reading comprehension; Reading (process); Context (archaeology); Measure (data warehouse); Test (biology); Psychology; Language proficiency; Sample (material); Vocabulary development; Mathematics education; Language assessment; Computer science; Linguistics; Teaching method","retraction":null,"screen_n_in":null,"score":{"opus":0.09731813955004215,"gpt":0.406059529447716,"spread":0.3087413898976738,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003130488,0.0001159232,0.0001701122,0.0001833873,0.0000918697,0.0000286197,0.0001204321,0.00008907282,0.000684517],"category_scores_gemma":[0.001180775,0.0001191045,0.00003596853,0.0003366669,0.00002430502,0.0002627604,0.00001592381,0.0001812512,0.00001850328],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001189025,"about_ca_system_score_gemma":0.0001165607,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001404246,"about_ca_topic_score_gemma":0.00006481753,"domain_scores_codex":[0.9987161,0.000245617,0.0002819153,0.0002590074,0.0002365545,0.0002608483],"domain_scores_gemma":[0.9990966,0.0001957119,0.0001609505,0.0002437646,0.0002571928,0.00004576704],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0000292533,0.0002303176,0.01357269,0.0001056226,0.0000321651,0.00003624922,0.0734766,0.003579995,0.04331007,0.000230644,0.000007649212,0.8653888],"study_design_scores_gemma":[0.01145313,0.0005833753,0.7942628,0.001905661,0.0003484047,0.0002122004,0.1017742,0.07069571,0.01734526,0.0004423854,0.00005584531,0.0009210572],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.961946,0.003043764,0.0008335939,0.0000104177,0.0001758834,0.0002867959,0.000001583915,0.0000956072,0.03360632],"genre_scores_gemma":[0.9922594,1.295822e-7,0.007199365,0.0001514823,0.0002222943,0.00005180746,0.00002797069,0.00003035981,0.00005718436],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8644677,"threshold_uncertainty_score":0.7494982,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2129988987","doi":"10.1177/0265532208092433","title":"Test review: College English Test (CET) in China","year":2008,"lang":"en","type":"article","venue":"Language Testing","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":192,"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":"Test (biology); College English; China; Psychology; Language assessment; Test of English as a Foreign Language; Mathematics education; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.01727451502830387,"gpt":0.2757766721564041,"spread":0.2585021571281003,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002831015,0.0001062119,0.0001488902,0.00008491719,0.000125452,0.0000126508,0.0005765132,0.00004747985,0.00002264514],"category_scores_gemma":[0.005635848,0.0001009612,0.00002291014,0.0009279926,0.00004414333,0.0002050505,0.0001571483,0.0002469618,0.00003266115],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005216243,"about_ca_system_score_gemma":0.0001308074,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005520729,"about_ca_topic_score_gemma":0.00001649271,"domain_scores_codex":[0.9991082,0.00002652666,0.0002159245,0.0002697317,0.0001467905,0.0002328238],"domain_scores_gemma":[0.9985433,0.0008735562,0.00008397402,0.0003911617,0.0000654395,0.00004255298],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[4.736991e-7,0.0007546929,0.9472071,0.0002532321,0.000006408845,0.001056227,0.004956536,0.00001436461,0.002156925,0.009492432,0.01671873,0.01738289],"study_design_scores_gemma":[0.0007210357,0.0002374557,0.9806531,0.002421604,0.00001057539,0.0007350305,0.0006214603,0.005557359,0.002512983,0.001761639,0.004042175,0.0007256031],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8762404,0.02024066,0.002485356,0.008441632,0.0007210257,0.0009573118,0.00002954325,0.002326088,0.08855796],"genre_scores_gemma":[0.9365644,0.0001212005,0.0617414,0.0008107263,0.0001173232,0.0000390768,0.000003876175,0.000007766623,0.0005942316],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08796372,"threshold_uncertainty_score":0.6747043,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2159609851","doi":"10.1191/0265532204lt288oa","title":"ESL/EFL instructors’ classroom assessment practices: purposes, methods, and procedures","year":2004,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":184,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Alberta; Queen's University","funders":"","keywords":"Psychology; English as a foreign language; Mathematics education; Tertiary level; Pedagogy; Second language; Language assessment; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.06401151606627827,"gpt":0.4566933338742858,"spread":0.3926818178080075,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001259492,0.0001602916,0.0001926141,0.00007346245,0.0006034473,0.0003334264,0.0002522462,0.00009594835,0.00007815419],"category_scores_gemma":[0.004276884,0.000147588,0.00003457235,0.0003582949,0.0001592115,0.0004974992,0.000119671,0.000250586,0.000009544773],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001551925,"about_ca_system_score_gemma":0.0004715754,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005643806,"about_ca_topic_score_gemma":0.002757241,"domain_scores_codex":[0.9983709,0.0002323285,0.000230163,0.0003362223,0.0004226617,0.0004077303],"domain_scores_gemma":[0.9984898,0.0007326702,0.0003897482,0.0001591591,0.00008756423,0.0001410448],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001790745,0.0003189038,0.571827,0.0002207024,0.0001452855,0.0001653321,0.1659635,0.00002255949,0.03583076,0.04561178,0.0003762064,0.1795001],"study_design_scores_gemma":[0.004627989,0.0005317705,0.5233849,0.0008702152,0.0004278018,0.00009897222,0.417603,0.0002602982,0.002466767,0.02205466,0.025515,0.002158609],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8899671,0.001086575,0.0009146828,0.001141978,0.0003414859,0.0005238556,0.000003373891,0.0003853021,0.1056356],"genre_scores_gemma":[0.6683367,0.00004521573,0.3304719,0.0002061011,0.0004152005,0.00002397043,0.000004750965,0.00001726388,0.0004788553],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3295572,"threshold_uncertainty_score":0.8531784,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2135663628","doi":"10.1177/0265532208097336","title":"Cognitive diagnostic assessment of L2 reading comprehension ability: Validity arguments for Fusion Model application to <i>LanguEdge</i> assessment","year":2008,"lang":"en","type":"article","venue":"Language Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":175,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Reading comprehension; Psychology; Cognition; Test (biology); Profiling (computer programming); Dependability; Comprehension; Reading (process); Cognitive psychology; Computer science; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.4824119830634888,"gpt":0.5159648486704076,"spread":0.03355286560691884,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005607481,0.000233323,0.000566214,0.0003597184,0.0004046212,0.00006082846,0.000509802,0.00009164836,0.00002018725],"category_scores_gemma":[0.1339034,0.0001909338,0.0001271029,0.001691146,0.00007605331,0.0001761379,0.0003434014,0.0001987398,0.000007722714],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000126566,"about_ca_system_score_gemma":0.0001264369,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00012942,"about_ca_topic_score_gemma":0.000006228173,"domain_scores_codex":[0.9961553,0.0004159203,0.001004955,0.0008236492,0.001181548,0.0004186427],"domain_scores_gemma":[0.9124872,0.0851215,0.000710158,0.0006375588,0.0008672672,0.0001762872],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003764888,0.0003782486,0.6663653,0.00008788145,0.00002349876,0.0000170198,0.002140289,0.01127237,0.1008787,0.00018096,0.0006185359,0.2179995],"study_design_scores_gemma":[0.0009548515,0.0003929524,0.4863092,0.000242554,0.00005246457,0.00002515564,0.002254783,0.501873,0.003340098,0.00416316,0.00004141528,0.0003504124],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5328215,0.00002596287,0.4643598,0.00004385342,0.0001190898,0.0007316874,0.00004269867,0.00007494368,0.001780404],"genre_scores_gemma":[0.6565592,0.000003071366,0.342977,0.0001693233,0.0000724407,0.0001264413,0.0000228743,0.00001553337,0.00005407738],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4906006,"threshold_uncertainty_score":0.873392,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2156770411","doi":"10.1177/0265532209104666","title":"Interacting in pairs in a test of oral proficiency: Co-constructing a better performance","year":2009,"lang":"en","type":"article","venue":"Language Testing","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":163,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Test (biology); Psychology; Meaning (existential); Context (archaeology); Negotiation; Social psychology; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.06423787690980926,"gpt":0.4414124951436066,"spread":0.3771746182337974,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008745365,0.00007059228,0.0001284538,0.0001686429,0.00009090579,0.00002749879,0.0001334796,0.00004145074,0.00005734649],"category_scores_gemma":[0.006404143,0.00007187673,0.00001720052,0.0006446068,0.00007185568,0.0001782217,0.00001146488,0.0002035043,0.000005524333],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007201653,"about_ca_system_score_gemma":0.0001842248,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007276883,"about_ca_topic_score_gemma":0.001565743,"domain_scores_codex":[0.9990702,0.00007881753,0.0002915865,0.000133915,0.000136962,0.0002885158],"domain_scores_gemma":[0.9990865,0.0006114139,0.0001418829,0.00008159674,0.00003845003,0.00004018604],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000001628253,0.00004121177,0.6462628,0.00001222895,2.976236e-7,0.000004617653,0.1330892,0.0000109161,0.002875242,0.00005682594,0.000002569546,0.2176425],"study_design_scores_gemma":[0.0007903318,0.0001270993,0.7268726,0.001049932,0.000005274872,0.00001463379,0.2554966,0.008204082,0.006666456,0.0001019732,0.0002691103,0.0004019037],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9365197,0.00002648729,0.000001658372,0.0003017929,0.00005918647,0.0001288944,7.818629e-7,0.00004161773,0.06291988],"genre_scores_gemma":[0.9940374,8.091318e-7,0.005483535,0.0002789349,0.00009623174,0.000003649228,0.000001073518,0.00000451924,0.00009381247],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2172406,"threshold_uncertainty_score":0.9993337,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2123012719","doi":"10.1177/026553220101800302","title":"Examining dialogue: another approach to content specification and to validating inferences drawn from test scores","year":2001,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":162,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Construct (python library); Psychology; Test (biology); Sociocultural evolution; Point (geometry); Cognition; Cognitive psychology; Content (measure theory); Mathematics education; Inference; Construct validity; Linguistics; Natural language processing; Social psychology; Computer science; Artificial intelligence; Psychometrics; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.2440808435611483,"gpt":0.2751833738551075,"spread":0.03110253029395921,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000356283,0.0001473481,0.0001675434,0.00009360186,0.0005268126,0.000680113,0.0001330382,0.00002768943,0.0001709287],"category_scores_gemma":[0.002197814,0.0001272882,0.00001556453,0.00007850124,0.00005963792,0.0001397787,0.00007019342,0.0001813371,0.00005379833],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001953491,"about_ca_system_score_gemma":0.00001280019,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006368421,"about_ca_topic_score_gemma":0.0006014605,"domain_scores_codex":[0.9990053,0.00007466658,0.0002211531,0.0003156986,0.0001430903,0.0002401453],"domain_scores_gemma":[0.998731,0.000869108,0.00009656652,0.0001709535,0.00005267454,0.00007975198],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000009034814,0.00008175262,0.3287088,0.00002462453,0.00001838361,0.00001513893,0.5310782,0.0001536866,0.0258181,0.002228043,0.00021681,0.1116474],"study_design_scores_gemma":[0.001411497,0.0009103002,0.4466537,0.001995737,0.0001279567,0.00004309851,0.5033068,0.007240548,0.002041281,0.0005905167,0.03346477,0.002213776],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9266044,0.0001622523,0.001301419,0.00009141493,0.00008273293,0.0001555744,0.00001115817,0.0002283933,0.07136269],"genre_scores_gemma":[0.9839895,8.609571e-7,0.01264573,0.0004961099,0.0009940148,0.00002227285,0.00003100993,0.00002753582,0.001792926],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.117945,"threshold_uncertainty_score":0.9627189,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2105451781","doi":"10.1191/0265532204lt287oa","title":"Teacher formative assessment and talk in classroom contexts: assessment as discourse and assessment of discourse","year":2004,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":159,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Formative assessment; Psychology; Pedagogy; Discourse analysis; Applied linguistics; Mathematics education; Assessment for learning; Systemic functional linguistics; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.03032002706073061,"gpt":0.4286908790492009,"spread":0.3983708519884703,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001395828,0.0002527186,0.0004196861,0.000153969,0.0003267148,0.0001746399,0.0002231136,0.000107439,0.0001004092],"category_scores_gemma":[0.0002220149,0.0002263329,0.00005128387,0.000374452,0.000480266,0.0006579131,0.0002028843,0.0004338013,0.000001621174],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004164316,"about_ca_system_score_gemma":0.0005470617,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007168258,"about_ca_topic_score_gemma":0.003914115,"domain_scores_codex":[0.9977274,0.0002386689,0.0004745305,0.0003772012,0.0006771764,0.0005050161],"domain_scores_gemma":[0.9987569,0.0004686003,0.0003281348,0.0001996349,0.00008004533,0.0001666697],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000009927094,0.0005870681,0.8351541,0.00007148167,0.00007616492,0.000109113,0.09371276,0.00004194808,0.002073702,0.03619533,0.00002822531,0.03194013],"study_design_scores_gemma":[0.002046388,0.0002367015,0.6623685,0.0003387458,0.00006070044,0.000006857245,0.332113,0.0005101711,0.00003959758,0.001887427,0.0000566251,0.0003353139],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9221969,0.0003018323,0.000573084,0.001138191,0.0001137976,0.0005960678,0.00001091914,0.00006635585,0.07500283],"genre_scores_gemma":[0.9810621,0.00006002154,0.01820872,0.000108062,0.0001288971,0.00005495936,0.0000127587,0.00002223536,0.0003422976],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2384002,"threshold_uncertainty_score":0.9994431,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2082767250","doi":"10.1177/0265532208101010","title":"An investigation into native and non-native teachers' judgments of oral English performance: A mixed methods approach","year":2009,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":137,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology; Pronunciation; Rasch model; Consistency (knowledge bases); Grammar; Mathematics education; First language; Internal consistency; Multimethodology; Linguistics; Psychometrics; Developmental psychology; Computer science; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.04154220340015909,"gpt":0.3200407438186342,"spread":0.2784985404184751,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001155387,0.0001420475,0.0001992127,0.0001099408,0.0003225747,0.0000983085,0.0001067728,0.00004438387,0.00002396408],"category_scores_gemma":[0.0007698403,0.0001239099,0.00002288274,0.00007973056,0.000142393,0.0004091733,0.0000250243,0.0003415127,7.06702e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002950846,"about_ca_system_score_gemma":0.00002206755,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007746941,"about_ca_topic_score_gemma":0.00003919139,"domain_scores_codex":[0.9989463,0.0003177896,0.0002082213,0.0002180045,0.0001390702,0.0001706197],"domain_scores_gemma":[0.9993135,0.0001982529,0.0001766146,0.0001218346,0.0001271684,0.00006257673],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000005036533,0.00004507543,0.004986664,0.00004149772,0.00001603269,0.000001039187,0.8422319,0.00007775443,0.003015193,0.0007998537,0.00002082888,0.1487592],"study_design_scores_gemma":[0.001566952,0.00156339,0.0506083,0.0005329209,0.0001354867,0.000006502383,0.8271337,0.106333,0.009431276,0.001483698,0.0003798519,0.0008249379],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9757374,0.000114781,0.0005273069,0.00001392735,0.00007283677,0.0001209268,0.000001913189,0.0001330943,0.02327783],"genre_scores_gemma":[0.9013217,6.711829e-7,0.09794325,0.00009590042,0.0002086236,0.000006673522,0.0000247728,0.00001338915,0.00038504],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1479342,"threshold_uncertainty_score":0.50529,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2116070328","doi":"10.1177/0265532210376379","title":"Think-aloud protocols in research on essay rating: An empirical study of their veridicality and reactivity","year":2010,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":121,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Think aloud protocol; Psychology; Protocol analysis; Perception; Empirical research; Sample (material); Qualitative research; Rating scale; Social psychology; Nomothetic and idiographic; Cognitive psychology; Applied psychology; Developmental psychology; Epistemology; Cognitive science","retraction":null,"screen_n_in":null,"score":{"opus":0.2923764194054679,"gpt":0.5510859877479262,"spread":0.2587095683424582,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006440684,0.000083672,0.0001775743,0.00007915629,0.0003393473,0.00009831074,0.0002602454,0.00008569542,0.00003783099],"category_scores_gemma":[0.003381105,0.00006686147,0.00001369408,0.0005312032,0.0002182454,0.00016641,0.0001173351,0.000689057,0.000001792791],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003987847,"about_ca_system_score_gemma":0.0001467037,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006873621,"about_ca_topic_score_gemma":0.0138442,"domain_scores_codex":[0.9975994,0.001020123,0.0001965475,0.0002779067,0.000615862,0.0002901692],"domain_scores_gemma":[0.9976435,0.001847386,0.00008194242,0.0002457829,0.00009258107,0.00008882382],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002434618,0.0009123872,0.6528844,0.000009748273,0.000002838422,0.0000147292,0.3272898,3.739633e-7,0.01173417,0.0001797215,0.000008902451,0.006938535],"study_design_scores_gemma":[0.0005041019,0.0004316432,0.7267394,0.00004266435,0.000001945941,3.425733e-7,0.2713432,0.00013408,0.0003647935,0.0003241998,0.00002581715,0.00008783205],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9803934,0.000002900598,7.588783e-7,0.000104596,0.00003270371,0.004111891,0.000001182957,0.00005459703,0.01529796],"genre_scores_gemma":[0.9991723,1.434238e-7,0.0003921033,0.0000196967,0.0001780559,0.0001737892,6.945806e-7,0.00000772996,0.00005546028],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07385501,"threshold_uncertainty_score":0.9997397,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2136852860","doi":"10.1177/026553220101800303","title":"Native- and nonnative-speaking EFL teachers’ evaluation of Chinese students’ English writing","year":2001,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":117,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology; Multivariate analysis of variance; First language; Foreign language; English as a foreign language; Point (geometry); Language assessment; Language proficiency; Linguistics; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.06659189121337757,"gpt":0.3396810487605434,"spread":0.2730891575471659,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002475633,0.0001469607,0.0001873518,0.0001134496,0.0003803516,0.0001737718,0.0001358549,0.00003119895,0.000311388],"category_scores_gemma":[0.004877461,0.0001271442,0.00003481481,0.00007338329,0.00009510405,0.0002701373,0.00007807865,0.0003178951,0.000002910089],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004196446,"about_ca_system_score_gemma":0.00001839471,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001026245,"about_ca_topic_score_gemma":0.0004789045,"domain_scores_codex":[0.998524,0.0002930881,0.0002647721,0.000215009,0.0004980306,0.000205087],"domain_scores_gemma":[0.9986761,0.0005903076,0.0002197775,0.0001215885,0.0003531381,0.0000390443],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000003157266,0.00004313478,0.2310523,0.0000239027,0.00002744929,0.000008723066,0.6878537,0.00008320155,0.0005194504,0.0004170695,0.000007268678,0.07996068],"study_design_scores_gemma":[0.003473714,0.0002927558,0.3036566,0.001395809,0.0002524569,0.00003606518,0.657928,0.02806009,0.0001336779,0.000828742,0.002993396,0.0009486682],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8948968,0.0007365703,0.00002344043,0.00001476823,0.0001012973,0.0001327896,0.000004095264,0.0001519426,0.1039383],"genre_scores_gemma":[0.9979767,0.000001883104,0.0005953821,0.00004456181,0.0008565212,0.000009658792,0.0000118359,0.00002481555,0.0004786291],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1034597,"threshold_uncertainty_score":0.5839129,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2122362059","doi":"10.1191/0265532206lt322oa","title":"Aiming for positive washback: a case study of international teaching assistants","year":2005,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":102,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval","funders":"","keywords":"Test (biology); Psychology; Language proficiency; Process (computing); Mathematics education; Empirical research; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.05418180584152639,"gpt":0.417552964658704,"spread":0.3633711588171776,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007105966,0.00006760054,0.0001042936,0.00005648265,0.0004002398,0.00006266831,0.000160414,0.0000264458,0.00003904515],"category_scores_gemma":[0.0009049757,0.00006699014,0.00003172235,0.000101326,0.00003463605,0.0001861593,0.00005213873,0.00009001935,0.000002092163],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008201926,"about_ca_system_score_gemma":0.00003300066,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005183866,"about_ca_topic_score_gemma":0.005065913,"domain_scores_codex":[0.9991629,0.00007986378,0.0001786317,0.0001430193,0.0002568784,0.0001787081],"domain_scores_gemma":[0.9990917,0.0005825684,0.0001277636,0.000068469,0.00009271935,0.00003674204],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001842531,0.0005926692,0.2616042,0.000008992158,0.00009007394,0.000386716,0.6251513,0.00003314335,0.002364049,0.0009155349,0.0001074889,0.1087273],"study_design_scores_gemma":[0.001206595,0.0001576592,0.0180886,0.00007474444,0.00005805515,0.00003765854,0.9781072,0.001685571,0.0001075276,0.00002420987,0.0002692359,0.0001829938],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9689544,0.00003273317,0.0002297826,0.0001360328,0.0001407786,0.0003280746,0.000006541906,0.00005797715,0.0301137],"genre_scores_gemma":[0.9845323,3.072115e-7,0.01427359,0.00004243792,0.0005142501,0.00001456001,0.00000259419,0.00000862757,0.0006113448],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3529558,"threshold_uncertainty_score":0.783649,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2109863812","doi":"10.1191/0265532204lt278oa","title":"A teacher-verification study of speaking and writing prototype tasks for a new TOEFL","year":2004,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":93,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Test of English as a Foreign Language; Active listening; Psychology; Formative assessment; Mathematics education; Test (biology); CLARITY; Reading (process); Second language writing; Language proficiency; Presentation (obstetrics); Language assessment; Pedagogy; Computer science; Second language; Linguistics; Communication","retraction":null,"screen_n_in":null,"score":{"opus":0.08046825576428888,"gpt":0.3114058708120012,"spread":0.2309376150477123,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003661521,0.00009192758,0.0001368742,0.0000523529,0.0003030353,0.0001014493,0.00006773145,0.0000200847,0.00003361943],"category_scores_gemma":[0.0006213356,0.00008420099,0.00002038952,0.00003605224,0.00002575096,0.0001085483,0.00002259408,0.0001523259,0.000001878259],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001741643,"about_ca_system_score_gemma":0.00002548229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005348182,"about_ca_topic_score_gemma":0.001010677,"domain_scores_codex":[0.9993507,0.0000351501,0.000204487,0.0001713831,0.00008817248,0.0001501598],"domain_scores_gemma":[0.9994986,0.0001594528,0.0001471806,0.0001163153,0.00004610915,0.00003231229],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002501725,0.000193915,0.008487372,0.0001722214,0.00002646607,0.00000701509,0.8562801,0.0001795212,0.003525621,0.004053048,0.00001679919,0.1270329],"study_design_scores_gemma":[0.005402894,0.003165028,0.01077141,0.001543824,0.0001963589,0.00003030716,0.9680936,0.002798713,0.0006503473,0.001676873,0.004861932,0.0008086776],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9929972,0.0001994275,0.0004205132,0.00004932929,0.00003359448,0.000639434,0.000002000766,0.0001431058,0.005515357],"genre_scores_gemma":[0.9936404,1.625203e-7,0.005305791,0.00003040754,0.0005671749,0.00003456566,0.000005429826,0.00002172561,0.0003943319],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1262242,"threshold_uncertainty_score":0.8084887,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2078562183","doi":"10.1177/0265532210384253","title":"Impact and consequences of school-based assessment (SBA): Students’ and parents’ views of SBA in Hong Kong","year":2011,"lang":"en","type":"article","venue":"Language Testing","topic":"Parental Involvement in Education","field":"Social Sciences","cited_by":92,"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":"Perception; Psychology; Competence (human resources); Certificate; Context (archaeology); Medical education; Developmental psychology; Mathematics education; Pedagogy; Social psychology; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.2119144028794248,"gpt":0.4574731512778527,"spread":0.2455587483984279,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008858168,0.00008307703,0.000169077,0.00009191225,0.00007375069,0.00002735764,0.000150262,0.00003621225,0.00007156334],"category_scores_gemma":[0.0006197664,0.00007525065,0.00001902342,0.0002532114,0.0002923575,0.0001717167,0.00004503408,0.00007789077,6.09541e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008274943,"about_ca_system_score_gemma":0.0001734601,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03319608,"about_ca_topic_score_gemma":0.004081465,"domain_scores_codex":[0.9989656,0.0001767526,0.0002719577,0.0001496084,0.0002617351,0.0001744069],"domain_scores_gemma":[0.9992792,0.0002501778,0.000239074,0.00009998431,0.00005103577,0.00008052085],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000003724774,0.0001567186,0.9646982,0.00004062565,0.000007618654,0.000003175888,0.02109743,0.000002179396,0.01262138,0.0000552694,0.000003792563,0.001309866],"study_design_scores_gemma":[0.0002301558,0.00007428443,0.9656995,0.0002343087,0.00001229038,4.270289e-7,0.03093889,0.00003594098,0.002516512,0.0001809223,0.000001422122,0.00007535989],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9955471,0.0006171659,0.00001308305,0.00001605929,0.0000681735,0.0002335523,0.000007498069,0.00001749544,0.003479867],"genre_scores_gemma":[0.9963095,0.00002553355,0.003579307,0.00002410074,0.00002589088,0.00001077952,0.000004747596,0.000005127164,0.00001504858],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02911461,"threshold_uncertainty_score":0.9732419,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2096343999","doi":"10.1191/0265532203lt248oa","title":"Does item-level DIF manifest itself in scale-level analyses? Implications for translating language tests","year":2003,"lang":"en","type":"article","venue":"Language Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":92,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Differential item functioning; Equivalence (formal languages); Psychology; Scale (ratio); Measurement invariance; Item response theory; Item analysis; Statistics; Test (biology); Psychometrics; Developmental psychology; Linguistics; Structural equation modeling; Mathematics; Confirmatory factor analysis","retraction":null,"screen_n_in":null,"score":{"opus":0.6773012489253456,"gpt":0.5390986227521983,"spread":0.1382026261731473,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006747382,0.0003296185,0.0006064455,0.0008460847,0.0003819273,0.0003147263,0.0008995,0.0001509855,0.0001260107],"category_scores_gemma":[0.2575959,0.0002099897,0.0001911787,0.004788593,0.00007689429,0.0003262023,0.00008801077,0.0003267178,0.0000220619],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006955244,"about_ca_system_score_gemma":0.0000954463,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004920563,"about_ca_topic_score_gemma":0.001119523,"domain_scores_codex":[0.9958195,0.0005375626,0.001269155,0.0009708789,0.0005923416,0.0008105782],"domain_scores_gemma":[0.9392038,0.05879252,0.0005797765,0.0009333737,0.0003084515,0.0001821001],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000007264762,0.00009430339,0.6871116,0.00003705266,0.00001642013,0.00002815883,0.004096707,0.0001968023,0.05214192,0.0004318964,0.0001117143,0.2557262],"study_design_scores_gemma":[0.001122805,0.00007757432,0.9640663,0.0001531694,0.00004474372,0.00008866646,0.01552727,0.003763278,0.002878072,0.01148491,0.0001887571,0.0006045162],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9085241,0.000791975,0.07255301,0.0005000946,0.0003716465,0.0006680805,0.0002236013,0.0002721901,0.0160953],"genre_scores_gemma":[0.6726789,0.000001488572,0.3261494,0.0001705375,0.0001277178,0.00006777591,0.00001047682,0.00003100654,0.0007626727],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2769547,"threshold_uncertainty_score":0.8563132,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2776964473","doi":"10.1177/0265532217716732","title":"The development of EFL examinations in Haiti: Collaboration and language assessment literacy development","year":2017,"lang":"en","type":"article","venue":"Language Testing","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":87,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; University of Ottawa","funders":"","keywords":"Christian ministry; Psychology; Literacy; Medical education; Professional development; English language; Pedagogy; Faculty development; Mathematics education; Language assessment; Language development; Political science; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.07883816509620944,"gpt":0.5035994185271453,"spread":0.4247612534309358,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.001114551,0.00006011703,0.00007460784,0.00005408183,0.001381724,0.0002659363,0.0001786356,0.00003110625,0.00001910078],"category_scores_gemma":[0.002803244,0.00005022089,0.000006474209,0.0001330138,0.00009920088,0.0001743359,0.00005289077,0.00007166943,0.000002822802],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001037394,"about_ca_system_score_gemma":0.000839283,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002074819,"about_ca_topic_score_gemma":0.02013881,"domain_scores_codex":[0.9992169,0.00009095631,0.0002437016,0.0001067428,0.0001802256,0.0001615337],"domain_scores_gemma":[0.9990162,0.000456596,0.0002014978,0.0001521098,0.0001239623,0.00004969623],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[7.507459e-7,0.000023737,0.0284036,0.00001234199,0.000002892629,0.000001364498,0.5091015,8.358121e-7,0.0009286687,0.001300756,0.000008670154,0.4602149],"study_design_scores_gemma":[0.0002613286,0.000006579311,0.8431256,0.0001261786,0.000003453973,4.801729e-7,0.1437469,0.0002833077,0.002099712,0.00004665881,0.01016879,0.0001310434],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9579504,0.0001555847,0.00004216155,0.0004230856,0.0001020322,0.0001753878,0.000001522817,0.00002208441,0.0411278],"genre_scores_gemma":[0.964879,0.000004001315,0.03330017,0.00003844248,0.00007508521,0.00002715913,0.000005382847,0.000004767629,0.001666006],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.814722,"threshold_uncertainty_score":0.9999183,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2104209819","doi":"10.1177/0265532212436659","title":"Topical knowledge and ESL writing","year":2012,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":77,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of British Columbia","funders":"","keywords":"Impromptu; Cohesion (chemistry); Language proficiency; Psychology; Test (biology); Mathematics education; Language assessment; Second language writing; Test of English as a Foreign Language; English for academic purposes; Task (project management); Pedagogy; Second language; Linguistics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.06154722270013897,"gpt":0.2921705855996226,"spread":0.2306233628994836,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003336346,0.00007742867,0.0000874784,0.00003109875,0.0003456884,0.0001023494,0.00004706771,0.00002214154,0.0003762667],"category_scores_gemma":[0.000451741,0.00006679497,0.00001790385,0.00001890478,0.0000506766,0.0001463821,0.00004903193,0.0002191526,0.00006799751],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000009313579,"about_ca_system_score_gemma":0.000005039362,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001990142,"about_ca_topic_score_gemma":0.00004590266,"domain_scores_codex":[0.9994435,0.00005457129,0.0001072321,0.00009262245,0.0000519409,0.0002500801],"domain_scores_gemma":[0.9994588,0.0003470677,0.00003433146,0.00007512679,0.00001832852,0.00006630452],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000001900768,0.00006159063,0.1143796,0.0001036248,0.00001331639,0.00001248446,0.5248106,0.000001055848,0.0009571558,0.07670829,0.0005474025,0.282403],"study_design_scores_gemma":[0.001668624,0.0003294702,0.09240823,0.001279063,0.0001700916,0.0002934793,0.2695688,0.005051067,0.0005936438,0.0008662308,0.6257617,0.002009529],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7490291,0.002679795,0.00002012471,0.00004477178,0.0001423373,0.00002650454,7.261738e-7,0.0001887864,0.2478679],"genre_scores_gemma":[0.9911574,6.853463e-7,0.001380415,0.0001336435,0.002457032,0.00000280753,0.000002507026,0.00001444235,0.004851068],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6252143,"threshold_uncertainty_score":0.4119857,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2005408543","doi":"10.1191/0265532204lt292oa","title":"Test decisions over time: tracking validity","year":2004,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":69,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Test (biology); Psychology; Sample (material); Active listening; Test validity; Applied psychology; Social psychology; Mathematics education; Psychometrics; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.1004204092065655,"gpt":0.2939022965666889,"spread":0.1934818873601234,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0004003863,0.0001371782,0.000142537,0.00006679029,0.0006815977,0.0002795259,0.0001496499,0.00003576185,0.001591122],"category_scores_gemma":[0.00457736,0.0001198853,0.00006053765,0.00006109421,0.00006603329,0.0001996025,0.00004755471,0.0003634228,0.0005117817],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004464213,"about_ca_system_score_gemma":0.0000301434,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001115818,"about_ca_topic_score_gemma":0.0002256885,"domain_scores_codex":[0.9990923,0.00004395917,0.0001912657,0.000207482,0.0001821669,0.000282794],"domain_scores_gemma":[0.9982728,0.001321398,0.00008957538,0.0001997668,0.00005006103,0.00006644046],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001355728,0.0006795053,0.03277876,0.0000868864,0.00006419853,0.000793916,0.6610945,0.002074301,0.03906832,0.02491748,0.003123549,0.235305],"study_design_scores_gemma":[0.01943282,0.00356865,0.2340919,0.01532846,0.001250235,0.001230249,0.2057081,0.03342386,0.0125521,0.04692489,0.4131491,0.01333958],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8992832,0.000153095,0.000195114,0.00009807546,0.000161516,0.00006820215,0.00001140488,0.0005819982,0.09944735],"genre_scores_gemma":[0.9915507,5.549055e-7,0.002472898,0.0002869051,0.0008213656,0.000003526356,0.00001312138,0.00003343383,0.004817429],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4553863,"threshold_uncertainty_score":0.9993216,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2145055698","doi":"10.1177/0265532210368717","title":"Explaining ESL essay holistic scores: A multilevel modeling approach","year":2010,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":64,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"Educational Testing Service","keywords":"Psychology; Argumentation theory; Context (archaeology); Multilevel model; Set (abstract data type); Multilevel modelling; Social psychology; Epistemology; Statistics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.1195594706556013,"gpt":0.2960164806540859,"spread":0.1764570099984846,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006918146,0.0002151494,0.0002137296,0.0001198984,0.0008408013,0.0003802165,0.0002492951,0.00006621973,0.0003535916],"category_scores_gemma":[0.002272333,0.0001906683,0.00006203242,0.00004895132,0.00009599784,0.0002241068,0.00008060793,0.001038103,0.00006712238],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002022778,"about_ca_system_score_gemma":0.00004769389,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00190087,"about_ca_topic_score_gemma":0.0002490589,"domain_scores_codex":[0.9986669,0.00007160731,0.0002718268,0.0003442783,0.00021134,0.0004340002],"domain_scores_gemma":[0.9989648,0.000447406,0.0001148494,0.0002964509,0.00007893623,0.00009749181],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001272887,0.0001495046,0.002585647,0.0002176799,0.00004208168,0.00009398563,0.8471537,0.01300773,0.008911178,0.05369629,0.0001106889,0.07401884],"study_design_scores_gemma":[0.0003722018,0.00002909834,0.00006990362,0.0001830618,0.00003059929,0.00003521718,0.04492009,0.9527322,0.00003440423,0.0003272926,0.0008539393,0.0004119936],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8514761,0.000141738,0.008339768,0.00001759399,0.0004266923,0.0001099715,0.000007719938,0.0006814672,0.1387989],"genre_scores_gemma":[0.9679496,2.161449e-7,0.02734209,0.0001550435,0.001756844,0.00002280277,0.00002715376,0.0000608441,0.00268538],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9397244,"threshold_uncertainty_score":0.7775228,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2100317885","doi":"10.1177/0265532215570924","title":"How do young students with different profiles of reading skill mastery, perceived ability, and goal orientation respond to holistic diagnostic feedback?","year":2015,"lang":"en","type":"article","venue":"Language Testing","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":61,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology; Reading (process); Goal orientation; Cognition; Developmental psychology; Orientation (vector space); Mathematics education; Cognitive psychology; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.06663303109817183,"gpt":0.3901718268217531,"spread":0.3235387957235813,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003248621,0.0001691118,0.0002156316,0.0000863722,0.00006201849,0.0001050472,0.0001778473,0.00005525085,0.00007128862],"category_scores_gemma":[0.001692213,0.000120983,0.00001770329,0.0002345885,0.00007103882,0.00008921753,0.00009234608,0.000128885,0.00001317494],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005710201,"about_ca_system_score_gemma":0.00001847985,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002013405,"about_ca_topic_score_gemma":0.00002845465,"domain_scores_codex":[0.9985784,0.0001807359,0.000218006,0.00040296,0.0003492672,0.0002706499],"domain_scores_gemma":[0.9981331,0.001153575,0.0001382516,0.0002362126,0.0001525657,0.0001863187],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001257889,0.0003171522,0.9595342,0.0000372531,0.0000286805,0.00003497314,0.03374704,0.000003293566,0.002302727,0.0001817832,0.00007500747,0.003612128],"study_design_scores_gemma":[0.0007882874,0.0005630148,0.9668079,0.0001710416,0.0000343192,0.00004307867,0.03107253,0.000004826328,0.00004511747,0.000314847,0.000006313504,0.0001487414],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9963343,0.00008910333,0.0002165802,0.0002015094,0.0003202758,0.0004706662,0.00002833617,0.00004549558,0.002293699],"genre_scores_gemma":[0.9956874,5.531966e-7,0.002630657,0.00009449535,0.0001487703,0.0001025248,0.0000314683,0.00001819553,0.001285938],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007273709,"threshold_uncertainty_score":0.4933544,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2611914815","doi":"10.1177/0265532217703433","title":"Developing a user-oriented second language comprehensibility scale for English-medium universities","year":2017,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":59,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Alberta; Concordia University","funders":"","keywords":"Operationalization; Formative assessment; Psychology; Construct (python library); English for academic purposes; Scale (ratio); Language proficiency; Point (geometry); Task (project management); Focus (optics); Rating scale; Mathematics education; Linguistics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.04907756014366304,"gpt":0.2904557913624791,"spread":0.241378231218816,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.0004504075,0.0001774303,0.0002318207,0.00006704454,0.002093313,0.0004315582,0.0003136562,0.00005273215,0.000496938],"category_scores_gemma":[0.001064982,0.0001705791,0.00007504365,0.00002058194,0.0002203053,0.0003892338,0.0001489197,0.0002964202,0.00001363224],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008813496,"about_ca_system_score_gemma":0.00009195081,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001557359,"about_ca_topic_score_gemma":0.003623045,"domain_scores_codex":[0.9989377,0.00006617502,0.0001976974,0.0002996495,0.0001361618,0.0003626014],"domain_scores_gemma":[0.9985062,0.0005547095,0.0002225109,0.0004407121,0.0002097371,0.00006608824],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00003727242,0.00004886645,0.01608638,0.0004466877,0.00005523515,0.00006038382,0.9575025,0.00001815544,0.001619005,0.01663235,0.001005734,0.006487359],"study_design_scores_gemma":[0.001810862,0.0001327712,0.01041566,0.000644811,0.00008275243,0.00001010248,0.8136135,0.002083429,0.001696033,0.0002203129,0.1684095,0.0008803142],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9378775,0.0001243117,0.0005802072,0.00009844609,0.0006295813,0.0001887052,0.00005883514,0.0004301362,0.06001224],"genre_scores_gemma":[0.970728,2.53415e-7,0.008520718,0.0001862512,0.001372422,0.000009180869,0.00004603298,0.00003279318,0.01910432],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1674038,"threshold_uncertainty_score":0.9992058,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4251941390","doi":"10.1177/0265532220929918","title":"Automated scoring of junior and senior high essays using Coh-Metrix features: Implications for large-scale language testing","year":2020,"lang":"en","type":"article","venue":"Language Testing","topic":"Software Engineering Research","field":"Computer Science","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre for Advancing Health Outcomes; University of Alberta","funders":"","keywords":"Natural language processing; Artificial intelligence; Rating scale; Computer science; Scale (ratio); Quality (philosophy); Construct (python library); Psychology; Computational linguistics; Disadvantaged; Machine learning; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.0384669575825602,"gpt":0.314934657209404,"spread":0.2764676996268438,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0005099743,0.0001834195,0.0002773508,0.0001726493,0.00020698,0.0001450411,0.0005523913,0.00007625024,0.000002406511],"category_scores_gemma":[0.01280623,0.0001876516,0.00003671898,0.001711649,0.00003180611,0.0002565515,0.0004246361,0.0002298309,0.000001665597],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006699291,"about_ca_system_score_gemma":0.00009463333,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002521761,"about_ca_topic_score_gemma":0.000007400031,"domain_scores_codex":[0.9984274,0.0000440952,0.0002874923,0.0004702182,0.0002178771,0.0005529008],"domain_scores_gemma":[0.9955913,0.00342109,0.0001453042,0.0004265477,0.0002128271,0.0002028593],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000109954,0.0000819273,0.113956,0.001104267,0.0000600877,0.00007170271,0.03106241,0.00501727,0.8071012,0.0009538756,0.0001414415,0.04043881],"study_design_scores_gemma":[0.001025519,0.0001706915,0.1006129,0.0003204192,0.00002718051,0.0001053082,0.001706215,0.8777049,0.01778612,0.00008529035,0.00001038726,0.0004450818],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7929544,0.0004436481,0.2043096,0.0001651695,0.00006152351,0.0003297323,0.00005130433,0.001643458,0.00004123371],"genre_scores_gemma":[0.6143281,3.807035e-7,0.3854439,0.0000633857,0.0000993271,0.00002344203,0.000005428649,0.00002625172,0.00000985091],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8726876,"threshold_uncertainty_score":0.9955093,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2145523583","doi":"10.1177/026553220101800206","title":"ESL/EFL instructors' practices for writing assessment: specific purposes or general purposes?","year":2001,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":57,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Interview; Psychology; Mathematics education; English for academic purposes; Pedagogy; Writing assessment; Language assessment; Process (computing); Sociology; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.1428657225810786,"gpt":0.3633483596418732,"spread":0.2204826370607945,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007561034,0.0002604644,0.0002806921,0.0001327026,0.001178777,0.0009452029,0.0002615569,0.00006246429,0.001399808],"category_scores_gemma":[0.001865258,0.000214575,0.00009042829,0.0000873961,0.00009756287,0.0006297161,0.00007227982,0.0004794273,0.00003341283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006109334,"about_ca_system_score_gemma":0.00006700716,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001358762,"about_ca_topic_score_gemma":0.000695334,"domain_scores_codex":[0.9982345,0.0001438124,0.0004192309,0.0004061192,0.0002513018,0.0005450682],"domain_scores_gemma":[0.9975294,0.001360317,0.0006192697,0.0002644674,0.0001330962,0.00009344234],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000144646,0.0003386282,0.04630851,0.0005085911,0.0001888515,0.0007174264,0.1527911,0.0002330297,0.01571698,0.07772937,0.003080449,0.7022424],"study_design_scores_gemma":[0.001878438,0.0006267045,0.003009985,0.0006728903,0.0001450655,0.0003998153,0.09129248,0.008085134,0.0003858677,0.0006337703,0.8916006,0.001269243],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8652254,0.0006813967,0.0002836392,0.0001914587,0.000535248,0.0003391727,0.00001621498,0.000580724,0.1321468],"genre_scores_gemma":[0.9288244,0.00001736844,0.04191982,0.0003419911,0.006073201,0.00006382285,0.00006567508,0.0000873225,0.02260636],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8885202,"threshold_uncertainty_score":0.999513,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2137984677","doi":"10.1177/0265532207071510","title":"A confirmatory approach to differential item functioning on an ESL reading assessment","year":2006,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":57,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Alberta Advanced Education","funders":"","keywords":"Differential item functioning; Psychology; Reading (process); Test (biology); Language proficiency; Item response theory; Flagging; Cognitive psychology; Psychometrics; Developmental psychology; Linguistics; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.04018946173988295,"gpt":0.2719887309013568,"spread":0.2317992691614739,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003283558,0.0001712687,0.0001679814,0.0001269849,0.0007569516,0.0004306865,0.0001187387,0.00003516833,0.0003042379],"category_scores_gemma":[0.0001429073,0.0001544034,0.00004287659,0.00004956797,0.00003298739,0.0001479602,0.0000345337,0.0003789527,0.00004515198],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006243904,"about_ca_system_score_gemma":0.000017979,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002071881,"about_ca_topic_score_gemma":0.0001009964,"domain_scores_codex":[0.9988267,0.000133504,0.0002173383,0.0002979494,0.0002178588,0.0003066141],"domain_scores_gemma":[0.9993432,0.0002398286,0.00009873646,0.0002019359,0.00004080019,0.00007556553],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00006704642,0.001330337,0.03656032,0.0003391904,0.000110818,0.0001342017,0.3439125,0.0116457,0.03161876,0.4849252,0.004294096,0.0850618],"study_design_scores_gemma":[0.005339117,0.003028467,0.1635192,0.002937029,0.0004675487,0.000197098,0.3803941,0.3092317,0.001603548,0.001251674,0.1260356,0.005994899],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6617964,0.00001896131,0.001521015,0.00001865965,0.0002064411,0.00009409395,0.000003259159,0.0004304296,0.3359108],"genre_scores_gemma":[0.9885792,3.62004e-8,0.003605275,0.0002652948,0.001968866,0.0000223817,0.00005916184,0.00004007219,0.005459712],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4836735,"threshold_uncertainty_score":0.6296386,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2027399021","doi":"10.1177/0265532207076363","title":"The challenges of the Ontario Secondary School Literacy Test for second language students","year":2007,"lang":"en","type":"article","venue":"Language Testing","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":57,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University","funders":"","keywords":"Literacy; Test (biology); Psychology; Mathematics education; Vocabulary; Context (archaeology); Reading (process); English as a second language; Language proficiency; Language assessment; Pedagogy; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.02266960080524535,"gpt":0.3419876094318603,"spread":0.319318008626615,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001527666,0.0001972514,0.0002089431,0.00007172694,0.00036965,0.00008999265,0.0006713927,0.0001099033,0.00968408],"category_scores_gemma":[0.001711345,0.0001270248,0.0001367495,0.0001998077,0.00008539326,0.0001019499,0.0001373759,0.0006059461,0.0000649857],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008002898,"about_ca_system_score_gemma":0.00007224191,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001972474,"about_ca_topic_score_gemma":0.0104911,"domain_scores_codex":[0.9983564,0.0001147018,0.0004515934,0.0003185431,0.0002505334,0.0005082351],"domain_scores_gemma":[0.9956551,0.003123608,0.0003273544,0.0006830052,0.0001126306,0.00009833123],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000167446,0.0004094151,0.1368472,0.0001930498,0.0002900216,0.0003029247,0.4866394,0.000003225331,0.02598303,0.001457716,0.002375237,0.3453313],"study_design_scores_gemma":[0.001756078,0.0001974341,0.8851979,0.0001678875,0.00004955269,0.0001045111,0.09748836,0.000009586689,0.002934234,0.00009413287,0.01169293,0.000307444],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9249005,0.01903411,0.00004053805,0.0001493482,0.0005367137,0.0004682063,0.00002331365,0.00009866417,0.05474858],"genre_scores_gemma":[0.9743323,8.785945e-7,0.0007906856,0.002058849,0.0004704866,0.00004068418,0.000009012924,0.00004261497,0.02225452],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7483506,"threshold_uncertainty_score":0.9912212,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3040446934","doi":"10.1177/0265532220937830","title":"More efficient processes for creating automated essay scoring frameworks: A demonstration of two algorithms","year":2020,"lang":"en","type":"article","venue":"Language Testing","topic":"Topic Modeling","field":"Computer Science","cited_by":51,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Rubric; Artificial intelligence; Machine learning; Computer science; Support vector machine; Convolutional neural network; Feature engineering; Deep learning; Artificial neural network; Natural language processing; Meaning (existential); Feature (linguistics); Strengths and weaknesses; F1 score; Algorithm; Mathematics; Mathematics education; Linguistics; Psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.0348837319221498,"gpt":0.309711483163347,"spread":0.2748277512411972,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002309982,0.0001175483,0.0001689332,0.00004232086,0.0001169198,0.00009745584,0.0003680387,0.00006002286,0.000001162452],"category_scores_gemma":[0.002601815,0.0001135787,0.00003051833,0.0005957231,0.00001890998,0.0001700886,0.0001225902,0.0001566914,9.166187e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002445424,"about_ca_system_score_gemma":0.0001417599,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006518512,"about_ca_topic_score_gemma":0.000002610978,"domain_scores_codex":[0.9989178,0.00002229097,0.0003065706,0.0003387753,0.0001801548,0.0002344434],"domain_scores_gemma":[0.9987445,0.0005921955,0.0001989598,0.0002060337,0.0001895798,0.00006876143],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000007934903,0.00009054314,0.008498736,0.001534831,0.0000353108,0.00004417334,0.07838282,0.6210707,0.06189445,0.005106845,0.00001136717,0.2233223],"study_design_scores_gemma":[0.0002342855,0.00004007664,0.0001509813,0.0003069196,0.000008587147,0.00000740604,0.00059788,0.9859968,0.01245269,0.0000868758,8.432411e-7,0.0001166918],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2182092,0.0001881094,0.7799336,0.0002883697,0.00004923395,0.0002091117,0.000002442844,0.0008561254,0.0002638107],"genre_scores_gemma":[0.5445089,1.099434e-7,0.4552332,0.0001353325,0.00009661452,0.00001453082,0.000001620078,0.000007476328,0.000002146754],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.3649261,"threshold_uncertainty_score":0.4631606,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3087153552","doi":"10.1177/0265532220957298","title":"Hanyu Shuiping Kaoshi (HSK): A multi-level, multi-purpose proficiency test","year":2020,"lang":"en","type":"article","venue":"Language Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":50,"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":"Language proficiency; Test (biology); Psychology; Argument (complex analysis); Language assessment; Scale (ratio); Mathematics education; Linguistics; Geography","retraction":null,"screen_n_in":null,"score":{"opus":0.8227491867455936,"gpt":0.5099737195301289,"spread":0.3127754672154647,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.005035285,0.0003886459,0.0006416252,0.0004224829,0.0005002351,0.0005397485,0.001728567,0.000150213,0.0003227926],"category_scores_gemma":[0.5866742,0.0002983867,0.0001700977,0.006194631,0.0001416515,0.0004147852,0.0006769855,0.0005704049,0.0004743075],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005077856,"about_ca_system_score_gemma":0.0001198002,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001810563,"about_ca_topic_score_gemma":0.00002195938,"domain_scores_codex":[0.9948754,0.0004547962,0.001214768,0.001278155,0.001273394,0.0009034931],"domain_scores_gemma":[0.9553905,0.04188103,0.0008040175,0.0009346224,0.0004909616,0.0004988471],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00001202431,0.000205641,0.3600578,0.0000548075,0.00001276255,0.0003469287,0.007743378,0.000233849,0.05963743,0.0000365781,0.0007123502,0.5709465],"study_design_scores_gemma":[0.003430956,0.0007012363,0.4814237,0.0003851758,0.00006025032,0.0002136085,0.02398005,0.4791605,0.006773615,0.000457817,0.001710408,0.001702705],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6171786,0.003029125,0.3663179,0.001247611,0.0008284577,0.0009753024,0.00008678201,0.001484194,0.008851971],"genre_scores_gemma":[0.5595,0.000002265452,0.4384884,0.001054371,0.000295155,0.00002014056,0.000003002021,0.00003422099,0.0006024119],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5816389,"threshold_uncertainty_score":0.9999468,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2318200755","doi":"10.1177/0265532213509810","title":"Examining the impact of L2 proficiency and keyboarding skills on scores on TOEFL-iBT writing tasks","year":2013,"lang":"en","type":"article","venue":"Language Testing","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":45,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Test of English as a Foreign Language; Psychology; Test (biology); Language proficiency; Task (project management); Context (archaeology); English language; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.04433657734822507,"gpt":0.3557343872788868,"spread":0.3113978099306617,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001226496,0.0001169974,0.0001442264,0.00008196316,0.0007270469,0.0001978314,0.000168688,0.0000533656,0.00005256522],"category_scores_gemma":[0.005429271,0.00008267799,0.00003961862,0.000318105,0.0001621295,0.0001405333,0.0000348574,0.0001987888,0.00001398422],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008147863,"about_ca_system_score_gemma":0.0001000152,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01037951,"about_ca_topic_score_gemma":0.00005112878,"domain_scores_codex":[0.9987514,0.0001798932,0.0002214735,0.0002193006,0.0002716983,0.0003562355],"domain_scores_gemma":[0.997257,0.002212979,0.000187157,0.0001655378,0.0001054072,0.00007190803],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006716156,0.0001736779,0.4198041,0.00008381053,0.00002802822,0.000008499903,0.21222,0.0003940443,0.02504918,0.002970464,0.0005379997,0.3387235],"study_design_scores_gemma":[0.0003613042,0.000528318,0.7732381,0.002010431,0.00002424047,0.000009936045,0.2186345,0.001930977,0.002191415,0.0006602874,0.000009711766,0.0004008465],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9411235,0.000116044,0.000008362837,0.0001618631,0.00006094724,0.0002501327,0.000002437507,0.0001081984,0.05816849],"genre_scores_gemma":[0.9984141,0.000003414015,0.000697928,0.00008703225,0.000489264,0.00002248504,0.00000201522,0.0000150847,0.0002687064],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.353434,"threshold_uncertainty_score":0.9962105,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2751665091","doi":"10.1177/0265532217725776","title":"Developing and evaluating a computerized adaptive testing version of the Word Part Levels Test","year":2017,"lang":"en","type":"article","venue":"Language Testing","topic":"Second Language Acquisition and Learning","field":"Psychology","cited_by":45,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Affix; Test (biology); Vocabulary; Computerized adaptive testing; Natural language processing; Computer science; Strengths and weaknesses; Word (group theory); Vocabulary development; Psychology; Artificial intelligence; Linguistics; Psychometrics; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.1617082178408199,"gpt":0.3865890369439115,"spread":0.2248808191030916,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.000790643,0.000162312,0.0002278304,0.00005119954,0.0008287188,0.0001021234,0.0003649237,0.00007422241,0.0009806536],"category_scores_gemma":[0.006763358,0.000129223,0.00004254717,0.0001616689,0.0001406091,0.0001251,0.0003329338,0.0002690206,0.00002311434],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003465663,"about_ca_system_score_gemma":0.00005882418,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004465764,"about_ca_topic_score_gemma":0.000009796505,"domain_scores_codex":[0.9987025,0.0001930062,0.000304242,0.0003162948,0.0001861042,0.0002978563],"domain_scores_gemma":[0.9969165,0.001758629,0.0005989927,0.0005382026,0.0001361888,0.00005146461],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00005352444,0.0000640029,0.1665237,0.0001040353,0.00009121967,0.0001821852,0.03009133,0.00004992958,0.08650853,0.0005274448,0.0001435773,0.7156605],"study_design_scores_gemma":[0.001919582,0.0001703422,0.9700329,0.001293133,0.0000598167,0.0001388607,0.01479074,0.009332729,0.001687327,0.0001035056,0.0001032619,0.0003677672],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.985706,0.0006465132,0.0005732497,0.0001113346,0.0003308207,0.0002103145,0.00001156913,0.0001019624,0.01230828],"genre_scores_gemma":[0.9532248,1.43564e-7,0.04534487,0.0007049604,0.0002424073,0.00001027271,0.000002501944,0.00002430771,0.0004456935],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8035092,"threshold_uncertainty_score":0.9999326,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2131222006","doi":"10.1177/0265532210364380","title":"Use of tree-based regression in the analyses of L2 reading test items","year":2010,"lang":"en","type":"article","venue":"Language Testing","topic":"Educational Technology and Assessment","field":"Computer Science","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Reading (process); Cognition; Psychology; Interpretation (philosophy); Cognitive psychology; Test (biology); Tree (set theory); Regression; Regression analysis; Natural language processing; Artificial intelligence; Computer science; Machine learning; Linguistics; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.1065583235754579,"gpt":0.3862878005927542,"spread":0.2797294770172963,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003532011,0.00005269352,0.00008678131,0.000113463,0.00003535407,0.0000175126,0.0004178238,0.00003928541,0.000003445648],"category_scores_gemma":[0.001931721,0.00003362502,0.00002020452,0.0005838154,0.00004787209,0.0001184292,0.00004634073,0.0001721514,6.056441e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000005460204,"about_ca_system_score_gemma":0.00005182611,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003274851,"about_ca_topic_score_gemma":0.00007328141,"domain_scores_codex":[0.9994681,0.00003758151,0.0001667527,0.0001149816,0.0001235235,0.00008900646],"domain_scores_gemma":[0.9972418,0.002189643,0.0001543395,0.0003570463,0.00004635795,0.00001084406],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[6.703237e-7,0.0001308049,0.2848825,0.00001644372,0.000002413642,0.00001200938,0.001244647,0.00006304986,0.6914383,0.0028929,0.00003269767,0.01928355],"study_design_scores_gemma":[0.0002231377,0.0001033221,0.4095229,0.0003797789,0.00001237367,0.00001663616,0.00125326,0.06037855,0.5269459,0.0009841033,0.00003135062,0.0001486324],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961209,0.00003536544,0.002765309,0.0004215768,0.00004787805,0.00005265056,0.000001155401,0.00004014962,0.0005150454],"genre_scores_gemma":[0.8886801,2.169788e-7,0.1112188,0.00005749081,0.00001686807,0.000004652825,0.000001529832,0.000002087322,0.00001823502],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1644924,"threshold_uncertainty_score":0.231259,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4387429753","doi":"10.1177/02655322231202947","title":"Our validity looks like justice. Does yours?","year":2023,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Lethbridge","funders":"","keywords":"Economic Justice; Oppression; Psychology; Licensure; Set (abstract data type); Social psychology; Mathematics education; Pedagogy; Law; Political science; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.1345071057179136,"gpt":0.4132933080995925,"spread":0.2787862023816789,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009659495,0.00009823933,0.0001186746,0.00006772931,0.0006002703,0.0001531346,0.0002785468,0.000071526,0.0001168319],"category_scores_gemma":[0.001352648,0.00007979859,0.00004718095,0.0006186633,0.00004147402,0.0001366406,0.0001072283,0.0001505593,0.000480638],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004373112,"about_ca_system_score_gemma":0.00007037711,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002054527,"about_ca_topic_score_gemma":0.0009748512,"domain_scores_codex":[0.9986535,0.0001098266,0.0001318718,0.0002061942,0.0004596371,0.0004389512],"domain_scores_gemma":[0.9992421,0.000374879,0.00007408189,0.0001468281,0.00006604948,0.0000960645],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.000008392523,0.00009093931,0.7409391,0.0001066006,0.00004430742,0.0004566314,0.1979117,0.00002210043,0.005274525,0.002637563,0.02542272,0.02708534],"study_design_scores_gemma":[0.000614701,0.00006505467,0.2255498,0.0001364411,0.0002020808,0.000002362614,0.7611564,0.0006111523,0.0002490338,0.001368878,0.009372399,0.0006716757],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8966624,0.00004253489,0.00002493187,0.0008389241,0.001366069,0.0001577929,0.000005546059,0.000987027,0.09991485],"genre_scores_gemma":[0.9799092,0.000009063267,0.001238313,0.0001896213,0.001612206,0.00001018265,0.000008395496,0.00001531933,0.01700765],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5632446,"threshold_uncertainty_score":0.617779,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2917835789","doi":"10.1177/0265532219828252","title":"The Test of English for International Communication (TOEIC <sup>®</sup> )","year":2019,"lang":"en","type":"article","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":22,"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":"American Psychological Association; Educational Testing Service","keywords":"TOEIC; Psychology; Test (biology); Linguistics; International communication; Language proficiency; Language assessment; Mathematics education; Communication; Philosophy; Reading (process)","retraction":null,"screen_n_in":null,"score":{"opus":0.02662448091769125,"gpt":0.2592560255326446,"spread":0.2326315446149533,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005150306,0.00006016335,0.0000752822,0.00002708993,0.0002996054,0.000117574,0.0002655998,0.00001740106,0.0001968751],"category_scores_gemma":[0.007995432,0.00004411166,0.00003730037,0.00001941591,0.00005772751,0.00009304401,0.00005034401,0.0001587022,0.00001952353],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001380432,"about_ca_system_score_gemma":0.00001154627,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002372016,"about_ca_topic_score_gemma":0.00006937172,"domain_scores_codex":[0.9995065,0.00004196972,0.0001625412,0.00008246148,0.00009406458,0.0001124611],"domain_scores_gemma":[0.9963285,0.003011285,0.0001255288,0.0002166987,0.0003050616,0.00001291392],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002043021,0.0001103533,0.04517521,0.0001245988,0.00007232754,9.093382e-7,0.7968059,0.001316549,0.0008581142,0.06969133,0.003490851,0.0823334],"study_design_scores_gemma":[0.001450443,0.000313038,0.002293167,0.0006549507,0.00005591299,0.000003043616,0.2929147,0.09292159,0.0003862727,0.001753984,0.606787,0.0004658419],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7852494,0.000457268,0.00002441726,0.000168062,0.0003846665,0.0001964909,0.00002004363,0.0001601474,0.2133395],"genre_scores_gemma":[0.9891253,0.000002002299,0.001304572,0.00007407359,0.0007186623,0.0000116062,0.00002465114,0.0000144013,0.008724702],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6032962,"threshold_uncertainty_score":0.9571856,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2097625210","doi":"10.1177/0265532211404195","title":"Test review: ACCESS for ELLs <sup>®</sup>","year":2011,"lang":"en","type":"article","venue":"Language Testing","topic":"Digital Accessibility for Disabilities","field":"Social Sciences","cited_by":16,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Ell; Psychology; Test (biology); Language proficiency; Mathematics education; Teaching method; Geology","retraction":null,"screen_n_in":null,"score":{"opus":0.1337333077180752,"gpt":0.3811891609412508,"spread":0.2474558532231756,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001037435,0.00014157,0.0002274705,0.00003839436,0.0003029229,0.0002027643,0.0007528855,0.00005873586,0.00052089],"category_scores_gemma":[0.02243293,0.0001296657,0.0001051325,0.0004351246,0.0002944165,0.001050471,0.0001460225,0.0000910075,0.00007847642],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007245631,"about_ca_system_score_gemma":0.0001149795,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003177557,"about_ca_topic_score_gemma":0.0005711977,"domain_scores_codex":[0.9985617,0.00005889945,0.0003228203,0.0003198603,0.0002648245,0.0004718684],"domain_scores_gemma":[0.997835,0.001400597,0.0001316215,0.0003184953,0.0001832787,0.0001309501],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002309784,0.0008700556,0.4358202,0.004529818,0.00005005013,0.00003362805,0.1875175,0.000008624997,0.0001629313,0.01352007,0.01677255,0.3406915],"study_design_scores_gemma":[0.004293988,0.001884846,0.06081007,0.02067779,0.001044333,0.0000512759,0.4171112,0.003021856,0.006448138,0.2746371,0.2019059,0.008113514],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2912447,0.003688116,0.00007261761,0.0003570868,0.0001154107,0.001334959,0.00007815366,0.0006427832,0.7024662],"genre_scores_gemma":[0.9862995,0.00005961528,0.006145985,0.001249018,0.0004111899,0.0001500319,0.00001587488,0.00003567201,0.005633131],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.696833,"threshold_uncertainty_score":0.9858015,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4383069305","doi":"10.1177/02655322231179134","title":"Fairness of using different English accents: The effect of shared L1s in listening tasks of the Duolingo English test","year":2023,"lang":"en","type":"article","venue":"Language Testing","topic":"Phonetics and Phonology Research","field":"Psychology","cited_by":15,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Brock University","funders":"","keywords":"Active listening; Psychology; Stress (linguistics); Test (biology); Interlanguage; Vocabulary; Dictation; Linguistics; Task (project management); Hindi; Communication","retraction":null,"screen_n_in":null,"score":{"opus":0.03822191359303156,"gpt":0.3483421313789877,"spread":0.3101202177859562,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0008786324,0.0001458727,0.0003815934,0.0001429863,0.00006970653,0.00001457166,0.0006334386,0.0001137518,0.00007897416],"category_scores_gemma":[0.009390903,0.00008804037,0.00009271014,0.001026229,0.0001571846,0.00003172339,0.0003855859,0.0004250576,0.000001841431],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002391685,"about_ca_system_score_gemma":0.00002439923,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00161162,"about_ca_topic_score_gemma":0.0001456935,"domain_scores_codex":[0.9983354,0.0004021143,0.0004349853,0.0002219876,0.0002315361,0.0003740347],"domain_scores_gemma":[0.9947923,0.004171662,0.0002871632,0.0005583448,0.0001597648,0.00003075531],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004835302,0.00007894072,0.8246284,0.0004663288,0.00005442847,0.00002409659,0.0846558,0.0002144238,0.08812476,0.00001681029,0.00003569039,0.001652003],"study_design_scores_gemma":[0.001350578,0.0002277478,0.9487026,0.001034373,0.0000666193,0.000006415669,0.006844816,0.003722264,0.03784671,0.00004536225,0.000004057939,0.0001484726],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.997453,0.0002652556,0.000003578032,0.00001367124,0.0003868962,0.0003732477,0.0000648381,0.00004614301,0.0013934],"genre_scores_gemma":[0.9997041,0.000001449327,0.00003500965,0.00001127064,0.0001188582,0.00002514596,0.000009581993,0.00002603434,0.00006849899],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1240742,"threshold_uncertainty_score":0.9989534,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2782384206","doi":"10.1177/0265532217750692","title":"Examining sources of variability in repeaters’ L2 writing scores: The case of the PTE Academic writing section","year":2018,"lang":"en","type":"article","venue":"Language Testing","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Test (biology); Psychology; Context (archaeology); Boston Naming Test; Language assessment; Language proficiency; Mathematics education; Cognition","retraction":null,"screen_n_in":null,"score":{"opus":0.3740090240133829,"gpt":0.4424069973241175,"spread":0.0683979733107346,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04395105,0.000150941,0.0003540737,0.0002827357,0.0003838981,0.00007337797,0.0009004445,0.0001184916,0.00003911367],"category_scores_gemma":[0.4279945,0.00008396868,0.00007881052,0.003934137,0.0004176251,0.000187578,0.00050096,0.0005721146,0.000001565338],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000366806,"about_ca_system_score_gemma":0.00004893656,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001882361,"about_ca_topic_score_gemma":0.0001243278,"domain_scores_codex":[0.9947868,0.00238961,0.00140078,0.0005058862,0.0005480374,0.0003688909],"domain_scores_gemma":[0.8812267,0.1158117,0.001495027,0.001013234,0.0004048634,0.000048406],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006502964,0.000009143778,0.779924,0.00002368205,0.000004543639,0.00003758929,0.008377585,0.0001139314,0.01290258,0.00004890832,0.000008312566,0.1985432],"study_design_scores_gemma":[0.0004777123,0.0001543054,0.8202249,0.0009758305,0.0000330756,0.001429487,0.1217552,0.03262131,0.01640105,0.005641297,0.00001124032,0.0002746601],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916693,0.0002110218,0.002710139,0.00006900854,0.0003475023,0.0002003829,0.000004981368,0.00005400499,0.004733643],"genre_scores_gemma":[0.984133,0.000001063758,0.01533073,0.00007202217,0.000409187,0.000006122288,2.022241e-7,0.000012225,0.00003543585],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3840435,"threshold_uncertainty_score":0.9844536,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4407931412","doi":"10.1177/02655322251319284","title":"The relationship between English language proficiency test scores and academic achievement: A longitudinal study of two tests","year":2025,"lang":"en","type":"article","venue":"Language Testing","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"York University","funders":"","keywords":"Psychology; Language proficiency; Test (biology); Language assessment; Achievement test; Academic achievement; Mathematics education; English language; Test of English as a Foreign Language; Standardized test","retraction":null,"screen_n_in":null,"score":{"opus":0.1073036299005186,"gpt":0.4529902981757598,"spread":0.3456866682752412,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002328164,0.0001025564,0.0001320046,0.00009088909,0.001164699,0.0001345145,0.000347256,0.00005616074,0.000008894818],"category_scores_gemma":[0.05939497,0.00008349853,0.00001636122,0.001056104,0.0002297645,0.000265949,0.0001063877,0.0004853095,0.000004470698],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000503164,"about_ca_system_score_gemma":0.0001958426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004684316,"about_ca_topic_score_gemma":0.001946799,"domain_scores_codex":[0.9983624,0.0004655701,0.0003153876,0.0002452123,0.0003403035,0.0002711549],"domain_scores_gemma":[0.9808006,0.01849031,0.0002943639,0.0002178268,0.0001387672,0.00005815417],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000001580898,0.00006401684,0.8198246,0.00001545275,0.000008235719,0.000001833293,0.1757089,0.000003336502,0.00009713515,0.001594217,0.00001883568,0.002661781],"study_design_scores_gemma":[0.0002238726,0.00005792698,0.8783645,0.00008695196,0.00004788189,2.585959e-7,0.1208417,0.00000606608,0.00002120016,0.0001745028,0.0001020989,0.00007299994],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916908,0.0009025748,0.00001466173,0.000652093,0.0001263449,0.0004682189,0.000002441846,0.0001423895,0.00600044],"genre_scores_gemma":[0.9974722,0.000002597353,0.0005632708,0.0000238835,0.0002798169,0.00003536675,0.000001987813,0.000009082518,0.001611793],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05853987,"threshold_uncertainty_score":0.9485282,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3039959558","doi":"10.1177/0265532220930348","title":"Change in home language environment and English literacy achievement over time: A multi-group latent growth curve modeling investigation","year":2020,"lang":"en","type":"article","venue":"Language Testing","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Latent growth modeling; Literacy; Psychology; Longitudinal study; Home language; Competence (human resources); Population; Academic achievement; Mathematics education; Immigration; Developmental psychology; Pedagogy; Social psychology; Demography; Sociology; Geography; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.08900675997372587,"gpt":0.3676722898153619,"spread":0.278665529841636,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004671992,0.0001352522,0.0001455285,0.0000763698,0.0001477048,0.0001026439,0.0001195078,0.00006784312,0.0001359341],"category_scores_gemma":[0.001167412,0.0001405461,0.00002518815,0.0002685021,0.00005500153,0.0003104684,0.00007851041,0.0001858794,0.00002636755],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009675854,"about_ca_system_score_gemma":0.00003844917,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01660976,"about_ca_topic_score_gemma":0.0003797191,"domain_scores_codex":[0.9987689,0.0001822005,0.0002393303,0.000287847,0.0002242567,0.0002974913],"domain_scores_gemma":[0.9994025,0.0001493591,0.00009099339,0.00009458257,0.00003226719,0.0002303135],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000005001341,0.00005340033,0.1064487,0.00004387823,0.000004701338,0.00001306705,0.8747073,0.00007086137,0.002528952,0.0001679262,0.000007454189,0.01594871],"study_design_scores_gemma":[0.002724653,0.0001521907,0.1623411,0.0004256626,0.00004589073,0.000001753283,0.05355741,0.7783238,0.0003510605,0.0001212912,0.0008601656,0.001095026],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9973543,0.0006442155,0.00006875468,0.0009083477,0.00006481742,0.0003744605,0.00001174089,0.0001333601,0.0004399661],"genre_scores_gemma":[0.991568,0.00003230218,0.005209423,0.002271262,0.000718851,0.00004030769,0.00003606604,0.00001892883,0.0001048365],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8211499,"threshold_uncertainty_score":0.9899387,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3214027731","doi":"10.1177/02655322211052680","title":"Investigating and optimizing score dependability of a local ITA speaking test across language groups: A generalizability theory approach","year":2021,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Generalizability theory; Dependability; Language proficiency; Psychology; Variance (accounting); Test (biology); Construct (python library); Formative assessment; Computer science; Mathematics education; Developmental psychology; Accounting","retraction":null,"screen_n_in":null,"score":{"opus":0.05086898553121055,"gpt":0.3497865942429446,"spread":0.2989176087117341,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002905199,0.0001911051,0.0003354749,0.0000317507,0.0005742435,0.0001790089,0.0002677217,0.0001136641,0.00006458023],"category_scores_gemma":[0.006835991,0.0001870258,0.0000761772,0.0006511075,0.0007264764,0.0002795396,0.0003837637,0.0002698928,0.000001508345],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001011251,"about_ca_system_score_gemma":0.0001679251,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003139915,"about_ca_topic_score_gemma":0.001562892,"domain_scores_codex":[0.997613,0.000506481,0.0003970895,0.0005083492,0.0004474028,0.0005277197],"domain_scores_gemma":[0.9972788,0.001854535,0.0002230849,0.0003255481,0.0001689809,0.0001490236],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.000004317801,0.0001563806,0.6197887,0.0002300039,0.00002669257,0.00004828407,0.3206521,0.00007943944,0.02788865,0.001711013,0.000004185023,0.02941021],"study_design_scores_gemma":[0.0006748703,0.00004565737,0.09744235,0.0002598083,0.00006085772,0.00002414841,0.8915778,0.003956292,0.004598945,0.0008943815,0.000008816892,0.0004561183],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9803843,0.001152884,0.004781025,0.00004414874,0.00006582489,0.0002325634,0.00001666114,0.0001578173,0.01316473],"genre_scores_gemma":[0.9309824,0.000004254037,0.06854533,0.0001099143,0.0002212388,0.00001201781,0.00001568341,0.00002101095,0.00008821429],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5709257,"threshold_uncertainty_score":0.8183813,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4414209357","doi":"10.1177/02655322251348685","title":"Advancing language assessment for teaching and learning in the era of the artificial intelligence (AI) revolution: Promises and challenges","year":2025,"lang":"en","type":"article","venue":"Language Testing","topic":"Second Language Learning and Teaching","field":"Arts and Humanities","cited_by":10,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"Japan Society for the Promotion of Science; Waseda University","keywords":"Language assessment; Language proficiency; Language acquisition; Applications of artificial intelligence; Assessment for learning; Teaching method","retraction":null,"screen_n_in":null,"score":{"opus":0.0365263602611972,"gpt":0.3059178929668351,"spread":0.2693915327056379,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001757634,0.0001118067,0.0001548855,0.0000586237,0.0007475544,0.0001159314,0.0001396504,0.00002938337,0.00001663336],"category_scores_gemma":[0.002600129,0.00007082817,0.00003228499,0.00005018956,0.0001259148,0.0001122084,0.00007478649,0.0008792472,1.660746e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001988695,"about_ca_system_score_gemma":0.0000312204,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001429765,"about_ca_topic_score_gemma":0.002114912,"domain_scores_codex":[0.9989185,0.0003702898,0.0002306891,0.0001923736,0.00009647573,0.0001916847],"domain_scores_gemma":[0.9978697,0.001828024,0.0001155477,0.0001441543,0.0000288906,0.0000136808],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.000005948374,0.00002595493,0.001391932,0.0002963231,0.00001176557,0.00000435305,0.3203359,0.00009481444,0.0007880256,0.05944712,0.000008399749,0.6175895],"study_design_scores_gemma":[0.0002025884,0.000115103,0.005608365,0.001309743,0.00005652868,0.00001632228,0.9757233,0.01262298,0.000211843,0.001914747,0.002014853,0.0002036162],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9741301,0.01030394,0.0006319328,0.002710881,0.0001117608,0.000340868,0.000003098197,0.00006943182,0.011698],"genre_scores_gemma":[0.9961234,0.00001565312,0.002917601,0.0003752391,0.000198031,0.00003035032,0.000002378198,0.00001039782,0.0003269205],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6553874,"threshold_uncertainty_score":0.574966,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2337150116","doi":"10.1177/0265532214559115","title":"Design in four diagnostic language assessments","year":2015,"lang":"en","type":"article","venue":"Language Testing","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Language assessment; Language proficiency; Psychology; Educational assessment; Process (computing); Intervention (counseling); Educational research; Mathematics education; Management science; Computer science; Pedagogy","retraction":null,"screen_n_in":null,"score":{"opus":0.2169898390405153,"gpt":0.4566473782595891,"spread":0.2396575392190738,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007257408,0.0001494996,0.0001647942,0.00009358831,0.00003452504,0.0000345979,0.000237739,0.00008543235,0.0007884694],"category_scores_gemma":[0.001703368,0.0001292658,0.00002396037,0.0004209597,0.00002648387,0.00008381988,0.00004606538,0.0002456873,0.0007392911],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007258174,"about_ca_system_score_gemma":0.00005724497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007418139,"about_ca_topic_score_gemma":0.00002382573,"domain_scores_codex":[0.9985015,0.0003299797,0.0002512382,0.0003275388,0.0002141629,0.0003756204],"domain_scores_gemma":[0.9971808,0.002264897,0.00009315529,0.0002619827,0.0000477164,0.0001514326],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006337519,0.0009274896,0.8644369,0.00001936628,0.00005099646,0.002507317,0.02806913,0.0002056492,0.001733516,0.0009175239,0.006765139,0.0943036],"study_design_scores_gemma":[0.002171991,0.0004366001,0.9734071,0.0001243053,0.00002388082,0.0001648063,0.01967905,0.0003269118,0.00008796591,0.002760428,0.0003349666,0.0004819754],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9064298,0.0006309916,0.002269101,0.0001852939,0.0006810406,0.0002722091,0.000005234132,0.0001438678,0.08938245],"genre_scores_gemma":[0.9737221,5.469306e-7,0.02211025,0.0007883082,0.0002640341,0.0001202889,0.00001667827,0.00002136328,0.002956407],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1089702,"threshold_uncertainty_score":0.9502338,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2168715918","doi":"10.1177/02655322090260020602","title":"Test review: The Versant Spanish <sup>TM</sup> Test","year":2009,"lang":"en","type":"article","venue":"Language Testing","topic":"Neurobiology of Language and Bilingualism","field":"Neuroscience","cited_by":8,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Carleton University","funders":"","keywords":"Test (biology); Psychology; Geology","retraction":null,"screen_n_in":null,"score":{"opus":0.03386165576883628,"gpt":0.2910663115460739,"spread":0.2572046557772376,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0004847054,0.0002847272,0.0002989214,0.00005055487,0.0003848228,0.00008257599,0.0007425413,0.00007886782,0.0001843066],"category_scores_gemma":[0.02949,0.0001874139,0.0001070639,0.0006165411,0.000168146,0.0001402303,0.000123005,0.0004982287,0.0002976029],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002741672,"about_ca_system_score_gemma":0.00004636549,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006992889,"about_ca_topic_score_gemma":0.000006897839,"domain_scores_codex":[0.9979866,0.0001604463,0.00038084,0.0006023324,0.0002737935,0.0005960529],"domain_scores_gemma":[0.9945009,0.004446598,0.0001918994,0.0007069816,0.00005035773,0.0001032031],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.000005754302,0.0002320295,0.001291096,0.000112768,0.00000326894,0.006567173,0.002810745,0.00001834519,0.9376535,0.00003729121,0.01138912,0.03987885],"study_design_scores_gemma":[0.004172631,0.004306541,0.008966379,0.007169669,0.0006270901,0.03521606,0.00570645,0.006792856,0.8342755,0.001395591,0.08682118,0.004549991],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9511229,0.01108701,0.0000110087,0.006976518,0.0001995137,0.0009214503,0.00007281004,0.001027612,0.02858112],"genre_scores_gemma":[0.9458494,0.0003100868,0.000481365,0.05171693,0.0005264709,0.00001179045,0.000008989965,0.00002979082,0.001065242],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.103378,"threshold_uncertainty_score":0.978685,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Language proficiency; Psychology; Test (biology); Sophistication; Linguistics; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.4041793967710166,"gpt":0.4111835522633015,"spread":0.007004155492284891,"validation_status":"score_only:v0-immature-baseline"},"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"},"labels":[],"label_agreement":null},{"id":"W1979398480","doi":"10.1177/0265532214560799","title":"A prototype of a receptive lexical test for a polysynthetic heritage language: The case of Inuttitut in Labrador","year":2014,"lang":"en","type":"article","venue":"Language Testing","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"The Scarborough Hospital; University of Toronto; Carleton University","funders":"","keywords":"Heritage language; Linguistics; Vocabulary; Psychology; Test (biology); Comprehension; Noun; Language proficiency; First language; Natural language processing; Computer science; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.01657557252071323,"gpt":0.2924684211711534,"spread":0.2758928486504402,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008631768,0.0001479754,0.0002647644,0.0001220079,0.00006213992,0.00004062253,0.0006869013,0.0000767355,0.000005810576],"category_scores_gemma":[0.006430248,0.0001034157,0.00005826448,0.0007090424,0.0001224164,0.0001761099,0.0002088232,0.0002196644,0.000001592853],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000283952,"about_ca_system_score_gemma":0.0000630674,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001911266,"about_ca_topic_score_gemma":0.0005231013,"domain_scores_codex":[0.9988532,0.0001157721,0.0003498149,0.0002857773,0.0001332111,0.0002622275],"domain_scores_gemma":[0.9971865,0.001887448,0.0002488495,0.0005162024,0.0001227498,0.00003824364],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0001319795,0.0004387789,0.002218223,0.001145561,0.00002478493,0.0009872088,0.07159334,0.00001471745,0.2960268,0.02366296,0.00007269823,0.6036829],"study_design_scores_gemma":[0.003050524,0.005163533,0.001178501,0.003294995,0.00009121018,0.003038223,0.01366062,0.196867,0.7483045,0.02382121,0.00009509802,0.001434582],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8981039,0.002339057,0.09549961,0.0005780943,0.00006966079,0.002017459,0.00002387462,0.0007474693,0.0006208789],"genre_scores_gemma":[0.6692018,2.32114e-7,0.3305553,0.00006901882,0.00003368786,0.0000973397,6.471516e-7,0.00001421942,0.00002778406],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6022483,"threshold_uncertainty_score":0.7698072,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2162394022","doi":"10.1177/0265532212469178","title":"Differential importance of language components in determining secondary school students’ Chinese reading literacy performance","year":2013,"lang":"en","type":"article","venue":"Language Testing","topic":"Reading and Literacy Development","field":"Psychology","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Dictation; Psychology; Reading (process); Copying; Reading comprehension; Literacy; Mathematics education; Chinese characters; Linguistics; Pedagogy; Computer science; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.01812018530720183,"gpt":0.3170660889351399,"spread":0.298945903627938,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002662383,0.0002502109,0.0003731654,0.0002442005,0.00007254863,0.00009927581,0.0004003354,0.00007972239,0.002066661],"category_scores_gemma":[0.0003056398,0.0002185042,0.00005023323,0.0003738977,0.00003462559,0.0003704548,0.0001552567,0.0003978766,0.0001758299],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006587046,"about_ca_system_score_gemma":0.00002408929,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001044276,"about_ca_topic_score_gemma":0.00001467605,"domain_scores_codex":[0.998078,0.0000987646,0.0006920455,0.0003878261,0.0002695629,0.000473748],"domain_scores_gemma":[0.9987798,0.000331315,0.000307218,0.0004071322,0.00005777083,0.0001168118],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001393026,0.00009916065,0.888968,0.00008557476,0.0000250747,0.0001208842,0.03972847,0.000002881238,0.0246125,0.000002987279,0.00003018866,0.04631038],"study_design_scores_gemma":[0.0009429338,0.00004835351,0.9947724,0.0004688483,0.000006946192,0.00005335868,0.002020498,0.001000119,0.0004340401,0.000007810401,0.000007172287,0.0002375281],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9909829,0.0002663504,0.00002065973,0.000008354644,0.0003215339,0.0003004298,0.000006310785,0.0001098363,0.007983626],"genre_scores_gemma":[0.9945767,0.000001382941,0.003325519,0.0001323141,0.0001512092,0.00007978603,0.00004466158,0.00003866211,0.001649776],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1058044,"threshold_uncertainty_score":0.9988456,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4399235559","doi":"10.1177/02655322241249754","title":"A scoping review of research on second language test preparation","year":2024,"lang":"en","type":"review","venue":"Language Testing","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":4,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université de Sherbrooke; Western University","funders":"","keywords":"Psychology; Test (biology); Language assessment; Language proficiency; Linguistics; Mathematics education","retraction":null,"screen_n_in":null,"score":{"opus":0.2975427459650408,"gpt":0.5064660563792833,"spread":0.2089233104142425,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002857443,0.0003316712,0.00113086,0.0004036841,0.0002979753,0.0002411573,0.000313806,0.000109282,0.001923733],"category_scores_gemma":[0.005236507,0.0002462392,0.0002625293,0.0002987779,0.0001214631,0.00009054151,0.0001337426,0.001538868,0.0006245426],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008695065,"about_ca_system_score_gemma":0.0002720932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002095975,"about_ca_topic_score_gemma":0.0002565476,"domain_scores_codex":[0.997327,0.0005851205,0.0007761786,0.0004817913,0.0004569193,0.0003729511],"domain_scores_gemma":[0.995402,0.003518136,0.0003899886,0.0004972785,0.0001346633,0.00005791002],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[5.984866e-7,0.0000288953,3.62018e-7,0.654337,0.00003354335,0.00008406914,0.02093856,2.046347e-7,0.000001883581,0.0009199551,0.002124677,0.3215302],"study_design_scores_gemma":[0.00002612383,0.0001190496,3.345196e-8,0.7317721,0.0002025616,0.00001831344,0.001542157,0.000009102262,0.000001402808,0.000007601841,0.2661115,0.0001899974],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00002899147,0.8539596,4.019504e-7,0.00001098345,0.0001858459,0.0008732346,0.00005461525,0.000259941,0.1446264],"genre_scores_gemma":[0.0001100699,0.9759648,0.000155739,0.0000969822,0.00175085,0.0001452366,0.0001509457,0.0001124038,0.02151299],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.3213402,"threshold_uncertainty_score":0.999999,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4381740816","doi":"10.1177/02655322231179128","title":"Rethinking student placement to enhance efficiency and student agency","year":2023,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Université de Montréal; Carleton University; University of Ottawa","funders":"University of Ottawa","keywords":"Agency (philosophy); Context (archaeology); Psychology; Process (computing); Test (biology); Mathematics education; Language proficiency; Pedagogy; Computer science; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.0478632348052326,"gpt":0.4160427232575863,"spread":0.3681794884523537,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001602769,0.0001337004,0.000153979,0.0001175375,0.0007081409,0.0002146893,0.0003651093,0.000045539,0.0001070074],"category_scores_gemma":[0.0006315287,0.0001309165,0.00002855947,0.0009903419,0.00005970634,0.0001030892,0.0003558068,0.0001326139,0.0001850033],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001054118,"about_ca_system_score_gemma":0.00005064946,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005803019,"about_ca_topic_score_gemma":0.0004697135,"domain_scores_codex":[0.9979709,0.0001123081,0.0002277678,0.0003614791,0.0008100573,0.000517514],"domain_scores_gemma":[0.9990919,0.0004510641,0.00007926141,0.000170709,0.00005885933,0.0001482048],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000003987525,0.00008559251,0.1962958,0.00001941549,0.00002164129,0.0001023355,0.7764701,0.00007105194,0.005299756,0.001007949,0.0004322748,0.02019017],"study_design_scores_gemma":[0.0005216069,0.0002973473,0.688284,0.0003312432,0.00005710832,0.0000018609,0.3067774,0.0002314834,0.0004572754,0.0005923039,0.001687935,0.0007603486],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9497128,0.0002179289,0.0001059555,0.0004611493,0.0003495526,0.0004817287,0.000001324167,0.0004371943,0.04823238],"genre_scores_gemma":[0.9938977,0.00002863117,0.002103886,0.0002629336,0.00040765,0.0000448326,0.00000249621,0.00001571508,0.003236095],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4919883,"threshold_uncertainty_score":0.5446519,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4404521715","doi":"10.1177/02655322241291764","title":"Review of the Canadian English Language Proficiency Index Program (CELPIP)","year":2024,"lang":"en","type":"article","venue":"Language Testing","topic":"Multilingual Education and Policy","field":"Social Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa","funders":"","keywords":"Psychology; Language proficiency; Index (typography); Language assessment; Linguistics; Mathematics education; Computer science; Philosophy","retraction":null,"screen_n_in":null,"score":{"opus":0.06094631709999317,"gpt":0.4570166644197495,"spread":0.3960703473197564,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008019099,0.00007301444,0.0000946068,0.00006603318,0.0002904725,0.00009248687,0.0002993015,0.00005246398,0.0003693493],"category_scores_gemma":[0.006670138,0.00005301023,0.000055435,0.001021299,0.0001315451,0.00006578924,0.00002684972,0.0001938374,0.00002129558],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001201846,"about_ca_system_score_gemma":0.001719958,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.52345,"about_ca_topic_score_gemma":0.3617387,"domain_scores_codex":[0.9990414,0.000130754,0.0001757909,0.0001409557,0.0002245486,0.000286565],"domain_scores_gemma":[0.999315,0.0001739821,0.00005688514,0.0001958712,0.0001245906,0.0001336352],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[3.003819e-7,0.00003652089,0.00646415,0.002309049,0.000007950159,0.00001297997,0.4624062,0.000001014016,0.00009973927,0.004638097,0.00342061,0.5206034],"study_design_scores_gemma":[0.000154009,0.00004967563,0.009794569,0.01948521,0.0000804322,0.000009522319,0.1107338,0.0009294776,0.0004657567,0.0001335711,0.8576055,0.0005584374],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2110727,0.06271833,0.000001386185,0.002894062,0.001402289,0.001285487,0.00002472232,0.00056519,0.7200358],"genre_scores_gemma":[0.9939931,0.0000995364,0.0003642062,0.001491181,0.0006807329,0.00003448559,0.000004396866,0.00001238293,0.003319982],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8541849,"threshold_uncertainty_score":0.798526,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4412969591","doi":"10.1177/02655322251348956","title":"Investigating construct representativeness and linguistic equity of automated oral reading fluency assessment with prosody","year":2025,"lang":"en","type":"article","venue":"Language Testing","topic":"Reading and Literacy Development","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Christian Studies","funders":"Iran Science Elites Federation; University of Toronto","keywords":"Prosody; Fluency; Psychology; Ell; Natural language processing; Linguistics; Reading comprehension; Reading (process); Cognitive psychology; Computer science; Artificial intelligence; Mathematics education; Speech recognition; Teaching method; Vocabulary development","retraction":null,"screen_n_in":null,"score":{"opus":0.03767559474110727,"gpt":0.4197115117810315,"spread":0.3820359170399242,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005905341,0.0001595972,0.0002892983,0.0001309257,0.0001339846,0.00005739102,0.0001224568,0.00005154975,0.00002796008],"category_scores_gemma":[0.001464815,0.0001345098,0.00001576145,0.0005225188,0.0001723755,0.00005518448,0.0001334808,0.0001881072,0.000001514176],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006048804,"about_ca_system_score_gemma":0.0001546884,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001128564,"about_ca_topic_score_gemma":0.00001027027,"domain_scores_codex":[0.9986468,0.0001679299,0.0003940305,0.0003623433,0.0001525348,0.000276393],"domain_scores_gemma":[0.998535,0.0007705357,0.0002423141,0.0002467401,0.000143951,0.00006141149],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001490088,0.00004736005,0.916972,0.0003527353,0.0001207863,0.0001221841,0.02369158,0.000028536,0.0189666,0.006431818,0.00004191571,0.03320953],"study_design_scores_gemma":[0.001836812,0.0001943628,0.9677733,0.003299157,0.0001291546,0.0001636559,0.0102658,0.008894078,0.00549959,0.001460838,0.00003580002,0.0004474245],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9338176,0.0001451094,0.0005653152,0.00002280596,0.0001579204,0.0002457548,0.000006272701,0.0003478624,0.06469138],"genre_scores_gemma":[0.9152763,2.566853e-7,0.08413101,0.00006670257,0.00002402311,0.00003345989,0.00001211076,0.00001417085,0.0004419631],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08356569,"threshold_uncertainty_score":0.5485149,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4412676320","doi":"10.1177/02655322251359320","title":"Book review: Innovation in Learning-Oriented Language Assessment ChongS.ReindersH. (Eds.), Innovation in Learning-Oriented Language Assessment. Palgrave MacMillan, 2023. 333 pp. ISBN 978-3-031-18949-4 (hbk) US$169.99 ISBN 978-3-031-18952-4 (sbk) US$169.99 ISBN 978-3-031-18950-0 (ebk) US$129.99","year":2025,"lang":"en","type":"article","venue":"Language Testing","topic":"Student Assessment and Feedback","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":"University of Saskatchewan","funders":"","keywords":"Psychology; Linguistics; Sociology; Philosophy","retraction":null,"screen_n_in":null,"score":{"opus":0.01741045735498279,"gpt":0.3375408814479155,"spread":0.3201304240929327,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.007998494,0.001879939,0.002420376,0.002352894,0.001695652,0.000857047,0.001874603,0.001139836,0.002981932],"category_scores_gemma":[0.00837872,0.002022994,0.000453341,0.01953098,0.0006879977,0.001019182,0.001206,0.004428082,0.0002496684],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002595348,"about_ca_system_score_gemma":0.001904123,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0189951,"about_ca_topic_score_gemma":0.004550829,"domain_scores_codex":[0.9835112,0.002675777,0.004356312,0.002864254,0.00338429,0.003208152],"domain_scores_gemma":[0.991594,0.001674978,0.003065192,0.001579561,0.001681993,0.0004043325],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002629676,0.001121699,0.8396774,0.002089095,0.0006597587,0.001658584,0.03067411,0.0007213392,0.01934605,0.009898183,0.05337339,0.04051743],"study_design_scores_gemma":[0.01972423,0.001885326,0.4432375,0.01700424,0.001256147,0.00009305384,0.1819585,0.0174547,0.002045948,0.0006542268,0.3059624,0.008723797],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7594275,0.09312202,0.002163674,0.006341017,0.0031513,0.008493182,0.0001724569,0.002834865,0.1242939],"genre_scores_gemma":[0.9216686,0.01345022,0.009134124,0.01143327,0.002136762,0.001303249,0.005094005,0.0005349128,0.03524487],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3964399,"threshold_uncertainty_score":0.9999741,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}