{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":10,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":10,"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","author_layer_release":"2026-06-26"},"query_hash":"a7f5fc1f824e","filters":{"venue":"ETS Research Report Series"}},"results":[{"id":"W1964735878","doi":"10.1002/j.2333-8504.2005.tb01990.x","title":"ANALYSIS OF DISCOURSE FEATURES AND VERIFICATION OF SCORING LEVELS FOR INDEPENDENT AND INTEGRATED PROTOTYPE WRITTEN TASKS FOR THE NEW TOEFL®","year":2005,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Software Engineering Research","field":"Computer Science","cited_by":55,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Christian Studies; University of Toronto","funders":"","keywords":"Test of English as a Foreign Language; Computer science; Natural language processing; Psychology; Artificial intelligence; Mathematics education; Language assessment","authors":[{"name":"Alister Cumming","is_ca":true},{"name":"Robert Kantor","is_ca":false},{"name":"Kyoko Baba","is_ca":true},{"name":"Keanre Eouanzoui","is_ca":true},{"name":"Usman Erdosy","is_ca":true},{"name":"Mark Jamse","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.075963810214805,"gpt":0.3964993958148846,"spread":0.3205355856000797,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002325503,0.00009337194,0.0002290504,0.000387119,0.0001441025,0.0001349958,0.0004175851,0.00005773546,0.000003968843],"category_scores_gemma":[0.002871457,0.00006504294,0.00005791264,0.0009244829,0.0002176304,0.000404753,0.0002159826,0.0001896613,1.189946e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004085656,"about_ca_system_score_gemma":0.0003229892,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002624266,"about_ca_topic_score_gemma":0.0002016873,"domain_scores_codex":[0.9984255,0.00005007491,0.000300699,0.0003434134,0.000604879,0.0002754164],"domain_scores_gemma":[0.9974728,0.001101339,0.0001050182,0.0005986843,0.0006315527,0.00009067286],"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":[0.00145126,0.0003381527,0.120562,0.001509829,0.003612379,0.00005676824,0.01757843,0.005264806,0.06851996,0.03216276,0.004518662,0.7444251],"study_design_scores_gemma":[0.0006414898,0.0008266608,0.8658916,0.0001377187,0.0001261196,0.00009286679,0.0008664462,0.02441253,0.09696329,0.001284616,0.008498325,0.0002583765],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5566862,0.001961216,0.4309771,0.006573038,0.00008805842,0.003559347,0.00004442733,0.00007461809,0.00003601002],"genre_scores_gemma":[0.9746764,0.00007626827,0.02381977,0.000003237956,0.00004705236,0.0003226921,0.00001110969,0.00001068321,0.001032779],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7453296,"threshold_uncertainty_score":0.343761,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1980407112","doi":"10.1002/j.2333-8504.2009.tb02187.x","title":"THE SPEAKING SECTION OF THE TOEFL IBT™ (SSTIBT): TEST‐TAKERS' REPORTED STRATEGIC BEHAVIORS","year":2009,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":32,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Victoria; York University; Institute for Christian Studies; University of Toronto","funders":"","keywords":"Test of English as a Foreign Language; Psychology; Test (biology); Metacognition; Construct (python library); Cognition; Variety (cybernetics); Social psychology; Cognitive psychology; Mathematics education; Applied psychology; Language assessment; Computer science; Artificial intelligence","authors":[{"name":"Merrill Swain","is_ca":true},{"name":"Li‐Shih Huang","is_ca":true},{"name":"Khaled Barkaoui","is_ca":true},{"name":"Lindsay Brooks","is_ca":true},{"name":"Sharon Lapkin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1443413160757023,"gpt":0.3729416411252792,"spread":0.2286003250495769,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.004067208,0.0001570593,0.0001933114,0.00010479,0.002841794,0.0005466018,0.0003375469,0.00007048049,0.0002973302],"category_scores_gemma":[0.001813677,0.00008985545,0.0001325018,0.0001896505,0.0008294169,0.0003237411,0.00008796868,0.001145668,0.000009415438],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007959481,"about_ca_system_score_gemma":0.0002828659,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001782838,"about_ca_topic_score_gemma":0.002777038,"domain_scores_codex":[0.9971548,0.0004110753,0.0006146646,0.0003178506,0.001014215,0.0004873972],"domain_scores_gemma":[0.997868,0.0003553775,0.0004001861,0.0007699071,0.0005373438,0.00006917653],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0006388489,0.001535036,0.06806846,0.0003552498,0.0005239923,0.004827368,0.3024724,0.0004337766,0.04994818,0.4570373,0.03886177,0.07529759],"study_design_scores_gemma":[0.0003101083,0.001156317,0.05569619,0.0003838345,0.00006758446,0.001237258,0.07278277,0.00006677418,0.004299241,0.01785676,0.8456807,0.0004625227],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8078617,0.0002406608,0.000002195338,0.003013247,0.000660957,0.0003030699,0.000003075752,0.0001169436,0.1877981],"genre_scores_gemma":[0.9400244,0.00004348408,0.0000215551,0.00002324498,0.0006478191,0.00001331439,0.00000918851,0.00001958304,0.05919744],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8068188,"threshold_uncertainty_score":0.9984564,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1504933570","doi":"10.1002/ets2.12050","title":"Test Takers' Writing Activities During the<i><scp>TOEFL iBT</scp><sup>®</sup></i>Writing Tasks: A Stimulated Recall Study","year":2015,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Writing and Handwriting Education","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Test of English as a Foreign Language; Test (biology); Task (project management); Psychology; Second language writing; Mathematics education; Recall; English language; Cognitive psychology; Linguistics; Second language","authors":[{"name":"Khaled Barkaoui","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1097165021020219,"gpt":0.4172894513180939,"spread":0.3075729492160719,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.02048649,0.0003190058,0.000445477,0.0003681246,0.004881889,0.00168364,0.0007284533,0.0002223905,0.00004990469],"category_scores_gemma":[0.0403089,0.0002890061,0.0001204186,0.001517614,0.00122298,0.001313814,0.0004046568,0.001345979,0.00007762029],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005457953,"about_ca_system_score_gemma":0.001774438,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005776494,"about_ca_topic_score_gemma":0.002016699,"domain_scores_codex":[0.9915826,0.002073101,0.0009412749,0.0008497308,0.002905735,0.00164751],"domain_scores_gemma":[0.9920149,0.004593248,0.0003726342,0.0008143476,0.001642799,0.0005620204],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00009618516,0.001585794,0.470497,0.000312267,0.0002344677,0.00278136,0.4987831,0.0006751416,0.001664592,0.001383823,0.01668504,0.005301282],"study_design_scores_gemma":[0.0006766269,0.0004154456,0.02034014,0.0003757244,0.00002777424,0.0002966138,0.9442359,0.0001601757,0.0006693769,0.001278314,0.03130058,0.0002233955],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9291077,0.0003646374,0.000004474733,0.003074386,0.0002167611,0.001105557,0.00001302504,0.0004951194,0.06561836],"genre_scores_gemma":[0.969063,0.0001077817,0.0002619346,0.00003683464,0.001924188,0.0002608782,0.00002559253,0.00006938612,0.02825037],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4501569,"threshold_uncertainty_score":0.9999562,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2069986276","doi":"10.1002/j.2333-8504.2004.tb01952.x","title":"TOWARD ACCESSIBLE COMPUTER‐BASED TESTS: PROTOTYPES FOR VISUAL AND OTHER DISABILITIES","year":2004,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Digital Accessibility for Disabilities","field":"Social Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"University of Toronto; Advanced Technology Research Council","keywords":"Formative assessment; Blindness; Psychology; Test (biology); Learning disability; Visual impairment; Medical education; Computerized adaptive testing; Applied psychology; Assistive technology; Computer science; Mathematics education; Human–computer interaction; Clinical psychology; Developmental psychology; Optometry; Psychometrics; Medicine","authors":[{"name":"Eric G. Hansen","is_ca":false},{"name":"Douglas C. Forer","is_ca":false},{"name":"Moon J. Lee","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1843955837931545,"gpt":0.4780755513786063,"spread":0.2936799675854518,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.003111931,0.000185731,0.0002974705,0.0001461485,0.000984117,0.00118136,0.0004099578,0.0001264451,0.0001396103],"category_scores_gemma":[0.005444096,0.0001599095,0.0001068984,0.0004505113,0.004327635,0.001796404,0.0002226171,0.000201211,0.00001345293],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003019136,"about_ca_system_score_gemma":0.001289616,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004380303,"about_ca_topic_score_gemma":0.007363474,"domain_scores_codex":[0.9968027,0.0001944458,0.0004706237,0.0006145991,0.001124631,0.0007929993],"domain_scores_gemma":[0.997733,0.0009533557,0.00009839323,0.0003693312,0.0005921012,0.0002537988],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001164665,0.002260959,0.421461,0.00361994,0.000160401,0.0001505179,0.07920618,0.0000955545,0.0002979466,0.4637354,0.002380896,0.02546652],"study_design_scores_gemma":[0.001190255,0.001846881,0.01335456,0.0003745156,0.00001545997,0.00004182957,0.06661376,0.00003965817,0.006294918,0.644321,0.2651275,0.0007796875],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9511189,0.0001608944,0.0003632047,0.007080005,0.0002171421,0.003457689,0.00005825857,0.000307891,0.03723599],"genre_scores_gemma":[0.9917195,0.00001065504,0.002965639,0.00008572623,0.0003822946,0.00109118,0.00001724123,0.00003437316,0.0036934],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4081065,"threshold_uncertainty_score":0.9998555,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3088847275","doi":"10.1002/ets2.12307","title":"Reflections on<scp>Equity‐Centered</scp>Design","year":2020,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":8,"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":"Verisimilitude; Equity (law); Relevance (law); Test (biology); Management science; Computer science; Psychology; Political science; Economics","authors":[{"name":"María Elena Oliveri","is_ca":false},{"name":"Jessica Nastal","is_ca":false},{"name":"David Slomp","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5692019926705136,"gpt":0.6024174074277371,"spread":0.03321541475722345,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0054982,0.0001219495,0.000170635,0.0001401756,0.002182084,0.0006630099,0.0005269752,0.0001245952,0.0008313556],"category_scores_gemma":[0.03034585,0.0001221826,0.00006479416,0.001443457,0.00071203,0.0009797631,0.0002063587,0.0009373097,0.001429113],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002320183,"about_ca_system_score_gemma":0.001943566,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001164912,"about_ca_topic_score_gemma":0.00051527,"domain_scores_codex":[0.9944441,0.002080021,0.0003350042,0.0005106632,0.001908906,0.0007212447],"domain_scores_gemma":[0.9963706,0.001858983,0.0001960724,0.0003929071,0.0006332685,0.0005481839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000865306,0.0003407056,0.005120411,0.00005842329,0.0001156898,0.0009978901,0.2689483,0.0001516368,0.002154675,0.04601003,0.6723492,0.003666463],"study_design_scores_gemma":[0.0000576321,0.0002400161,0.001589889,0.00001758244,0.000004846385,0.00002313796,0.02419969,0.000002107792,0.0008296015,0.001176995,0.9718056,0.00005292914],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.0244606,0.0001399032,0.0003627645,0.1516748,0.0008806431,0.0006796742,0.00000417323,0.0004441757,0.8213533],"genre_scores_gemma":[0.818301,0.000321313,0.002826511,0.0007535476,0.001519826,0.0001451566,0.00001289767,0.00004077998,0.1760789],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.7938404,"threshold_uncertainty_score":0.9993484,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3082102071","doi":"10.1002/ets2.12303","title":"Perspectives on Social and Emotional Learning in Tertiary Education","year":2020,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Emotional Intelligence and Performance","field":"Psychology","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Numeracy; Complementarity (molecular biology); Higher education; Institution; Psychology; Social emotional learning; Literacy; Competition (biology); Mathematics education; Tertiary institution; Pedagogy; Sociology; Medical education; Developmental psychology; Economic growth; Social science; Economics","authors":[{"name":"Catherine M. Millett","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1898114876669374,"gpt":0.4860659654415337,"spread":0.2962544777745962,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005558484,0.00006352143,0.00007892685,0.0001198427,0.0002192084,0.00003592034,0.00006729963,0.00006189572,0.001133406],"category_scores_gemma":[0.0003634687,0.00006101958,0.00002124684,0.0002233201,0.0001985538,0.0001885379,0.00004049131,0.0005417507,0.0001718837],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000404336,"about_ca_system_score_gemma":0.0001793514,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001210113,"about_ca_topic_score_gemma":0.00002106745,"domain_scores_codex":[0.9988604,0.0001759479,0.0001665095,0.00028194,0.0003159151,0.0001993287],"domain_scores_gemma":[0.9995957,0.00008168248,0.00003753891,0.00006999804,0.0001528041,0.00006228875],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006457024,0.0005449735,0.6372293,0.0001151313,0.00006009851,0.0006397369,0.1580217,0.00003805913,0.001052233,0.1223074,0.02742839,0.05191736],"study_design_scores_gemma":[0.0000674316,0.0002043687,0.9340359,0.00003472754,8.188248e-7,0.0001367031,0.03706847,0.00001987717,0.0001828737,0.001445752,0.02671964,0.00008344485],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9362589,0.0003925133,0.000009483839,0.02155674,0.0001033634,0.0001207159,0.00000125239,0.00002600141,0.04153102],"genre_scores_gemma":[0.9905525,0.0001406125,0.00004945426,0.0001834417,0.0004739234,0.00003626151,0.00001929774,0.0000086665,0.008535773],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2968066,"threshold_uncertainty_score":0.9997797,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1971125305","doi":"10.1002/j.2333-8504.2003.tb01910.x","title":"INVESTIGATING THE VALIDITY OF TOEFL: A FEASIBILITY STUDY USING CONTENT AND CRITERION‐RELATED STRATEGIES","year":2003,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"EFL/ESL Teaching and Learning","field":"Arts and Humanities","cited_by":6,"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; Context (archaeology); Psychology; Test (biology); Content validity; Language proficiency; Test validity; Sample (material); Computer science; Mathematics education; Language assessment; Natural language processing; Psychometrics; Clinical psychology","authors":[{"name":"Michael Rosenfeld","is_ca":false},{"name":"Philip K. Oltman","is_ca":false},{"name":"Ken Sheppard","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5438925237843821,"gpt":0.4462797001089897,"spread":0.09761282367539248,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.007013302,0.0001298127,0.0002445516,0.00008450144,0.001573997,0.0005526277,0.0001264494,0.00003764588,0.0002595421],"category_scores_gemma":[0.003022952,0.00008642272,0.00004898327,0.00008944524,0.001376146,0.0004868943,0.0001114993,0.0006635842,0.000001660367],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004090185,"about_ca_system_score_gemma":0.0002171462,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002599136,"about_ca_topic_score_gemma":0.0009764915,"domain_scores_codex":[0.9966228,0.001699173,0.0004985254,0.0002989809,0.0005765628,0.0003039604],"domain_scores_gemma":[0.9985308,0.0003407855,0.0001933455,0.0003955294,0.0004635819,0.00007593726],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00004020946,0.0005395424,0.3158163,0.0003090967,0.0001961901,0.0002988027,0.5237395,0.00006486661,0.005306699,0.1528761,0.0002153473,0.0005973483],"study_design_scores_gemma":[0.0005570443,0.001230132,0.07803874,0.0002380291,0.00004806123,0.0003857601,0.8791475,0.0001200095,0.0008419356,0.02351139,0.01555473,0.0003266314],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9841573,0.0002134551,0.000005046367,0.0002421663,0.0001366207,0.0004641621,0.000002801223,0.00005115774,0.01472728],"genre_scores_gemma":[0.9971981,0.000008505272,0.000104216,0.000009125568,0.00005503007,0.00001248074,0.000003175457,0.00001655578,0.002592792],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.355408,"threshold_uncertainty_score":0.9997258,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4307159924","doi":"10.1002/ets2.12357","title":"Mapping<i>TOEFL</i>®<i>Essentials</i>™ Test Scores to the Canadian Language Benchmarks","year":2022,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":2,"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; Test (biology); Construct (python library); Psychology; Mathematics education; Computer science; Language assessment; English language; Test score; Natural language processing; Standardized test; Programming language","authors":[{"name":"Spiros Papageorgiou","is_ca":false},{"name":"Larry Davis","is_ca":false},{"name":"Renka Ohta","is_ca":false},{"name":"Pablo Garcia Gomez","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09859811587093853,"gpt":0.4617409910682972,"spread":0.3631428751973587,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.006416193,0.00009045105,0.0001162375,0.0003120672,0.006611443,0.0003840403,0.0007106931,0.00004267483,0.009824703],"category_scores_gemma":[0.001921398,0.00007704415,0.00005149372,0.001273948,0.0004081313,0.0002053988,0.0002971046,0.0005445959,0.00008046195],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005281602,"about_ca_system_score_gemma":0.006739535,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.7591901,"about_ca_topic_score_gemma":0.9640564,"domain_scores_codex":[0.9962129,0.0006559732,0.0002401745,0.0003160696,0.001832283,0.0007425827],"domain_scores_gemma":[0.9982103,0.0005616222,0.00006714972,0.0004099106,0.0003520547,0.0003989399],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001593019,0.0001096059,0.1279547,0.00001697038,0.00003527167,0.001091036,0.1433097,0.00003268715,0.0006132636,0.02694616,0.6979513,0.001923415],"study_design_scores_gemma":[0.00002405048,0.00004440483,0.02741419,0.000005083078,0.000001607637,0.00004111898,0.08407428,9.840101e-7,0.00005419144,0.001237123,0.887009,0.00009402836],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.3261531,0.0004856556,0.000002959477,0.2413963,0.001309176,0.0009804117,0.00008441723,0.00005317849,0.4295348],"genre_scores_gemma":[0.9034816,0.00003666118,0.000108991,0.0006017162,0.000853192,0.000362953,0.000160996,0.00001227051,0.0943816],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5773286,"threshold_uncertainty_score":0.9988914,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2342001105","doi":"10.1002/ets2.12097","title":"The Prediction of Labor Force Status: Implications From International Adult Skill Assessments","year":2016,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Labor market dynamics and wage inequality","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Literacy; Categorical variable; Educational attainment; Adult literacy; Probit model; Demographic economics; Psychology; Immigration; Regression analysis; Probit; Demography; Econometrics; Geography; Economics; Sociology; Economic growth; Statistics; Pedagogy","authors":[{"name":"Tongyun Li","is_ca":false},{"name":"Matthias von Davier","is_ca":false},{"name":"Gregory R. Hancock","is_ca":false},{"name":"Irwin S. Kirsch","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06499529246291003,"gpt":0.3576600144154923,"spread":0.2926647219525823,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001598333,0.0000781847,0.0001586239,0.0001050321,0.0001918051,0.0001014549,0.0003192364,0.00006291908,0.0001428169],"category_scores_gemma":[0.00244129,0.00005337605,0.00005716192,0.0002262234,0.0002129549,0.0004421597,0.0001716626,0.0001309252,0.00001961774],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001796779,"about_ca_system_score_gemma":0.0001091477,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008807806,"about_ca_topic_score_gemma":0.0003292356,"domain_scores_codex":[0.9984235,0.00005460923,0.0007264224,0.0003221194,0.0001762221,0.0002971227],"domain_scores_gemma":[0.9978065,0.0003959437,0.0003773891,0.0006161075,0.0007268909,0.00007719605],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005314968,0.00006107199,0.542774,0.000006695924,0.0001074238,0.000006419707,0.0001380258,0.000001036605,0.0003869871,0.4521174,0.001101049,0.003246723],"study_design_scores_gemma":[0.0002082923,0.00004765026,0.5783812,0.00001963479,0.000001406614,0.000003847319,0.0001329107,0.00006857082,0.0002193995,0.2901357,0.1307195,0.00006178291],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9601452,0.0002149859,0.00252527,0.0116213,0.0007029309,0.0002857095,0.004980517,0.00004762433,0.01947639],"genre_scores_gemma":[0.9881896,0.002849421,0.0004383742,0.00002483465,0.0001186445,0.00009634352,0.0001289058,0.00001930502,0.008134532],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1619817,"threshold_uncertainty_score":0.2922629,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2066488918","doi":"10.1002/j.2333-8504.2000.tb01828.x","title":"ASSESSING TIME TRENDS IN SEX DIFFERENCES IN SWIMMING AND RUNNING","year":2000,"lang":"en","type":"article","venue":"ETS Research Report Series","topic":"Sports Analytics and Performance","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"","keywords":"Measure (data warehouse); Event (particle physics); Statistics; Psychology; Econometrics; Demography; Computer science; Mathematics; Data mining","authors":[{"name":"Howard Wainer","is_ca":false},{"name":"Catherine Njue","is_ca":true},{"name":"Samuel Palmer","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1388809068291307,"gpt":0.3539880512241354,"spread":0.2151071443950047,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002250326,0.0001039603,0.0003414222,0.0007523603,0.0001460671,0.0003245921,0.000113992,0.00007265299,0.00245425],"category_scores_gemma":[0.00006755826,0.0001080881,0.00002878066,0.0006821931,0.0001322332,0.0007674419,0.00005609413,0.0003270693,0.00005237253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006516193,"about_ca_system_score_gemma":0.00003107712,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001527707,"about_ca_topic_score_gemma":0.0003307195,"domain_scores_codex":[0.9984713,0.00001529863,0.0006019433,0.000389777,0.0001020301,0.0004197196],"domain_scores_gemma":[0.999541,0.00004083107,0.0001038606,0.0002269352,0.00002051148,0.00006679275],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001016096,0.00003334177,0.9776309,0.00001542934,0.000006689378,0.0004427144,0.0009375484,0.00004246416,0.000009595589,0.001085707,0.00007170135,0.01971378],"study_design_scores_gemma":[0.0001524505,0.00003712216,0.9684941,0.00005727028,4.997629e-7,0.00006381619,0.0002768944,0.005870818,0.00002327612,0.002530904,0.02233597,0.0001568544],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9292924,0.0009235239,0.00000279756,0.0004567064,0.0000316372,0.00005044745,0.000002771235,0.00001105764,0.06922868],"genre_scores_gemma":[0.9644923,0.0006596765,0.0001125288,0.00001307833,0.00004876229,0.0000118056,0.00001233919,0.00001241192,0.03463707],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03519995,"threshold_uncertainty_score":0.9984577,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}