{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":15,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":15,"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":"7a547a233b1e","filters":{"venue":"International Journal of Assessment Tools in Education"}},"results":[{"id":"W4309232742","doi":"10.21449/ijate.1124382","title":"Automatic story and item generation for reading comprehension assessments with transformers","year":2022,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Topic Modeling","field":"Computer Science","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Concordia University of Edmonton; University of Alberta","funders":"University of Alberta","keywords":"Fluency; Reading comprehension; Computer science; Comprehension; Literacy; Reading (process); Mathematics education; Multimedia; Psychology; Pedagogy; Linguistics","authors":[{"name":"Okan Bulut","is_ca":true},{"name":"Seyma N. Yildirim‐Erbasli","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04583102892320276,"gpt":0.3728767693014102,"spread":0.3270457403782074,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002884775,0.001187671,0.000641768,0.002062967,0.0003214782,0.001798764,0.001286402,0.0007955207,0.008571384],"category_scores_gemma":[0.02511345,0.0004978523,0.0008321128,0.001232827,0.0003329142,0.002896593,0.001690983,0.0008385849,0.003819399],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000520754,"about_ca_system_score_gemma":0.0007168832,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001359273,"about_ca_topic_score_gemma":0.001838058,"domain_scores_codex":[0.997273,0.001388299,0.000310145,0.0005236345,0.0004245114,0.00008028856],"domain_scores_gemma":[0.9872714,0.008587195,0.0005099711,0.001365371,0.002020429,0.0002457183],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001045596,0.0005025618,0.01598691,0.0006811723,0.0001447017,0.0005826881,0.003084545,0.02409746,0.04016727,0.005469442,0.01167971,0.8965579],"study_design_scores_gemma":[0.0002898469,0.0006519863,0.01289856,0.0001193955,0.0001309873,0.0007524031,0.001241709,0.8778957,0.07065073,0.01185296,0.02338322,0.0001324746],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1067284,0.0001275557,0.8482456,0.0001174348,0.0001025213,0.0008557831,0.002273449,0.03776142,0.003787797],"genre_scores_gemma":[0.3964682,0.00009370769,0.5943442,0.00005144345,0.00002250783,0.001046879,0.004533636,0.001364277,0.002075128],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.008571384,"threshold_uncertainty_score":0.02867413,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2945378006","doi":"10.21449/ijate.515085","title":"Explanatory Item Response Models for Polytomous Item Responses","year":2019,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":25,"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":"Polytomous Rasch model; Item response theory; Rating scale; Psychology; Econometrics; Psychometrics; Scale (ratio); Statistics; Clinical psychology; Mathematics; Developmental psychology","authors":[{"name":"Luke Stanke","is_ca":false},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4102124714905701,"gpt":0.560403145971193,"spread":0.150190674480623,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06149807,0.003046072,0.002382534,0.004500817,0.001227441,0.004068957,0.006504051,0.003284614,0.01565661],"category_scores_gemma":[0.1810189,0.001593545,0.004386027,0.008170955,0.002300805,0.005013618,0.00367441,0.005911338,0.005294409],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002728054,"about_ca_system_score_gemma":0.002578959,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004149108,"about_ca_topic_score_gemma":0.003719381,"domain_scores_codex":[0.9266729,0.05862176,0.002721849,0.006046461,0.004838783,0.001098257],"domain_scores_gemma":[0.8457049,0.1175753,0.01074786,0.01578909,0.00967396,0.0005089776],"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":[0.0006413733,0.0007375019,0.05247834,0.002033867,0.002082619,0.0007385229,0.006899258,0.1807721,0.0008056705,0.496372,0.0216287,0.23481],"study_design_scores_gemma":[0.0002252714,0.0003039312,0.01155144,0.0006659207,0.0003893334,0.0005271669,0.0009718718,0.6512028,0.0004324313,0.3148923,0.01864525,0.0001922842],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01666648,0.0004317158,0.9759446,0.0008736208,0.0001569813,0.001442194,0.001722534,0.0006521786,0.002109571],"genre_scores_gemma":[0.3319737,0.001076126,0.6369758,0.0009146166,0.0002456687,0.0137539,0.007107706,0.0004318553,0.007520634],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.06149807,"threshold_uncertainty_score":0.3252367,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4390265350","doi":"10.21449/ijate.1406304","title":"A dialectic on validity: Explanation-focused and the many ways of being human","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"Social Sciences and Humanities Research Council of Canada; Canada Research Chairs; University of Cambridge","keywords":"Test (biology); External validity; Classical test theory; Test validity; Epistemology; Construct validity; Psychology; Variation (astronomy); Dialectic; Predictive validity; Cognitive psychology; Item response theory; Social psychology; Psychometrics","authors":[{"name":"Bruno D. Zumbo","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1025470264963527,"gpt":0.4401993006598252,"spread":0.3376522741634725,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09125129,0.001139564,0.002100781,0.005506187,0.006507259,0.02451015,0.005427922,0.008669615,0.002644651],"category_scores_gemma":[0.116294,0.0008769141,0.00149282,0.003531501,0.1433982,0.03648521,0.01277655,0.02174717,0.0006657321],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009232245,"about_ca_system_score_gemma":0.01222823,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003954532,"about_ca_topic_score_gemma":0.00261889,"domain_scores_codex":[0.8537795,0.1222246,0.003629445,0.003621922,0.01453108,0.002213455],"domain_scores_gemma":[0.7050012,0.2702194,0.004790052,0.009859952,0.008003931,0.002125601],"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":[0.00001446642,0.00001501231,0.0002702187,0.0002216368,0.00002113368,0.00005789311,0.03065726,0.0001375852,0.00004767574,0.9492981,0.007363495,0.01189538],"study_design_scores_gemma":[0.00001412571,0.00002812524,0.0002823086,0.0008999738,0.00001053784,0.0001449073,0.0113277,0.0004436999,0.0001032514,0.9176889,0.069027,0.00002943124],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.009452422,0.06638568,0.1483863,0.6978333,0.008435838,0.0001354608,0.00007761369,0.0002089242,0.0690846],"genre_scores_gemma":[0.699786,0.0411335,0.1142925,0.112,0.01646638,0.0009999733,0.0001496695,0.0007162261,0.01445559],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.09125129,"threshold_uncertainty_score":0.4825887,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4410004709","doi":"10.21449/ijate.1602294","title":"A review of automatic item generation techniques leveraging large language models","year":2025,"lang":"en","type":"review","venue":"International Journal of Assessment Tools in Education","topic":"Topic Modeling","field":"Computer Science","cited_by":15,"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":"Computer science; Natural language processing; Data science","authors":[{"name":"Bin Tan","is_ca":true},{"name":"Nour Armoush","is_ca":true},{"name":"Elisabetta Mazzullo","is_ca":true},{"name":"Okan Bulut","is_ca":true},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07488583176221998,"gpt":0.4522201303716182,"spread":0.3773342986093982,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01755542,0.00225258,0.003062878,0.01065265,0.00052806,0.002730724,0.002612467,0.001297042,0.004696497],"category_scores_gemma":[0.06382826,0.001179525,0.003354053,0.008911084,0.0008310736,0.004513082,0.001342273,0.001481896,0.002666786],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001147437,"about_ca_system_score_gemma":0.004008513,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003400883,"about_ca_topic_score_gemma":0.004948286,"domain_scores_codex":[0.9903853,0.005451645,0.001315812,0.0008204722,0.001901669,0.0001251643],"domain_scores_gemma":[0.9077246,0.0834993,0.002517563,0.001554414,0.004466215,0.0002378525],"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":[0.00007061646,0.00008498618,0.0009808944,0.04985424,0.0005192439,0.00005751894,0.0003002257,0.001150876,0.0003316297,0.00156384,0.005723718,0.9393622],"study_design_scores_gemma":[0.0003185026,0.001398328,0.02667657,0.1894797,0.00941641,0.002734108,0.00172687,0.01924842,0.005946143,0.02628515,0.7161906,0.000579092],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001354571,0.9732526,0.02172538,0.0007793853,0.0002198762,0.0002710902,0.0003009002,0.00032982,0.001766393],"genre_scores_gemma":[0.02001863,0.9077151,0.06840263,0.0007257974,0.0004520812,0.0008182878,0.0009832553,0.0001278529,0.0007563382],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.01755542,"threshold_uncertainty_score":0.092843,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2993603560","doi":"10.21449/ijate.627361","title":"Educational data mining: A tutorial for the rattle package in R","year":2019,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Online Learning and Analytics","field":"Computer Science","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 Alberta","funders":"","keywords":"Computer science; Educational data mining; Context (archaeology); Data mining; Flexibility (engineering); Graphical user interface; Field (mathematics); Big data; Data science; Software; Data stream mining; Random forest; Set (abstract data type); Interface (matter); Machine learning","authors":[{"name":"Okan Bulut","is_ca":true},{"name":"Hatice Çiğdem Yavuz","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04475368149601713,"gpt":0.4248154540721439,"spread":0.3800617725761268,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004827277,0.002705778,0.00183939,0.003189245,0.0005279492,0.002038771,0.003905131,0.001283733,0.1901449],"category_scores_gemma":[0.02251762,0.002344945,0.002652213,0.003031881,0.0005942307,0.003090703,0.002678544,0.00403334,0.1499977],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006020161,"about_ca_system_score_gemma":0.002317394,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001678754,"about_ca_topic_score_gemma":0.002733702,"domain_scores_codex":[0.9977908,0.0008722344,0.000297607,0.000319808,0.0006046951,0.0001147892],"domain_scores_gemma":[0.9878476,0.008898185,0.0006252552,0.001255138,0.001046171,0.0003275867],"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":[0.000141862,0.00009794081,0.0009483308,0.003076867,0.0002209032,0.0003455914,0.0002223656,0.003437931,0.002458181,0.01049447,0.8050081,0.1735475],"study_design_scores_gemma":[0.0002440001,0.00008599144,0.001691414,0.0009005841,0.00007792639,0.0007313197,0.00007537385,0.01175591,0.004063926,0.02941461,0.9508125,0.0001464644],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0008002712,0.002547096,0.6621853,0.00163039,0.0009996203,0.000799458,0.04623238,0.2706384,0.01416703],"genre_scores_gemma":[0.007298978,0.003342995,0.8073483,0.002876644,0.0006321029,0.006042799,0.03813893,0.1148853,0.01943394],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1901449,"threshold_uncertainty_score":0.636098,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4389923528","doi":"10.21449/ijate.1394194","title":"Language models in automated essay scoring: Insights for the Turkish language","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Topic Modeling","field":"Computer Science","cited_by":7,"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":"Turkish; Transformative learning; Computer science; Language model; Artificial intelligence; Natural language processing; Intersection (aeronautics); Transformer; Linguistics; Sociology; Engineering","authors":[{"name":"Tahereh Firoozi","is_ca":true},{"name":"Okan Bulut","is_ca":true},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0407797704534206,"gpt":0.3934933832112348,"spread":0.3527136127578142,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004112748,0.0004509258,0.0003452037,0.001449701,0.0005865339,0.003842706,0.0006441355,0.0005667455,0.001729318],"category_scores_gemma":[0.02496612,0.0002149431,0.0004169466,0.001279537,0.001041621,0.004149931,0.00136502,0.001147993,0.0006754607],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001689797,"about_ca_system_score_gemma":0.001754807,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007036024,"about_ca_topic_score_gemma":0.009395199,"domain_scores_codex":[0.9956204,0.003511669,0.0001372978,0.0002749255,0.0003478659,0.0001077991],"domain_scores_gemma":[0.9829196,0.01380828,0.001023424,0.0006548136,0.001362255,0.0002317357],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005091795,0.0003094597,0.0523949,0.0004619691,0.0001404223,0.0011472,0.01600488,0.1636373,0.006191288,0.339713,0.005529005,0.4139614],"study_design_scores_gemma":[0.00002213781,0.0001222591,0.01040009,0.0001166214,0.00004699943,0.0004193832,0.004952389,0.7994181,0.001692054,0.1688572,0.01387744,0.00007542494],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5608739,0.001829608,0.4081589,0.005322307,0.00009407947,0.000155569,0.0004805939,0.0005325463,0.02255251],"genre_scores_gemma":[0.9667667,0.0002277538,0.03166266,0.00007667572,0.00003046653,0.00004538659,0.0001372546,0.00004370776,0.001009474],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007036024,"threshold_uncertainty_score":0.02175057,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4389915531","doi":"10.21449/ijate.1321061","title":"A data pipeline for e-large-scale assessments: Better automation, quality assurance, and efficiency","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Mental Health Research Topics","field":"Psychology","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Northern Alberta Institute of Technology","funders":"International Atomic Energy Agency","keywords":"Workflow; Computer science; Pipeline (software); Scale (ratio); Quality assurance; Automation; Interface (matter); Quality (philosophy); Data science; Data quality; Risk analysis (engineering); Data mining; Database; Engineering; Operations management; Metric (unit)","authors":[{"name":"R.K. Schwarz","is_ca":false},{"name":"Hatice Cigdem Bulut","is_ca":true},{"name":"Charles ANİFOWOSE","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1654467129443049,"gpt":0.5848930899388985,"spread":0.4194463769945936,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04225607,0.002302454,0.001589155,0.005623942,0.00172232,0.009209383,0.003838272,0.001631485,0.01599478],"category_scores_gemma":[0.1215481,0.002291063,0.00202511,0.005277465,0.001507176,0.01128094,0.007990595,0.004801682,0.01468649],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002145598,"about_ca_system_score_gemma":0.01056926,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006779331,"about_ca_topic_score_gemma":0.003633433,"domain_scores_codex":[0.976834,0.008532993,0.003414843,0.003181518,0.007053092,0.0009835496],"domain_scores_gemma":[0.8532255,0.05763989,0.005471448,0.04790611,0.02960117,0.00615594],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001926159,0.0009685775,0.01268362,0.001589835,0.0001960003,0.0007173375,0.004249562,0.00633816,0.02394388,0.02183838,0.1317552,0.7937933],"study_design_scores_gemma":[0.001488392,0.001234634,0.02716717,0.002078613,0.0002112822,0.001311638,0.002438376,0.1525518,0.07341775,0.1025124,0.6345035,0.001084316],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00822758,0.0003268496,0.8860238,0.003246181,0.000282909,0.003309236,0.006168782,0.088494,0.003920744],"genre_scores_gemma":[0.04205591,0.0003253834,0.933937,0.000795547,0.0001488276,0.002067002,0.01051835,0.00758466,0.002567337],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.04225607,"threshold_uncertainty_score":0.2234741,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4396978444","doi":"10.21449/ijate.1376160","title":"The difference between estimated and perceived item difficulty: An empirical study","year":2024,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":2,"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":"Psychology; Statistics; Empirical research; Econometrics; Mathematics education; Mathematics","authors":[{"name":"Ayfer SAYIN","is_ca":false},{"name":"Okan Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08975046533286718,"gpt":0.5107784354602035,"spread":0.4210279701273363,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01432868,0.0003876636,0.0004572233,0.001698114,0.0003756679,0.001431433,0.0008024081,0.0007078693,0.002675716],"category_scores_gemma":[0.1241156,0.0003500449,0.0005610971,0.001577574,0.00126522,0.002087456,0.0009099299,0.001230655,0.0004585141],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006575521,"about_ca_system_score_gemma":0.0005531189,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001361479,"about_ca_topic_score_gemma":0.001648733,"domain_scores_codex":[0.9867541,0.006271604,0.001614691,0.001511036,0.003450953,0.0003975526],"domain_scores_gemma":[0.7186046,0.2400931,0.01662597,0.006829373,0.01607181,0.001775059],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000151052,0.00041378,0.9835125,0.00009524104,0.0001189512,0.0001330353,0.003422404,0.0003575368,0.0005071454,0.000190852,0.0001427115,0.01095498],"study_design_scores_gemma":[0.00001759589,0.0005880485,0.9911004,0.00005473735,0.00006016729,0.000335425,0.003192706,0.00325902,0.0007626127,0.0001834251,0.000418335,0.00002754265],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9974852,0.00009657616,0.001452317,0.00003407165,0.000005333918,0.00003409375,0.00008361876,0.00000574964,0.0008029802],"genre_scores_gemma":[0.9987751,0.0000426758,0.0009050508,0.00001117262,0.000004138855,0.00003577165,0.0000808625,0.000004804426,0.0001403352],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01432868,"threshold_uncertainty_score":0.07577825,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4390145121","doi":"10.21449/ijate.1266808","title":"The complexity of the grading system in Turkish higher education","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Educational Technology and Assessment","field":"Computer Science","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":"Grading (engineering); Higher education; Concordance; Turkish; Mathematics education; Quarter (Canadian coin); Psychology; Medical education; Medicine; Engineering; Political science","authors":[{"name":"Recep GÜR","is_ca":false},{"name":"Mustafa Köroğlu","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06223638291745076,"gpt":0.3912900597447815,"spread":0.3290536768273308,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01149052,0.0003214653,0.0003847693,0.003074632,0.001403294,0.003557156,0.001097151,0.0005266846,0.001428527],"category_scores_gemma":[0.02881897,0.0002603254,0.0003972686,0.004258655,0.001669326,0.001514478,0.001147055,0.000715077,0.0004953225],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003412779,"about_ca_system_score_gemma":0.00282961,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007649116,"about_ca_topic_score_gemma":0.006967196,"domain_scores_codex":[0.9757549,0.01332741,0.00304031,0.001721396,0.005411278,0.0007446762],"domain_scores_gemma":[0.9835525,0.00605137,0.002452344,0.001656604,0.005859442,0.0004277199],"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":[0.0007746112,0.0002516897,0.1530257,0.0007592243,0.0001147806,0.0005906747,0.008961873,0.01499191,0.007504901,0.0462469,0.01228703,0.7544906],"study_design_scores_gemma":[0.0001451207,0.001710491,0.7139331,0.0006329854,0.0001878399,0.00273072,0.01358576,0.07224349,0.009631623,0.07773437,0.1068141,0.0006503363],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8773587,0.001916906,0.07674024,0.001815806,0.0006482208,0.0005631861,0.0005478867,0.0003797202,0.04002926],"genre_scores_gemma":[0.9692351,0.0003821171,0.02809136,0.0001020831,0.0000743409,0.0001617296,0.0002916029,0.00003921774,0.001622451],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01149052,"threshold_uncertainty_score":0.06076843,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3121031673","doi":"10.21449/ijate.705426","title":"Examining the Measurement Invariance of TIMSS 2015 Mathematics Liking Scale through Different Methods","year":2021,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Rasch model; Measurement invariance; Mathematics; Statistics; Metric (unit); Homogeneity (statistics); Scale (ratio); Level of measurement; Descriptive statistics; Polytomous Rasch model; Scale invariance; Econometrics; Psychology; Structural equation modeling; Mathematics education; Social psychology; Item response theory; Confirmatory factor analysis; Psychometrics; Geography; Engineering","authors":[{"name":"Zafer Ertürk","is_ca":false},{"name":"Esra Oyar","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6714988377706251,"gpt":0.597784172952272,"spread":0.07371466481835309,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.024915,0.0005552142,0.0007936482,0.00238937,0.00115813,0.00190502,0.000760961,0.0004002158,0.00379181],"category_scores_gemma":[0.06288321,0.000325983,0.002469585,0.0025442,0.001334348,0.001532002,0.001457516,0.001257666,0.0004957932],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001223998,"about_ca_system_score_gemma":0.002787527,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005203453,"about_ca_topic_score_gemma":0.008446845,"domain_scores_codex":[0.9728381,0.01227976,0.002600961,0.002711243,0.008866199,0.0007036687],"domain_scores_gemma":[0.9642683,0.01956112,0.004467256,0.00418342,0.007098218,0.0004216311],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004335512,0.0004077037,0.6766713,0.0009792821,0.001443489,0.0001639008,0.01053935,0.001934449,0.005945752,0.01241531,0.002713714,0.2863522],"study_design_scores_gemma":[0.00009685137,0.001164288,0.9501399,0.0008020418,0.0006359096,0.0002372084,0.009455514,0.01059544,0.006871394,0.008523124,0.01133655,0.0001417468],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8627525,0.001229221,0.1128923,0.0007213764,0.0004307055,0.001407861,0.0009361539,0.0002067428,0.01942302],"genre_scores_gemma":[0.940789,0.0003722219,0.05422014,0.0001315547,0.00005697852,0.002236532,0.0009973142,0.00007446199,0.001121791],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.975085,"threshold_uncertainty_score":0.1317647,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4379929879","doi":"10.21449/ijate.1249297","title":"Automatic item generation for online measurement and evaluation: Turkish literature items","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Educational Technology and Assessment","field":"Computer Science","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 Alberta","funders":"","keywords":"Turkish; Computer science; Field (mathematics); Test (biology); Item bank; Item analysis; Subject-matter expert; Subject matter; Item response theory; Data science; Artificial intelligence; Statistics; Psychometrics; Curriculum; Expert system; Psychology; Mathematics","authors":[{"name":"Ayfer SAYIN","is_ca":false},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1047808688677508,"gpt":0.4375316124065066,"spread":0.3327507435387558,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01552993,0.001246164,0.001001037,0.006131188,0.0009967891,0.001400196,0.0013269,0.0006084226,0.006713116],"category_scores_gemma":[0.05633269,0.0005881862,0.001299735,0.005073262,0.0006764896,0.001537247,0.001756765,0.0007798601,0.003360545],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001402663,"about_ca_system_score_gemma":0.002453501,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001957287,"about_ca_topic_score_gemma":0.001994248,"domain_scores_codex":[0.979719,0.01119844,0.003979807,0.001340062,0.003200485,0.0005621973],"domain_scores_gemma":[0.9613953,0.01355043,0.001960424,0.004185921,0.01860139,0.0003063599],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001337241,0.001071573,0.04157918,0.002275782,0.0001766668,0.0003749204,0.004179399,0.003309284,0.01496092,0.004148552,0.01444819,0.9121382],"study_design_scores_gemma":[0.002223251,0.008177832,0.5890786,0.003225258,0.001369474,0.002666665,0.01358815,0.1040495,0.1049058,0.0200555,0.1497973,0.0008626974],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5767459,0.001414564,0.363019,0.0009154749,0.0006681514,0.02372806,0.005143011,0.003812202,0.02455371],"genre_scores_gemma":[0.5233265,0.0005589654,0.4433104,0.0001639862,0.00008130272,0.02192322,0.006632031,0.0003631559,0.003640528],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01552993,"threshold_uncertainty_score":0.08213109,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4323355119","doi":"10.21449/ijate.1212539","title":"The bibliometric journey of IJATE from local to global","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Publish or perish; Citation impact; Citation; Library science; Bibliometrics; Web of science; Impact factor; Publication; Visibility; Scopus; Citation analysis; Computer science; Political science; Publishing; Geography; MEDLINE; Law","authors":[{"name":"Orhan Karamustafaoğlu","is_ca":false},{"name":"Metin Orbay","is_ca":false},{"name":"İzzet Kara","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1766345083418794,"gpt":0.5757474426469738,"spread":0.3991129343050943,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["bibliometrics"],"consensus_categories":[],"category_scores_codex":[0.008871812,0.0003631259,0.001093431,0.0409553,0.002558774,0.01456175,0.0006479881,0.0007611114,0.004595672],"category_scores_gemma":[0.05064604,0.0002374186,0.0006767903,0.09841547,0.001259978,0.009318179,0.004084912,0.0009440653,0.002170297],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004196491,"about_ca_system_score_gemma":0.004875043,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005457124,"about_ca_topic_score_gemma":0.01040458,"domain_scores_codex":[0.9878632,0.002758944,0.001667069,0.000994038,0.006117258,0.0005995382],"domain_scores_gemma":[0.9500969,0.01480798,0.01068612,0.00284938,0.01895005,0.002609559],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001925579,0.0001648808,0.4864288,0.0025149,0.0004102082,0.0007302598,0.0492209,0.001061115,0.001579376,0.02234644,0.02895682,0.4063938],"study_design_scores_gemma":[0.00001479328,0.000223681,0.6846687,0.001467049,0.0002333008,0.001245998,0.07217387,0.002109807,0.002590845,0.007007273,0.2281095,0.0001550722],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8334666,0.0144432,0.00494536,0.017278,0.0004849661,0.0002187986,0.009282858,0.0005376303,0.1193426],"genre_scores_gemma":[0.9646001,0.01063204,0.007708157,0.0007958189,0.0005210853,0.000249098,0.004763872,0.0002085546,0.01052121],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9590447,"threshold_uncertainty_score":0.04691917,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4414775956","doi":"10.21449/ijate.1566093","title":"Construction and validation of a multilingual diagnostic instrument for neuromyths and their origins","year":2025,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Neuroscience, Education and Cognitive Function","field":"Neuroscience","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université du Québec à Montréal","funders":"Türkiye Bilimsel ve Teknolojik Araştırma Kurumu","keywords":"Relevance (law); Identification (biology); Robustness (evolution); Process (computing); Adaptation (eye); Key (lock); Qualitative research","authors":[{"name":"Oktay Cem Adıgüzel","is_ca":false},{"name":"Patrice Potvin","is_ca":true},{"name":"Sibel Küçükkayhan","is_ca":false},{"name":"Derya Atik Kara","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03563158567407632,"gpt":0.3779109001011886,"spread":0.3422793144271123,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02416455,0.0005353323,0.0006838565,0.005470821,0.001407377,0.001736857,0.00122808,0.0006811692,0.002388424],"category_scores_gemma":[0.06102411,0.0003377499,0.0009028617,0.001928817,0.001247624,0.00212626,0.004052168,0.001357756,0.00066677],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002341423,"about_ca_system_score_gemma":0.01001676,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003137859,"about_ca_topic_score_gemma":0.007096056,"domain_scores_codex":[0.9879345,0.005323735,0.002385372,0.0008440267,0.003019657,0.0004927706],"domain_scores_gemma":[0.9580791,0.01871695,0.003861689,0.002258309,0.01569758,0.001386347],"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":[0.0003569426,0.001018363,0.3398156,0.002058208,0.0001271031,0.0008529778,0.0572585,0.001104657,0.007921576,0.00863375,0.003968401,0.5768839],"study_design_scores_gemma":[0.0002941736,0.002268155,0.7964248,0.004424757,0.000324771,0.002695286,0.08894416,0.01024936,0.0151773,0.01232669,0.06646048,0.0004100282],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8628656,0.001181428,0.1076979,0.001611628,0.0003370359,0.008433428,0.001665863,0.0004872796,0.01571975],"genre_scores_gemma":[0.7347283,0.001027978,0.2446032,0.0004208112,0.0000644308,0.0147577,0.001576245,0.00007768555,0.002743652],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02416455,"threshold_uncertainty_score":0.1277959,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4401920683","doi":"10.21449/ijate.1475980","title":"The mental imagery scale for art students: Building and validating a short form","year":2024,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Creativity in Education and Neuroscience","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Northern Alberta Institute of Technology","funders":"","keywords":"Creativity; Scale (ratio); Psychology; Reliability (semiconductor); Construct (python library); Cognition; Mental image; Content validity; Construct validity; Visual arts education; Criterion validity; Computer science; Cognitive psychology; Applied psychology; Mathematics education; Psychometrics; Social psychology; Clinical psychology","authors":[{"name":"Handan Narin","is_ca":false},{"name":"Hatice Cigdem Bulut","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04279848651872874,"gpt":0.5057800219547152,"spread":0.4629815354359864,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002828846,0.0004233523,0.0005116469,0.001646597,0.0003965742,0.0007879715,0.000646801,0.0005240918,0.002403019],"category_scores_gemma":[0.01425968,0.0002585997,0.0008126782,0.0009583393,0.0004164514,0.0008095129,0.001131991,0.001185714,0.0006855871],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004013067,"about_ca_system_score_gemma":0.0009670988,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006106712,"about_ca_topic_score_gemma":0.001572448,"domain_scores_codex":[0.9984304,0.0003683901,0.000361361,0.0001103601,0.0006513327,0.00007817883],"domain_scores_gemma":[0.9932384,0.002758207,0.001511396,0.0004678948,0.001583183,0.0004407899],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0007651933,0.002202191,0.5382589,0.0008707931,0.0002836642,0.0004604468,0.005701908,0.002811637,0.01086735,0.002992705,0.009628919,0.4251563],"study_design_scores_gemma":[0.0002387444,0.001806894,0.9667826,0.0002236682,0.00008154535,0.001113638,0.002312057,0.007012652,0.002770819,0.002537004,0.01504708,0.00007340275],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9768652,0.0004528481,0.01115434,0.0003718846,0.0001232896,0.003854108,0.001307809,0.0001126977,0.005757862],"genre_scores_gemma":[0.8964045,0.0008817783,0.08454951,0.000296932,0.00007427867,0.01118333,0.003298364,0.00004497755,0.003266238],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.002828846,"threshold_uncertainty_score":0.01496059,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4389925007","doi":"10.21449/ijate.1359348","title":"Automatic item generation for non-verbal reasoning items","year":2023,"lang":"en","type":"article","venue":"International Journal of Assessment Tools in Education","topic":"Educational Assessment and Pedagogy","field":"Social Sciences","cited_by":0,"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":"Verbal reasoning; Computer science; Test (biology); Item analysis; Cognition; Item response theory; Homogeneous; Subject-matter expert; Natural language processing; Artificial intelligence; Subject matter; Psychology; Cognitive psychology; Psychometrics; Pedagogy; Expert system; Mathematics; Developmental psychology","authors":[{"name":"Ayfer SAYIN","is_ca":false},{"name":"Sabiha BOZDAĞ","is_ca":false},{"name":"Mark J. Gierl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08817372903914804,"gpt":0.4972479358557301,"spread":0.4090742068165821,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02336156,0.001758417,0.001543125,0.004638094,0.0005493676,0.001595341,0.002544604,0.001026058,0.005750282],"category_scores_gemma":[0.09864897,0.0006700027,0.001638967,0.003432713,0.0006237978,0.001540012,0.001850274,0.0012339,0.004555668],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007100459,"about_ca_system_score_gemma":0.001512976,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004610643,"about_ca_topic_score_gemma":0.0006673317,"domain_scores_codex":[0.9789555,0.01287478,0.002438391,0.001753783,0.003670365,0.0003072084],"domain_scores_gemma":[0.8852344,0.07012389,0.005273379,0.01595028,0.02286717,0.00055089],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00115536,0.001199746,0.03443938,0.001251562,0.0002016239,0.0003929079,0.002044761,0.005760952,0.01713522,0.003500544,0.01083002,0.9220879],"study_design_scores_gemma":[0.003003121,0.009654179,0.2186646,0.001933947,0.001087248,0.005309829,0.006242669,0.4304003,0.1903456,0.04424092,0.08819952,0.0009181393],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1955926,0.0003296204,0.7729322,0.0002582302,0.000338698,0.01394805,0.002893276,0.008202627,0.005504637],"genre_scores_gemma":[0.1867292,0.000174963,0.7894253,0.0001523482,0.0000638622,0.01415372,0.006707145,0.0006653521,0.001928042],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.02336156,"threshold_uncertainty_score":0.1235492,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}