{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":18,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":18,"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":"83ada4cbf0c3","filters":{"venue":"Studies In Educational Evaluation"}},"results":[{"id":"W2522686050","doi":"10.1016/j.stueduc.2016.08.007","title":"Meta-analysis of faculty's teaching effectiveness: Student evaluation of teaching ratings and student learning are not related","year":2016,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":584,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Mount Royal University","funders":"","keywords":"Psychology; Set (abstract data type); Meta-analysis; Student achievement; Sample (material); Mathematics education; Artifact (error); Correlation; Academic achievement; Computer science; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.4547744883637743,"gpt":0.6028893736354598,"spread":0.1481148852716855,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09268692,0.0002089714,0.0008769293,0.0007324184,0.0007431494,0.00004090412,0.0002553268,0.00009727685,0.0003099543],"category_scores_gemma":[0.04771985,0.0001601653,0.000286031,0.0005909876,0.0003777628,0.0007945515,0.000127663,0.0003362298,0.000004038066],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001150813,"about_ca_system_score_gemma":0.0005705137,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007520456,"about_ca_topic_score_gemma":0.001020639,"domain_scores_codex":[0.9763434,0.01726751,0.001105494,0.0005592721,0.004491648,0.0002326921],"domain_scores_gemma":[0.9865518,0.008753964,0.001846317,0.0002371685,0.002559078,0.00005164378],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005702274,0.0009973614,0.5414988,0.00008726221,0.0679618,1.627083e-7,0.297083,0.04020324,0.001776966,0.03294056,0.00006437979,0.01732937],"study_design_scores_gemma":[0.00103107,0.00007732966,0.8252358,0.0001478536,0.08254518,3.009152e-7,0.08416113,0.001747693,0.00007442099,0.004695204,0.00007860357,0.0002053968],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9863142,0.002505416,0.00007177076,0.007012738,0.0003367933,0.001244525,0.0000109117,0.0000218712,0.002481813],"genre_scores_gemma":[0.9981865,0.0001230629,0.0008152752,0.00002729689,0.00006460694,0.0004634708,0.00002934962,0.00001310642,0.0002772835],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.283737,"threshold_uncertainty_score":0.9603016,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2782375796","doi":"10.1016/j.stueduc.2017.12.008","title":"Re-conceptualizing classroom assessment fairness: A systematic meta-ethnography of assessment literature and beyond","year":2018,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":82,"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":"Conceptualization; Accountability; Construct (python library); Educational assessment; Psychology; Ethnography; Standards-based assessment; Process (computing); Pedagogy; Alternative assessment; Mathematics education; Sociology; Computer science; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.1824586292219303,"gpt":0.5132713437504184,"spread":0.330812714528488,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006089516,0.0001811818,0.0005254899,0.0003139467,0.0004747396,0.00009028619,0.0001976261,0.00008745705,0.0002659176],"category_scores_gemma":[0.0008890752,0.0001514035,0.0001183353,0.000955768,0.0008144926,0.0004169374,0.00008448522,0.0001396649,0.000003137873],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000424731,"about_ca_system_score_gemma":0.0006364452,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008026866,"about_ca_topic_score_gemma":0.001054059,"domain_scores_codex":[0.9961613,0.001157314,0.0006193485,0.0003465121,0.001476608,0.0002389314],"domain_scores_gemma":[0.9966658,0.001143116,0.0004174353,0.0002019972,0.001520175,0.00005148473],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001978682,0.0006333817,0.1340922,0.003170358,0.003342903,5.696983e-7,0.5813432,0.00002424203,0.0001119724,0.2698024,0.006290936,0.001168078],"study_design_scores_gemma":[0.001213354,0.0003106268,0.1888143,0.00246649,0.002823114,7.778413e-7,0.6834535,0.0007619327,0.00001496803,0.1191397,0.0005362915,0.0004650339],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6998417,0.05310109,0.0004897928,0.01513653,0.006444265,0.007136467,0.0000370961,0.00008246341,0.2177307],"genre_scores_gemma":[0.9896287,0.0008079219,0.00789719,0.000137214,0.0004133863,0.0008193729,0.00002981615,0.000009182502,0.0002572415],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.289787,"threshold_uncertainty_score":0.6174057,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1976592601","doi":"10.1016/j.stueduc.2013.10.006","title":"Towards a culture of inquiry for data use in schools: Breaking down professional learning barriers through intentional interruption","year":2013,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":78,"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":"","keywords":"Status quo; Context (archaeology); Interrupt; Psychology; Cognition; Professional development; Professional learning community; Social psychology; Pedagogy; Computer science; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.6261356046479121,"gpt":0.6152237617204723,"spread":0.01091184292743974,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006410231,0.0001816397,0.0003085441,0.0004248333,0.0001988042,0.000131089,0.0006383477,0.00008579212,0.0008905276],"category_scores_gemma":[0.03266252,0.0001424045,0.00007444229,0.0008044334,0.0001898691,0.002994863,0.0003768601,0.0002820936,0.00003279109],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006380085,"about_ca_system_score_gemma":0.001559298,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000301485,"about_ca_topic_score_gemma":0.0002597026,"domain_scores_codex":[0.9953836,0.0005143115,0.001225193,0.0006690272,0.001963032,0.0002448594],"domain_scores_gemma":[0.9938967,0.001626312,0.0006183388,0.0004212922,0.003382054,0.00005535666],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002282535,0.001393693,0.6410271,0.0001699273,0.0002892593,2.102071e-7,0.09045049,0.004427765,0.002267256,0.03118856,0.1548406,0.07371683],"study_design_scores_gemma":[0.001144853,0.000116368,0.5327826,0.0004131242,0.00003397467,0.00000214785,0.1774884,0.02236508,0.00004878639,0.2629205,0.002454131,0.000230111],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9795668,0.0008599313,0.002061425,0.01014537,0.004626671,0.002169377,0.00004713238,0.00001100515,0.0005122518],"genre_scores_gemma":[0.9819555,0.00006089274,0.01391618,0.0002681821,0.0004000742,0.001447568,0.0008185916,0.00001074125,0.001122236],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2317319,"threshold_uncertainty_score":0.9754857,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3128610517","doi":"10.1016/j.stueduc.2021.100977","title":"The long-term washback effects of the National Matriculation English Test on college English learning in China: Tertiary student perspectives","year":2021,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":31,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Matriculation; Active listening; Competence (human resources); Psychology; Test (biology); College English; Curriculum; Perception; Mathematics education; China; Medical education; Pedagogy; Medicine; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.03124067406168235,"gpt":0.419786253443572,"spread":0.3885455793818897,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002655667,0.0001183874,0.0001480522,0.0001058725,0.000732247,0.00005354341,0.0002329547,0.00005561558,0.00005094493],"category_scores_gemma":[0.02011385,0.00008689506,0.00007141533,0.000895014,0.000279045,0.0001947896,0.0001046705,0.0002032699,0.000003099964],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001445752,"about_ca_system_score_gemma":0.0009609932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002357298,"about_ca_topic_score_gemma":0.001280747,"domain_scores_codex":[0.9964652,0.0009977238,0.0003413808,0.0002683213,0.00173321,0.0001941352],"domain_scores_gemma":[0.9937923,0.003909969,0.000212659,0.0001067447,0.001957797,0.00002055175],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000007635882,0.0005537834,0.8097692,0.00002243506,0.00005447925,3.622915e-7,0.1735791,0.0009082835,0.00002818337,0.01403324,0.0006984872,0.0003448444],"study_design_scores_gemma":[0.0005483953,0.00002355031,0.8842094,0.0001373982,0.00002734641,1.344546e-7,0.1128406,0.00006991737,0.00004387378,0.001929078,0.00008752631,0.00008283731],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9809753,0.002069847,5.631072e-7,0.003772078,0.002308043,0.0009357651,0.000002761888,0.00001264553,0.009922994],"genre_scores_gemma":[0.9968901,0.0007253176,0.00002409624,0.00004292892,0.0009575095,0.0003014081,0.00001440337,0.000007443113,0.001036852],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07444018,"threshold_uncertainty_score":0.9881402,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2072591409","doi":"10.1016/s0191-491x(00)00011-0","title":"The Brazilian National Evaluation System of Basic Education: Context, process, and impact","year":2000,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"School Choice and Performance","field":"Social Sciences","cited_by":23,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université de Montréal","funders":"","keywords":"Context (archaeology); Process (computing); Process management; Mathematics education; Computer science; Psychology; Business; Geography","retraction":null,"screen_n_in":null,"score":{"opus":0.08028795285571905,"gpt":0.492372740274014,"spread":0.4120847874182949,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006582031,0.00009567924,0.0001257835,0.00009996874,0.0007661044,0.00004311547,0.0001465415,0.00004874146,0.0005977995],"category_scores_gemma":[0.002303023,0.00007492094,0.00003434894,0.0005070596,0.0003989842,0.0004356297,0.000008840138,0.00008665443,0.00002707194],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001076747,"about_ca_system_score_gemma":0.005958462,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004668137,"about_ca_topic_score_gemma":0.003173562,"domain_scores_codex":[0.9972491,0.0004905747,0.0003573762,0.0001971799,0.001542566,0.0001632144],"domain_scores_gemma":[0.9963559,0.0006660791,0.0001715706,0.0001126473,0.002649305,0.00004451728],"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":[0.0001525325,0.0003709833,0.1415307,0.0002009203,0.0001857117,3.946475e-8,0.08436802,0.002574185,0.0000129611,0.05047239,0.03257105,0.6875605],"study_design_scores_gemma":[0.001019039,0.00007297371,0.7868946,0.0004333956,0.0001262999,0.000005034957,0.1362868,0.005371088,0.00001922018,0.05404171,0.01546057,0.0002692749],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9259308,0.01264988,3.492613e-7,0.006432097,0.001057444,0.0009238896,0.000007202662,0.00001064767,0.05298766],"genre_scores_gemma":[0.9965025,0.0009493207,0.00002795101,0.00009409311,0.001094086,0.0006169604,0.0000331487,0.000005219897,0.0006767261],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6872912,"threshold_uncertainty_score":0.9996768,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2054415360","doi":"10.1016/j.stueduc.2009.12.005","title":"Evaluation for learning: A cross-case analysis of evaluator strategies","year":2009,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; Queen's University","funders":"","keywords":"Process (computing); Citizen journalism; Participatory evaluation; Reflection (computer programming); Program evaluation; Knowledge management; Computer science; Process management; Psychology; Sociology; Political science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.5328970254040822,"gpt":0.6892803130688502,"spread":0.1563832876647679,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05653324,0.0001989881,0.0005395718,0.001579867,0.000328773,0.0001655678,0.0003002257,0.00008562938,0.00214874],"category_scores_gemma":[0.03172893,0.0001668821,0.0002547673,0.003418691,0.0001882419,0.0009230479,0.00003422657,0.0001304046,0.000026486],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007306662,"about_ca_system_score_gemma":0.002908047,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001773224,"about_ca_topic_score_gemma":0.0003115428,"domain_scores_codex":[0.9903953,0.001414824,0.001568703,0.000627549,0.005737365,0.0002562412],"domain_scores_gemma":[0.9808043,0.003431085,0.0008838341,0.0004614738,0.01436255,0.00005672436],"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":[0.0001131356,0.0004764412,0.04781553,0.00001673139,0.0007925123,5.181113e-7,0.01177567,0.7467484,0.0002182783,0.01626085,0.001162219,0.1746197],"study_design_scores_gemma":[0.0009599401,0.0001968297,0.3201584,0.00001596046,0.001175001,0.000004327582,0.01445136,0.5031605,0.00003943631,0.1594791,0.0002264353,0.000132804],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9862316,0.001712077,0.001455247,0.002458384,0.0008283548,0.001668887,0.00002058414,0.00001381707,0.005611094],"genre_scores_gemma":[0.9965371,0.00005417718,0.001888819,0.0001228622,0.0001705878,0.0007142198,0.0001598917,0.000006811762,0.0003455035],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2723428,"threshold_uncertainty_score":0.9987634,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4385393971","doi":"10.1016/j.stueduc.2023.101290","title":"Evaluating student engagement and experiential learning in global classrooms: A qualitative case study","year":2023,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Education and Critical Thinking Development","field":"Social Sciences","cited_by":18,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; University of Toronto","funders":"","keywords":"Experiential learning; Student engagement; Psychology; Interactivity; Pedagogy; Mathematics education; Experiential education; Educational technology; Qualitative research; Sociology; Computer science; Multimedia; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.4136159316534047,"gpt":0.6442659677685991,"spread":0.2306500361151944,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01846163,0.0001140791,0.0001679062,0.0002024642,0.0007588324,0.00006580442,0.00008912475,0.00003780638,0.0001987407],"category_scores_gemma":[0.008501549,0.0001221037,0.00002015776,0.001089416,0.0002342089,0.0001676673,0.0001200478,0.0001765211,0.0000269241],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001523129,"about_ca_system_score_gemma":0.0009952474,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009062907,"about_ca_topic_score_gemma":0.006427837,"domain_scores_codex":[0.9936948,0.003909855,0.0004577198,0.0003782012,0.001274404,0.0002849525],"domain_scores_gemma":[0.9978409,0.001425563,0.00008889932,0.00008218075,0.000502387,0.00006004961],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.000006743495,0.0004210449,0.07959288,0.00001203428,0.00003797318,0.00001173324,0.8964037,0.0003976556,0.000001236167,0.01072616,0.0001410134,0.01224779],"study_design_scores_gemma":[0.0005990964,0.0000709852,0.1035535,0.00004978144,0.00002023199,0.000002886363,0.8735084,0.0003925677,2.036641e-7,0.02146497,0.0002240765,0.000113277],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9900445,0.000446342,0.000006828177,0.005050827,0.001249816,0.001149413,9.506362e-7,0.00003863247,0.00201263],"genre_scores_gemma":[0.997037,0.0001036917,0.0006672161,0.00009599055,0.0001791726,0.001446272,0.00001346629,0.000005752702,0.0004514621],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02396063,"threshold_uncertainty_score":0.9998503,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4210694249","doi":"10.1016/j.stueduc.2022.101126","title":"Item wording effects in self-report measures and reading achievement: Does removing careless respondents help?","year":2022,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Reading (process); Psychology; Mathematics education; Academic achievement; Achievement test; Standardized test; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.4604167025598831,"gpt":0.5511538330204064,"spread":0.09073713046052334,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04221355,0.0001458159,0.0003341896,0.001367512,0.0005501937,0.00008563319,0.000336154,0.00003323363,0.00005419972],"category_scores_gemma":[0.2109911,0.0001149442,0.0000402389,0.002773974,0.00006125894,0.000255212,0.0004589424,0.0002456495,0.000002149585],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001014748,"about_ca_system_score_gemma":0.0002736633,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008786227,"about_ca_topic_score_gemma":0.0001181409,"domain_scores_codex":[0.992645,0.002755892,0.000888682,0.0006861688,0.002749058,0.0002752471],"domain_scores_gemma":[0.953804,0.04461826,0.0005475607,0.0003700175,0.0006113136,0.00004891836],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004649122,0.000104547,0.9148682,0.00002872001,0.00004054688,0.00001090782,0.01004677,0.001707857,0.0001991311,0.0007059223,0.0004257289,0.07181523],"study_design_scores_gemma":[0.0005426827,0.00004821664,0.8817748,0.00009382405,0.00002191833,0.00002775789,0.03456986,0.002609948,0.000029232,0.07962693,0.00050307,0.0001517947],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9897216,0.003506068,0.0001373707,0.001666042,0.003201545,0.0005882257,0.000002351304,0.0000201748,0.001156602],"genre_scores_gemma":[0.9926624,0.00008767609,0.006230278,0.00009783083,0.000155506,0.0004476151,0.000007983026,0.00000931759,0.0003013858],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1687776,"threshold_uncertainty_score":0.9862427,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089211529","doi":"10.1016/j.stueduc.2007.04.002","title":"ANTIBULLYING PROGRAMS: A SURVEY OF EVALUATION ACTIVITIES IN PUBLIC SCHOOLS","year":2007,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Bullying, Victimization, and Aggression","field":"Psychology","cited_by":15,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"Social Sciences and Humanities Research Council of Canada; Ministère de l’Éducation, Gouvernement de l’Ontario","keywords":"Rigour; Psychological intervention; Program evaluation; Medical education; Poison control; Human factors and ergonomics; Suicide prevention; Medicine; Injury prevention; Psychology; Applied psychology; Nursing; Environmental health; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.2152842294067903,"gpt":0.4833329445328567,"spread":0.2680487151260665,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0169335,0.000171004,0.0002813429,0.0007370276,0.0001338815,0.00002462608,0.0001290212,0.0001244799,0.0005889743],"category_scores_gemma":[0.004802944,0.0001715758,0.00003748701,0.001381182,0.0001537998,0.0003537173,0.00004468393,0.0002098841,0.00001829794],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006870356,"about_ca_system_score_gemma":0.0007984756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008003027,"about_ca_topic_score_gemma":0.005237509,"domain_scores_codex":[0.9961162,0.001246727,0.0008041022,0.0004142516,0.00106719,0.0003514927],"domain_scores_gemma":[0.99659,0.0008947224,0.0004716663,0.0002628308,0.001738111,0.00004271468],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009920466,0.0008199416,0.9195666,0.00004265596,0.00007735344,4.385394e-7,0.01203707,0.0004382779,0.00009779746,0.002087371,0.000675038,0.0640582],"study_design_scores_gemma":[0.001222032,0.00005541112,0.9756876,0.0002778188,0.00003188814,0.000003450795,0.01660558,0.001439264,0.00004043442,0.00441672,0.000059987,0.0001598046],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.987801,0.003826321,0.0002213968,0.0005619457,0.00196194,0.001363913,0.000006280518,0.00001773659,0.00423945],"genre_scores_gemma":[0.9979821,0.0001266331,0.0006389584,0.00004426678,0.0001671772,0.0005142262,0.000386704,0.00001780109,0.0001220933],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0638984,"threshold_uncertainty_score":0.6996658,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4402569962","doi":"10.1016/j.stueduc.2024.101403","title":"Students’ perceptions of online peer feedback in process-oriented L2 writing: A qualitative inquiry","year":2024,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":12,"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":"Peer feedback; Perception; Mathematics education; Peer evaluation; Process (computing); Qualitative research; Psychology; Computer science; Pedagogy; Higher education; Sociology; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.230095691557151,"gpt":0.6111867833582557,"spread":0.3810910918011047,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005441125,0.0001223373,0.0002127348,0.0003920129,0.0001768961,0.00004403782,0.0002044106,0.00006280413,0.0004627903],"category_scores_gemma":[0.002770231,0.0001232184,0.00005414933,0.001374942,0.0004947422,0.0003749148,0.00006010638,0.0001769045,0.00003153294],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008463389,"about_ca_system_score_gemma":0.001124429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002086804,"about_ca_topic_score_gemma":0.002836022,"domain_scores_codex":[0.9962819,0.0006463556,0.0005465794,0.0003175082,0.001984732,0.0002229197],"domain_scores_gemma":[0.9973724,0.0009835773,0.0001081029,0.0000931884,0.001408244,0.0000344532],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00001406997,0.0006671702,0.1806117,0.00008253728,0.00007344085,3.280683e-7,0.8027387,0.0001099468,0.00002114655,0.01257349,0.001842303,0.001265141],"study_design_scores_gemma":[0.0002819034,0.00002361977,0.3954552,0.0002417237,0.00002926004,1.213977e-7,0.5918875,0.000146612,9.921818e-7,0.01163982,0.0002100108,0.00008323164],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9827357,0.001709014,0.00001407793,0.009389365,0.001544304,0.0007318429,0.00001752447,0.00002208807,0.003836108],"genre_scores_gemma":[0.9966377,0.0003550294,0.0005053199,0.00005231696,0.0006089288,0.0003654509,0.000148094,0.00000995379,0.001317196],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2148435,"threshold_uncertainty_score":0.5067229,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3176398386","doi":"10.1016/j.stueduc.2021.101058","title":"The development and psychometric properties of an educational development impact questionnaire","year":2021,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Evaluation of Teaching Practices","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; Alberta Advanced Education; University of Alberta","funders":"","keywords":"Exploratory factor analysis; Context (archaeology); Psychology; Scholarship; Psychometrics; Medical education; Applied psychology; Mathematics education; Knowledge management; Computer science; Medicine; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.3343641478523127,"gpt":0.5538289167297261,"spread":0.2194647688774134,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008577959,0.0001032622,0.0001261115,0.000195653,0.001227289,0.00007416687,0.000150993,0.00004220286,0.00009389518],"category_scores_gemma":[0.01765313,0.00008004971,0.00002172661,0.0006669822,0.000343933,0.0005054767,0.0000468152,0.0001064991,0.00001039098],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001012216,"about_ca_system_score_gemma":0.009209225,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001583722,"about_ca_topic_score_gemma":0.005382599,"domain_scores_codex":[0.9967064,0.001170352,0.0004830884,0.0002469179,0.001221129,0.0001721312],"domain_scores_gemma":[0.9967476,0.0008516349,0.0002809517,0.0001464904,0.001914746,0.00005855086],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00005889375,0.001180668,0.1614004,0.0001147404,0.0003490683,1.21654e-7,0.4025815,0.0007509282,0.0002178811,0.04572196,0.0009563895,0.3866675],"study_design_scores_gemma":[0.0002772074,0.00001922231,0.9387779,0.0001935584,0.00002904845,0.00000243423,0.04331513,0.00013935,0.0003872358,0.01151455,0.005192026,0.000152297],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9627375,0.01562583,0.000008897529,0.01889147,0.0008558395,0.0003506358,5.61299e-7,0.000008514252,0.001520725],"genre_scores_gemma":[0.9920597,0.0004049098,0.006197053,0.00004563406,0.0002053961,0.0003093843,0.00002217735,0.000006438929,0.0007492693],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7773775,"threshold_uncertainty_score":0.9964076,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4404020856","doi":"10.1016/j.stueduc.2024.101412","title":"Predicting the Mathematics Literacy of Resilient Students from High‐performing Economies: A Machine Learning Approach","year":2024,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Online Learning and Analytics","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":"McGill University","funders":"","keywords":"Mathematics education; Literacy; Computer science; Sociology; Pedagogy; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.05050611502294976,"gpt":0.4089772244966975,"spread":0.3584711094737477,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001957469,0.000102969,0.0001569856,0.0001428314,0.000162194,0.0001235625,0.00040895,0.00002175428,0.00001477114],"category_scores_gemma":[0.001054345,0.00007482368,0.00004317222,0.0003697357,0.00006078897,0.0002975823,0.0002075986,0.0002361578,0.00001068664],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002054305,"about_ca_system_score_gemma":0.0002224807,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000079797,"about_ca_topic_score_gemma":0.00001184344,"domain_scores_codex":[0.9984541,0.0001662189,0.0004175967,0.0002720653,0.0005688465,0.0001211707],"domain_scores_gemma":[0.9982306,0.001151381,0.0001496101,0.0002234873,0.0002281462,0.00001674988],"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":[0.000009589779,0.0008532226,0.1671019,0.0006869236,0.0007127994,0.000001283256,0.3835331,0.3381231,0.00005797323,0.05129631,0.0004327348,0.05719112],"study_design_scores_gemma":[0.0001294233,0.00002647635,0.01678946,0.0003079337,0.00004416929,0.000002150494,0.003398539,0.9512556,0.00001268849,0.02783192,0.000129605,0.0000720098],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9851983,0.006817024,0.003266619,0.003253856,0.0008131415,0.0002433029,0.000005228161,0.00003730425,0.0003652298],"genre_scores_gemma":[0.9737172,0.0001617512,0.02540285,0.00002583241,0.0002562448,0.00006838532,0.00004071586,0.000007496782,0.000319484],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6131326,"threshold_uncertainty_score":0.3051221,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4387762598","doi":"10.1016/j.stueduc.2023.101309","title":"Predicting reading comprehension performance based on student characteristics and item properties","year":2023,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Reading and Literacy Development","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":false,"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":"Reading comprehension; Reading (process); Affect (linguistics); Test (biology); Psychology; Multilingualism; Comprehension; Mathematics education; Cognition; Computer science; Cognitive psychology; Linguistics; Pedagogy; Communication","retraction":null,"screen_n_in":null,"score":{"opus":0.1548524743564766,"gpt":0.4384727952419538,"spread":0.2836203208854772,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001338717,0.0001089857,0.0001383054,0.0002458255,0.0001994944,0.00002172224,0.00005683206,0.00003223302,0.00004950771],"category_scores_gemma":[0.0005188575,0.0000939316,0.00001291123,0.0002656982,0.00006100549,0.00006866206,0.00003655904,0.0001005409,0.00008430984],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002318369,"about_ca_system_score_gemma":0.00008675364,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007174378,"about_ca_topic_score_gemma":0.000001964018,"domain_scores_codex":[0.9986784,0.0001619421,0.0003051885,0.0002656367,0.0004206934,0.0001681073],"domain_scores_gemma":[0.9990157,0.0005091775,0.00009852648,0.0001287471,0.0002237388,0.00002413723],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005881385,0.0001586625,0.921593,0.00009146996,0.00005017628,8.027949e-7,0.04995685,0.0009177314,0.00009736781,0.0003183778,0.001394798,0.025362],"study_design_scores_gemma":[0.0003665015,0.00006350585,0.9583269,0.0004408898,0.00001523455,0.00000188112,0.005471965,0.03480263,0.00001448093,0.0000793892,0.0003198937,0.00009674886],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9940193,0.0002625234,0.000001596245,0.001089198,0.001974434,0.0003907902,0.000002570505,0.00003937833,0.002220192],"genre_scores_gemma":[0.9978892,0.00008728525,0.0001092668,0.000160591,0.0002051288,0.0004802535,0.00007674286,0.00001075467,0.000980768],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04448488,"threshold_uncertainty_score":0.3830419,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1982497397","doi":"10.1016/s0191-491x(02)00014-7","title":"Matching the grade 8 TIMSS item pool to the Ontario curriculum","year":2002,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto; Lakehead University","funders":"","keywords":"Mathematics education; Curriculum; Matching (statistics); Secondary education; Psychology; Pedagogy; Mathematics; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.1656116291663467,"gpt":0.4685884905270746,"spread":0.3029768613607279,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002452534,0.00008436642,0.00008634964,0.00005343385,0.001055732,0.00007504377,0.0003132552,0.00002526144,0.001230334],"category_scores_gemma":[0.0006635452,0.00005228855,0.00003946322,0.0004400446,0.0001416056,0.0001502098,0.00006780592,0.0001341834,0.0002568586],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009908624,"about_ca_system_score_gemma":0.0002079393,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01374872,"about_ca_topic_score_gemma":0.2292784,"domain_scores_codex":[0.9980574,0.0003543982,0.000195315,0.0001731963,0.001015213,0.0002044794],"domain_scores_gemma":[0.9988517,0.0006296581,0.00007524994,0.0001414274,0.0002733558,0.00002860148],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000002029562,0.0002044615,0.09309825,0.000003202017,0.00006308185,1.091562e-7,0.7985937,0.000660968,0.000003020467,0.02902945,0.07384103,0.004500731],"study_design_scores_gemma":[0.0002427172,0.0000213347,0.5927417,0.00004426318,0.0000656798,5.644851e-7,0.3024482,0.0004367595,0.000001139861,0.02763575,0.07619528,0.0001666791],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8436687,0.001187729,0.000009681055,0.114558,0.002989085,0.001091214,0.000001021848,0.00001371278,0.03648089],"genre_scores_gemma":[0.9904101,0.000130586,0.0001660917,0.0007760128,0.00101876,0.0004442054,0.00000529327,0.000004617453,0.007044313],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4996434,"threshold_uncertainty_score":0.9996827,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W7104543262","doi":"10.1016/j.stueduc.2025.101530","title":"Stages of teaching expertise from routine to adaptive: A model for advancing teaching effectiveness","year":2025,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary","funders":"","keywords":"Conceptualization; Perspective (graphical); Focus group; Conceptual framework; Teaching method; Qualitative research; Multimethodology; Resource (disambiguation); Qualitative property; Professional development","retraction":null,"screen_n_in":null,"score":{"opus":0.2695009569417753,"gpt":0.5581773688283811,"spread":0.2886764118866058,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006135296,0.0001232443,0.0002603826,0.0002420689,0.0005312085,0.00001719359,0.0001591238,0.00004739529,0.00001502101],"category_scores_gemma":[0.01650255,0.0001284938,0.00005134465,0.000242554,0.0001483187,0.0001972506,0.00004914078,0.0001544265,0.000001611931],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0013455,"about_ca_system_score_gemma":0.001102219,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002249449,"about_ca_topic_score_gemma":0.003423155,"domain_scores_codex":[0.9976368,0.001051846,0.0003423993,0.0003164384,0.0004204465,0.0002320927],"domain_scores_gemma":[0.9962533,0.002801118,0.000125254,0.0001334816,0.0006471355,0.000039768],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001143057,0.000394906,0.03928142,0.00008593842,0.0001650869,1.896643e-8,0.8385293,0.01482056,0.0003357149,0.06746226,0.002385281,0.03642524],"study_design_scores_gemma":[0.001056942,0.00005150856,0.1792056,0.0009186288,0.0001244138,3.179188e-8,0.7291541,0.02428444,0.00011361,0.06429215,0.0005280753,0.0002704949],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8809176,0.005652737,0.04211203,0.06179068,0.002258651,0.002702137,0.00003047489,0.00004686886,0.004488783],"genre_scores_gemma":[0.9830466,0.00003267454,0.01454846,0.0002913877,0.0002180577,0.001325533,0.0000184043,0.000008198436,0.000510708],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1399242,"threshold_uncertainty_score":0.9917819,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1838440795","doi":"10.1016/j.stueduc.2015.09.002","title":"Evaluating intake variables for a teacher education programme: Improving student success and process efficiency","year":2015,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of the Fraser Valley","funders":"","keywords":"Practicum; Psychology; Medical education; Focus group; Process (computing); Program evaluation; Mathematics education; Pedagogy; Medicine; Computer science; Political science; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.433369157431125,"gpt":0.5880200743970577,"spread":0.1546509169659327,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01068082,0.000180673,0.0002263558,0.0002313998,0.0005652245,0.0001245558,0.0002092084,0.00008328414,0.00004098546],"category_scores_gemma":[0.01552691,0.0001791604,0.0000346372,0.0005915477,0.0004426334,0.0004246116,0.00004724993,0.000149372,0.000005019513],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001344154,"about_ca_system_score_gemma":0.006083424,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005979992,"about_ca_topic_score_gemma":0.001486735,"domain_scores_codex":[0.9969517,0.0006387579,0.0004532943,0.0004806954,0.001070289,0.0004052807],"domain_scores_gemma":[0.996253,0.0006350122,0.0002695199,0.0001469406,0.002555959,0.0001395557],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00003119211,0.001206123,0.09697319,0.0001617144,0.00005194665,2.44758e-8,0.7450688,0.000231262,0.00001103261,0.0116106,0.001269082,0.143385],"study_design_scores_gemma":[0.0009270007,0.0001141595,0.0494783,0.0001166043,0.0001169153,9.339442e-7,0.9197651,0.001068622,0.00000621967,0.0269041,0.001244038,0.0002580573],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.942794,0.01058065,0.0001544128,0.03545724,0.003177345,0.003588706,0.000002702353,0.00006144735,0.004183488],"genre_scores_gemma":[0.9894655,0.00006280304,0.003597945,0.0001911754,0.001152305,0.004648767,0.00003714469,0.00001674108,0.0008276068],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1746963,"threshold_uncertainty_score":0.9995512,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4413515290","doi":"10.1016/j.stueduc.2025.101512","title":"Stage-based professional development needs of canadian high school teachers: Insights from the dynamic model of educational effectiveness","year":2025,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Université du Québec à Montréal; Université du Québec en Outaouais","funders":"","keywords":"Professional development; Mathematics education; Faculty development; Psychology; Pedagogy","retraction":null,"screen_n_in":null,"score":{"opus":0.1620902212337623,"gpt":0.4620496655656992,"spread":0.2999594443319369,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002034662,0.0001369049,0.0002236934,0.0006206107,0.0005775909,0.0000173229,0.0002855808,0.00008531565,0.0003267236],"category_scores_gemma":[0.004231882,0.0001144923,0.0000464959,0.001403711,0.0006918202,0.0001301175,0.00003141579,0.000180238,0.000007465256],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002192486,"about_ca_system_score_gemma":0.02908864,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.05344728,"about_ca_topic_score_gemma":0.2730752,"domain_scores_codex":[0.9974627,0.00101457,0.0004499033,0.0002007821,0.0006791358,0.000192935],"domain_scores_gemma":[0.9957545,0.002649651,0.0002154925,0.0001826053,0.001132837,0.00006490628],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00004710466,0.0007541939,0.2416807,0.0001206264,0.0003389918,1.440369e-8,0.6447375,0.009330523,0.0001434919,0.09728352,0.003524622,0.002038628],"study_design_scores_gemma":[0.0004047405,0.000005895159,0.6925524,0.0003299743,0.00004597491,7.837109e-9,0.2771839,0.0005641629,0.00006759577,0.02788525,0.0008561269,0.000103963],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9150862,0.003033121,0.00004339372,0.07709365,0.002029809,0.0008793354,0.0000229583,0.000005505973,0.001806014],"genre_scores_gemma":[0.9957678,0.00004418316,0.0009389132,0.0003333515,0.0001036537,0.00074436,0.0001524342,0.000006559597,0.001908719],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4508716,"threshold_uncertainty_score":0.9764155,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W7117577854","doi":"10.1016/j.stueduc.2025.101557","title":"Assessing school readiness domains in a large cohort of refugee children: Validation and links with family factors","year":2025,"lang":"en","type":"article","venue":"Studies In Educational Evaluation","topic":"Early Childhood Education and Development","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":"Child, Adolescent and Family Mental Health","funders":"Universiti Sains Malaysia; British Academy","keywords":"Rasch model; Refugee; Polytomous Rasch model; Construct validity; Psychometrics; Sample (material); Cohort; Item analysis; Item response theory","retraction":null,"screen_n_in":null,"score":{"opus":0.05304186638219593,"gpt":0.4417620280352073,"spread":0.3887201616530114,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002678716,0.00009753802,0.0001778089,0.0003734312,0.0002632904,0.00005094863,0.00008137938,0.00008441163,0.00004753791],"category_scores_gemma":[0.002125195,0.00008958073,0.0000148507,0.0008759616,0.0001762158,0.0003925891,0.00002856744,0.0001527227,0.000001724366],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006701802,"about_ca_system_score_gemma":0.003064737,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005439629,"about_ca_topic_score_gemma":0.001782768,"domain_scores_codex":[0.9983786,0.000324984,0.0003502215,0.0002475513,0.0005362639,0.0001623995],"domain_scores_gemma":[0.9988177,0.0003441558,0.0001498125,0.0001028479,0.0005487329,0.00003676866],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000005844272,0.0001315286,0.9709764,0.00001433973,0.00004334005,2.232005e-8,0.01994953,0.00007612668,0.000009388834,0.007549754,0.0002443572,0.0009994184],"study_design_scores_gemma":[0.0004667572,0.000008890222,0.9570421,0.0002139466,0.00002743085,1.255063e-7,0.03703432,0.0000134825,0.00002441056,0.004974802,0.0001066859,0.00008705563],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9892971,0.0008828013,0.00004662101,0.002393199,0.0004477583,0.0007232614,0.000002688521,0.000009534356,0.006197067],"genre_scores_gemma":[0.9979897,0.000287341,0.001046121,0.0001369341,0.00007760557,0.000135116,0.00007974968,0.000004546557,0.0002429609],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0170848,"threshold_uncertainty_score":0.5436712,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}