{"meta":{"query_hash":"9f297251ca0a","filters":{"venue":"Educational Evaluation and Policy Analysis"},"cohort_total":14,"direct_labels_cover":0,"predictions_cover":14,"exported":14,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/9f297251ca0a","api":"https://metacan.xera.ac/api/v1/cohort?venue=Educational+Evaluation+and+Policy+Analysis"},"results":[{"id":"W1977534634","doi":"10.3102/0162373711398125","title":"Who Benefits From Kindergarten? Evidence From the Introduction of State Subsidization","year":2011,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Socioeconomic status; Subsidy; Immigration; Demographic economics; Revenue; Psychology; State (computer science); Mathematics education; Economic growth; Economics; Demography; Political science; Sociology; Population; Computer science","score_opus":0.08998701385613952,"score_gpt":0.37710806615415127,"score_spread":0.28712105229801177,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1977534634","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.91914886,0.0011889929,0.00039562013,0.07706423,0.0002902206,0.00026177018,0.000041378516,0.000014382986,0.0015945173],"genre_scores_gemma":[0.9955035,0.00084759813,0.0007942874,0.0006670085,0.001277168,0.000028693234,0.0002870333,0.000004826171,0.0005899102],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99805987,0.0005511189,0.0003234825,0.00025627806,0.0006845559,0.0001246834],"domain_scores_gemma":[0.99813014,0.00051735085,0.00026634935,0.00022046853,0.0007492099,0.00011645402],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014079309,0.000087703156,0.00014060526,0.00024907035,0.0004012878,0.00006499777,0.00016151651,0.00004929932,0.0062454366],"category_scores_gemma":[0.0026023518,0.000074379306,0.00006941144,0.0014766604,0.00014716045,0.00033841346,0.000018516497,0.000064997184,0.00003115092],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000030952273,0.00017422278,0.5034853,0.000002683899,0.0008652936,1.6863204e-8,0.40830466,0.0010435198,0.0000539504,0.041479312,0.010507074,0.03405301],"study_design_scores_gemma":[0.00008429023,0.0000065440872,0.9708068,0.000011850647,0.0004054239,4.3837336e-8,0.0035362514,0.0003467139,0.00013176451,0.023151716,0.0014350166,0.000083630934],"about_ca_topic_score_codex":0.07333943,"about_ca_topic_score_gemma":0.017463546,"teacher_disagreement_score":0.46732143,"about_ca_system_score_codex":0.0001272065,"about_ca_system_score_gemma":0.0021711895,"threshold_uncertainty_score":0.994663},"labels":[],"label_agreement":null},{"id":"W1980838979","doi":"10.3102/01623737027001001","title":"Success for All: First-Year Results From the National Randomized Field Trial","year":2005,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":84,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Randomized experiment; Quarter (Canadian coin); Test (biology); Psychology; Randomized controlled trial; Multilevel model; Control (management); Treatment and control groups; Reading (process); Mathematics education; Statistics; Medicine; Mathematics; Computer science; Political science; Geography","score_opus":0.09148252164579562,"score_gpt":0.47779666160513007,"score_spread":0.38631413995933445,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1980838979","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19424497,0.00023545256,0.00027325135,0.78531575,0.00030377592,0.0011214499,0.00018693479,0.0000168593,0.018301578],"genre_scores_gemma":[0.9798663,0.0002265717,0.00058177504,0.006394526,0.010067026,0.00034827398,0.00049030234,0.0000035322164,0.002021678],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9981626,0.00041120642,0.00033051154,0.00019445905,0.00076444965,0.00013673701],"domain_scores_gemma":[0.9920203,0.007086719,0.00017705956,0.00010158263,0.00054187287,0.00007243117],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0053062504,0.000074205986,0.00017730313,0.00016695602,0.00067227636,0.00014908987,0.00017244782,0.00006946439,0.0016297854],"category_scores_gemma":[0.013630019,0.00005553096,0.00019462191,0.0006047653,0.00010043073,0.0002461806,0.000010369508,0.00006821372,0.000034310116],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.2049459,0.00040387386,0.010972642,0.000007214694,0.003375257,1.5915546e-8,0.04875922,0.009509581,0.0000026787015,0.24645104,0.45970547,0.015867116],"study_design_scores_gemma":[0.32196534,0.000054601456,0.022138923,0.0000139534695,0.002718351,9.256178e-8,0.0015007405,0.05194182,0.000018171746,0.06947659,0.5298158,0.0003556511],"about_ca_topic_score_codex":0.009939523,"about_ca_topic_score_gemma":0.026292901,"teacher_disagreement_score":0.78562135,"about_ca_system_score_codex":0.000115341536,"about_ca_system_score_gemma":0.0009487715,"threshold_uncertainty_score":0.99928284},"labels":[],"label_agreement":null},{"id":"W1987197661","doi":"10.3102/0162373711424206","title":"Differential Effects of Literacy Instruction Time and Homogeneous Ability Grouping in Kindergarten Classrooms","year":2011,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology; Homogeneous; Literacy; Mathematics education; Developmental psychology; Early literacy; Cognition; Cognitive development; Differential effects; Longitudinal study; Early childhood; Pedagogy; Statistics","score_opus":0.025604482175742054,"score_gpt":0.36474640522516194,"score_spread":0.33914192304941987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1987197661","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.996866,0.00013315724,0.00003683785,0.0010998588,0.00007749071,0.00014221761,0.0000031799318,0.000005136266,0.0016361332],"genre_scores_gemma":[0.9991299,0.00012210582,0.00014934149,0.00006725391,0.00030633135,0.000022891682,0.000031375643,0.0000024408482,0.0001683719],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9989434,0.00029147798,0.0002044612,0.00016117128,0.00028963605,0.00010984308],"domain_scores_gemma":[0.9993614,0.00020623216,0.00010868424,0.00009087933,0.00016471205,0.00006807493],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0004885718,0.00006405917,0.0001379524,0.000443701,0.0001578032,0.00003053963,0.000055435536,0.000059075905,0.0010889757],"category_scores_gemma":[0.00048827485,0.000064136075,0.000047649177,0.0008498905,0.00011532406,0.00027163213,0.000012817931,0.00005895468,0.0000054359416],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046179866,0.00023165901,0.8623029,0.00005508895,0.00021115264,1.1347056e-7,0.058031056,0.00007081983,0.00046443884,0.0059490474,0.000043928216,0.07259363],"study_design_scores_gemma":[0.0002713807,0.000017324315,0.9876843,0.000012167732,0.00023495765,3.5632132e-7,0.000258493,0.004114683,0.00008805975,0.007143195,0.00010435606,0.00007069157],"about_ca_topic_score_codex":0.0075036073,"about_ca_topic_score_gemma":0.0032631624,"teacher_disagreement_score":0.12538145,"about_ca_system_score_codex":0.000076850345,"about_ca_system_score_gemma":0.00026840722,"threshold_uncertainty_score":0.99982417},"labels":[],"label_agreement":null},{"id":"W2124978570","doi":"10.3102/0162373707309073","title":"Early-Grade Retention and Children’s Reading and Math Learning in Elementary Years","year":2007,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":99,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Grade retention; Psychology; Reading (process); Retention rate; Mathematics education; Early childhood; Intervention (counseling); Developmental psychology; Academic achievement; Computer science","score_opus":0.026363589418701528,"score_gpt":0.382660152524002,"score_spread":0.35629656310530045,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124978570","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9868891,0.000254304,0.00003767754,0.010758891,0.000047917893,0.00016770136,0.0000016600391,0.000011729054,0.0018310128],"genre_scores_gemma":[0.997331,0.00039190493,0.00070833793,0.00039739936,0.00025388462,0.000011236974,0.000091995236,0.0000041312287,0.0008100789],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99881923,0.00019964187,0.0002247247,0.00021150569,0.00037423903,0.00017067899],"domain_scores_gemma":[0.9994901,0.00015209979,0.00009430854,0.0000531229,0.000079760866,0.0001306317],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028451697,0.00006946898,0.000102266364,0.00075114315,0.00031627069,0.00010144846,0.000037248043,0.00005023063,0.00019395274],"category_scores_gemma":[0.00038609677,0.000078902434,0.000027387829,0.0009069371,0.00008326621,0.00018708923,0.000016534981,0.00009714829,0.000005375621],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000037313282,0.00003206977,0.9102052,0.0000018663611,0.00008310622,6.4744356e-8,0.036058567,0.000027416863,0.0000113414435,0.027554432,0.00008960489,0.02593262],"study_design_scores_gemma":[0.00018768237,0.000008122168,0.99185944,0.00000806247,0.00010884028,8.755887e-7,0.0036599454,0.00018499345,0.0000019551912,0.0029353364,0.0009596765,0.00008507036],"about_ca_topic_score_codex":0.010957049,"about_ca_topic_score_gemma":0.0033935937,"teacher_disagreement_score":0.08165426,"about_ca_system_score_codex":0.00012991954,"about_ca_system_score_gemma":0.00045055078,"threshold_uncertainty_score":0.9956291},"labels":[],"label_agreement":null},{"id":"W2130850661","doi":"10.3102/01623737027003205","title":"Effects of Kindergarten Retention Policy on Children’s Cognitive Growth in Reading and Mathematics","year":2005,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":277,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Grade retention; Psychology; Propensity score matching; Multilevel model; Limiting; Promotion (chess); Academic achievement; Developmental psychology; Homogeneous; Harm; Reading (process); Mathematics education; Longitudinal study; Cohort; Social psychology; Political science; Mathematics; Statistics","score_opus":0.020192797292193084,"score_gpt":0.3748215988611014,"score_spread":0.3546288015689083,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2130850661","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.95326585,0.0001944196,0.00006512259,0.038331926,0.00003506835,0.00039015827,0.000006446614,0.000009283595,0.0077017182],"genre_scores_gemma":[0.9973404,0.00027258956,0.00058419036,0.00072097447,0.00043834222,0.000037273017,0.00006846745,0.000004792041,0.0005330158],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9986736,0.00026449317,0.00026656792,0.00018812253,0.0004661868,0.00014099524],"domain_scores_gemma":[0.99887615,0.0005418973,0.00016346165,0.00005508628,0.0002589502,0.00010446166],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009768583,0.000087467895,0.00015957643,0.0011838473,0.0001934814,0.000046134504,0.000051186686,0.000060498336,0.00014634871],"category_scores_gemma":[0.002685388,0.00008961403,0.000050449562,0.0013831017,0.00009881343,0.00015586977,0.000011373755,0.00006731094,0.0000098720375],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001786051,0.0008278751,0.21258718,0.00004388831,0.0006315956,6.393266e-8,0.18753462,0.00019553666,0.000039115832,0.5525708,0.0006255984,0.04492583],"study_design_scores_gemma":[0.00036650032,0.000015237107,0.9825962,0.000048713813,0.000210339,4.477389e-7,0.001545426,0.0003009549,0.00012725934,0.014638343,0.000054872784,0.00009566536],"about_ca_topic_score_codex":0.004493937,"about_ca_topic_score_gemma":0.00086105533,"teacher_disagreement_score":0.77000904,"about_ca_system_score_codex":0.00022291105,"about_ca_system_score_gemma":0.0014300431,"threshold_uncertainty_score":0.6793518},"labels":[],"label_agreement":null},{"id":"W2144062847","doi":"10.3102/01623737024003201","title":"Do Low-Achieving Students Benefit More from Small Classes? Evidence from the Tennessee Class Size Experiment","year":2002,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":96,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Class size; Differential effects; Class (philosophy); Mathematics education; Academic achievement; Randomized experiment; Psychology; Reading (process); Differential (mechanical device); Student achievement; Scale (ratio); Quarter (Canadian coin); Mathematics; Statistics; Computer science; Physics; Geography; Political science","score_opus":0.10885470082861255,"score_gpt":0.44710158325711225,"score_spread":0.3382468824284997,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2144062847","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9181305,0.0037785163,0.000018201043,0.07637128,0.00018943426,0.00021509913,0.000033953453,0.000015591377,0.0012474271],"genre_scores_gemma":[0.99083155,0.0015295633,0.00019751182,0.0028537395,0.0027127333,0.00012448622,0.00006800943,0.000007315917,0.0016751138],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.997378,0.00038753118,0.00030791175,0.00036799902,0.0013282497,0.00023031599],"domain_scores_gemma":[0.996651,0.0023149492,0.00019786769,0.0003421572,0.00033738118,0.00015662215],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001296953,0.00014535026,0.00018903275,0.00013381208,0.0009100875,0.00048209075,0.0005064966,0.00008486442,0.010076857],"category_scores_gemma":[0.0020920993,0.000116604795,0.00013439018,0.00119324,0.00014262556,0.00040394397,0.00006102394,0.00014002316,0.00012974764],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016764567,0.00020716428,0.9226873,0.0000036085992,0.00068042165,1.4944123e-7,0.061812498,0.001075859,0.00018068252,0.0014152215,0.0043131304,0.007607205],"study_design_scores_gemma":[0.00020630685,0.000008517618,0.98043805,0.000052449705,0.0006770285,6.973824e-8,0.0073735584,0.0045325845,0.000027919416,0.0015028052,0.005010586,0.00017014307],"about_ca_topic_score_codex":0.033431582,"about_ca_topic_score_gemma":0.0142991785,"teacher_disagreement_score":0.073517546,"about_ca_system_score_codex":0.00025790703,"about_ca_system_score_gemma":0.00032172186,"threshold_uncertainty_score":0.99082804},"labels":[],"label_agreement":null},{"id":"W2146700645","doi":"10.3102/0162373708328259","title":"Reading Instruction Time and Homogeneous Grouping in Kindergarten: An Application of Marginal Mean Weighting Through Stratification","year":2008,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Reading (process); Psychology; Homogeneous; Weighting; Mathematics education; Multilevel model; Reading comprehension; Longitudinal study; Set (abstract data type); Computer science; Statistics; Linguistics; Mathematics","score_opus":0.04578867278231784,"score_gpt":0.3879297195275477,"score_spread":0.3421410467452298,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146700645","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99240446,0.00011530111,0.00042146968,0.0039676866,0.000028269365,0.00018140877,0.000004467255,0.000010607413,0.002866334],"genre_scores_gemma":[0.99766994,0.00028807877,0.0011116868,0.000109191686,0.00048089566,0.0000366134,0.00017572727,0.000004367832,0.00012352709],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99865466,0.00026481794,0.00029563849,0.00023243422,0.00042054334,0.00013191356],"domain_scores_gemma":[0.9992295,0.00011857809,0.00021234564,0.00012585202,0.00024873368,0.000064975364],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011450794,0.00007770482,0.00014089493,0.0004475833,0.000479664,0.000035636313,0.00007055579,0.00006957082,0.00013754197],"category_scores_gemma":[0.00019191585,0.000086365195,0.00003293436,0.0013729343,0.00014059803,0.00064217526,0.000007519374,0.00007140198,0.0000053725216],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000050814753,0.00029343524,0.7063082,0.000040683866,0.00022469655,2.4136216e-7,0.112268955,0.007030978,0.0036891706,0.10408393,0.00012072263,0.06588816],"study_design_scores_gemma":[0.00036727075,0.000028735481,0.90244573,0.000014709877,0.00024029145,0.0000060786983,0.0032804417,0.08054835,0.00019817802,0.012063967,0.0006226008,0.00018364588],"about_ca_topic_score_codex":0.009075721,"about_ca_topic_score_gemma":0.0058067073,"teacher_disagreement_score":0.19613752,"about_ca_system_score_codex":0.00012750263,"about_ca_system_score_gemma":0.00043459333,"threshold_uncertainty_score":0.99752295},"labels":[],"label_agreement":null},{"id":"W2153308398","doi":"10.3102/01623737023003251","title":"Class Size and Eighth-Grade Math Achievement in the United States and Abroad","year":2001,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":82,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Class size; Mathematics education; Curriculum; Class (philosophy); Student achievement; Academic achievement; Math education; Psychology; Pedagogy; Computer science","score_opus":0.05050335955880637,"score_gpt":0.4362833388378362,"score_spread":0.38577997927902985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2153308398","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8555945,0.00033167232,0.0000048729457,0.1421307,0.000019014853,0.00011385569,0.000005027587,0.0000038020878,0.0017965097],"genre_scores_gemma":[0.99066556,0.0039803307,0.000056599954,0.0040025963,0.00033287014,0.000039025108,0.000092836715,0.0000020607301,0.0008281364],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9988443,0.00034440105,0.0001447561,0.00013767892,0.0003993803,0.00012944549],"domain_scores_gemma":[0.9991557,0.0005346884,0.00006073173,0.000082532744,0.000098819044,0.00006755573],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001679092,0.00006505414,0.00008553884,0.0003208914,0.0003847465,0.00012957398,0.00007135087,0.000035719866,0.0002974498],"category_scores_gemma":[0.0003644263,0.00004903362,0.000022578564,0.0015644976,0.000116082,0.00015985727,0.000010413583,0.00007100085,0.0000049885493],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000186381,0.00015498266,0.8749432,0.000008317875,0.00016420822,1.8468877e-7,0.05833197,0.000678323,0.000009743751,0.05839717,0.0025441323,0.0047491314],"study_design_scores_gemma":[0.00020077577,0.000010656438,0.9204629,0.000003954623,0.00017355497,5.6498055e-7,0.0072616707,0.008256371,6.5890663e-7,0.009025307,0.05453733,0.000066231674],"about_ca_topic_score_codex":0.019047076,"about_ca_topic_score_gemma":0.018792948,"teacher_disagreement_score":0.13812812,"about_ca_system_score_codex":0.000056064397,"about_ca_system_score_gemma":0.0002213735,"threshold_uncertainty_score":0.99911153},"labels":[],"label_agreement":null},{"id":"W2156855199","doi":"10.3102/01623737022004357","title":"Power and Politics in the Adoption of School Reform Models","year":2000,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":174,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Education reform; Power (physics); Perspective (graphical); Politics; Process (computing); Public administration; Sustainability; Set (abstract data type); Political science; Public relations; Economic growth; Economics; Higher education","score_opus":0.15073264855465782,"score_gpt":0.49702515498723526,"score_spread":0.3462925064325775,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2156855199","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9124329,0.0002161416,0.00007056484,0.05434098,0.00003065908,0.00017920518,0.000016137112,0.000001767463,0.03271164],"genre_scores_gemma":[0.9951637,0.00008668088,0.00032226415,0.0013964289,0.00013113538,0.000043224623,0.000064312015,0.0000023133503,0.0027899754],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.99767673,0.00026939838,0.0004950807,0.00021813136,0.0012283535,0.0001123183],"domain_scores_gemma":[0.99861276,0.00048659416,0.00014713532,0.0002515222,0.0004305583,0.00007145171],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003470145,0.00007908913,0.00014817595,0.00074268813,0.00011710767,0.00011278829,0.00017601394,0.000035036803,0.0043913717],"category_scores_gemma":[0.0005240427,0.0000517356,0.00006793295,0.0015459679,0.00007681164,0.00036485714,0.000014716551,0.000065776636,0.000033189288],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000025880934,0.0005600401,0.07123416,0.000007285634,0.00023508602,4.440465e-8,0.010456729,0.0071522966,0.00005195538,0.87242573,0.0036597447,0.034191042],"study_design_scores_gemma":[0.00015907384,0.000023004595,0.68724376,0.000002921161,0.00010354929,9.491096e-7,0.002757199,0.023019886,0.0000033927968,0.28506306,0.001566708,0.000056515633],"about_ca_topic_score_codex":0.00091951573,"about_ca_topic_score_gemma":0.0001460981,"teacher_disagreement_score":0.6160096,"about_ca_system_score_codex":0.00014489118,"about_ca_system_score_gemma":0.00060060975,"threshold_uncertainty_score":0.99651873},"labels":[],"label_agreement":null},{"id":"W2791023047","doi":"10.3102/0162373718760218","title":"School Improvement Grants in Ohio: Effects on Student Achievement and School Administration","year":2018,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Regression discontinuity design; Quarter (Canadian coin); Psychology; Demographic economics; Turnover; Regression analysis; Mathematics education; Economics; Statistics; Geography; Mathematics; Management","score_opus":0.032870814993456954,"score_gpt":0.4489629340119663,"score_spread":0.41609211901850934,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2791023047","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9646601,0.00010507304,0.000011355448,0.028913679,0.00014050391,0.00041563323,0.000004485418,0.000008570862,0.005740606],"genre_scores_gemma":[0.99440366,0.00019219509,0.000070672606,0.002330943,0.0016623675,0.00015290837,0.00004336508,0.0000039848696,0.0011399009],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99826616,0.0002744882,0.00024696192,0.0002723778,0.00074753445,0.00019250112],"domain_scores_gemma":[0.9991215,0.00018173108,0.00010745056,0.00013902901,0.00022486261,0.00022542979],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017870403,0.000105559055,0.00013881855,0.0005617331,0.00044797815,0.00015095128,0.00009248906,0.00005908761,0.0010548751],"category_scores_gemma":[0.000869147,0.00010195576,0.00004136054,0.0010268665,0.00009159167,0.00024448227,0.000018110279,0.000096222524,0.00011342146],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000077951925,0.00047569568,0.9434243,0.000020505971,0.00030914173,1.9236707e-7,0.0071299733,0.000055233362,0.00022680762,0.014933301,0.0020630094,0.031283893],"study_design_scores_gemma":[0.00048042182,0.00018925562,0.9925522,0.000019408013,0.00019385741,8.351221e-8,0.0008102916,0.00044796796,0.000146081,0.0022904165,0.0027573376,0.000112659654],"about_ca_topic_score_codex":0.0020926795,"about_ca_topic_score_gemma":0.016001998,"teacher_disagreement_score":0.049127925,"about_ca_system_score_codex":0.00031358557,"about_ca_system_score_gemma":0.0010207203,"threshold_uncertainty_score":0.9998583},"labels":[],"label_agreement":null},{"id":"W2804370236","doi":"10.3102/0162373718782648","title":"Course Choice and Achievement Effects of a System-Wide Vocational Education and Training Voucher Scheme for Young People","year":2018,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Education Systems and Policy","field":"Social Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Queen's University; National Centre for Vocational Education Research; Queen's University Belfast","keywords":"Voucher; Vocational education; Equity (law); School choice; Academic achievement; Training (meteorology); Quality (philosophy); Public economics; Psychology; Mathematics education; Economic growth; Medical education; Pedagogy; Political science; Business; Economics; Accounting","score_opus":0.044016428965278694,"score_gpt":0.4417211995699809,"score_spread":0.3977047706047022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2804370236","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9647971,0.0010251403,0.00094844896,0.026115086,0.0007125583,0.0009976419,0.000028741217,0.000017904553,0.0053573707],"genre_scores_gemma":[0.9923115,0.000041380576,0.0022392636,0.0003800159,0.0028174545,0.00043076542,0.00012299989,0.000008871996,0.0016477575],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9984038,0.00032674268,0.00034701847,0.00028250253,0.00045401836,0.00018592473],"domain_scores_gemma":[0.99732876,0.0009897075,0.00030411014,0.00014323906,0.0010457344,0.00018843453],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0018590778,0.00012088006,0.00023835976,0.0004547114,0.0006137711,0.00009667996,0.00007859156,0.0000825945,0.0000828384],"category_scores_gemma":[0.0016950371,0.00012539877,0.00006548029,0.00091115566,0.00023211133,0.00020761747,0.000015301477,0.00004431824,0.0000034896052],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000012996701,0.00035573324,0.37251014,0.0004093351,0.00074756634,4.471052e-9,0.17846078,0.000011972086,0.00013968899,0.42595512,0.004159544,0.017237114],"study_design_scores_gemma":[0.0004278323,0.00004658497,0.95834535,0.00008468175,0.0009197279,0.0000014580478,0.024169503,0.002710286,0.000016126527,0.0029790779,0.010124522,0.00017486954],"about_ca_topic_score_codex":0.0116907,"about_ca_topic_score_gemma":0.0052191955,"teacher_disagreement_score":0.58583516,"about_ca_system_score_codex":0.00016560516,"about_ca_system_score_gemma":0.0029899147,"threshold_uncertainty_score":0.9948905},"labels":[],"label_agreement":null},{"id":"W3008309614","doi":"10.3102/0162373720906217","title":"Teacher Coaching in a Simulated Environment","year":2020,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":181,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact","funders":"National Academy of Education; University of Virginia; Spencer Foundation","keywords":"Coaching; Psychology; Perception; Mathematics education; Medical education; Teacher education; Pedagogy; Applied psychology; Medicine","score_opus":0.21363410786154396,"score_gpt":0.4796332351777997,"score_spread":0.26599912731625575,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3008309614","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.27111402,0.00040614922,0.00009359148,0.7208598,0.0000329317,0.00018744686,0.0000023956018,0.000016375587,0.0072872355],"genre_scores_gemma":[0.9948868,0.00007113071,0.0000988315,0.003556749,0.00052189187,0.000022116852,0.00004011486,0.0000038237763,0.0007985095],"study_design_codex":"qualitative","study_design_gemma":"observational","domain_scores_codex":[0.99848276,0.0005955494,0.00018144798,0.00019879796,0.0003999477,0.00014146665],"domain_scores_gemma":[0.9994644,0.00020902633,0.000067272646,0.00006590744,0.00005368487,0.0001397349],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0011011901,0.000070940616,0.00012419288,0.000286652,0.00019223115,0.000046909565,0.0000717614,0.000044637836,0.0032322688],"category_scores_gemma":[0.0015107457,0.00007497884,0.00005661916,0.0010819451,0.00009763645,0.000104220504,0.000012135123,0.00010302849,0.00006201918],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009566335,0.0001821689,0.27761433,0.0000053263802,0.0002904202,6.464118e-8,0.68870246,0.0089554535,0.000012008881,0.015519673,0.0015928475,0.0071156677],"study_design_scores_gemma":[0.00043347804,0.000012476788,0.7834407,0.0000046091072,0.00049872027,7.9147796e-8,0.14143881,0.025402157,0.000001930832,0.00313585,0.045413002,0.00021816017],"about_ca_topic_score_codex":0.0046052374,"about_ca_topic_score_gemma":0.0011188927,"teacher_disagreement_score":0.7237728,"about_ca_system_score_codex":0.00020514903,"about_ca_system_score_gemma":0.00050866185,"threshold_uncertainty_score":0.99767894},"labels":[],"label_agreement":null},{"id":"W4390877660","doi":"10.3102/01623737231218735","title":"Beyond Prescriptive Reforms: An Examination of North Carolina’s Flexible School Restart Program","year":2024,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"School Choice and Performance","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Autonomy; Flexibility (engineering); Principal (computer security); Quarter (Canadian coin); School choice; Mathematics education; South carolina; Psychology; Pedagogy; Public administration; Sociology; Political science; Economics; Management; Computer science; Law; History","score_opus":0.056100336068339335,"score_gpt":0.4434867261868464,"score_spread":0.38738639011850706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390877660","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9642903,0.00066756987,0.000023572758,0.007987629,0.00016727915,0.00039071956,0.000035622506,0.00005633948,0.026380977],"genre_scores_gemma":[0.9933253,0.00029134637,0.0004652353,0.00017741353,0.001321167,0.00016311888,0.00042112236,0.000006319679,0.0038289826],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9982144,0.00033884507,0.00026550575,0.00025176269,0.0007621443,0.0001673006],"domain_scores_gemma":[0.9988592,0.00011486901,0.00009379089,0.00016897597,0.00061878527,0.00014437611],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0019044664,0.00009157133,0.00013713518,0.0008180619,0.0003137992,0.00013569683,0.00012485925,0.000061355306,0.0008482287],"category_scores_gemma":[0.0006361458,0.000084040636,0.00008427526,0.0025763377,0.00014088483,0.0009206339,0.00001280574,0.0001069914,0.000030932268],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000044581284,0.00050834654,0.32559788,0.00013482834,0.0010390984,2.8409292e-7,0.06759626,0.0024247086,0.000036818117,0.13626678,0.005675547,0.46067485],"study_design_scores_gemma":[0.00008782254,0.00006163336,0.9681155,0.0000149196485,0.0005293024,2.299233e-7,0.0014502662,0.011178225,0.00001824493,0.00730716,0.011130547,0.000106125386],"about_ca_topic_score_codex":0.008128035,"about_ca_topic_score_gemma":0.031194272,"teacher_disagreement_score":0.6425176,"about_ca_system_score_codex":0.0002519514,"about_ca_system_score_gemma":0.0020811323,"threshold_uncertainty_score":0.9984769},"labels":[],"label_agreement":null},{"id":"W4407384694","doi":"10.3102/01623737241311537","title":"Measuring Grading Standards at High Schools: A Methodological Contribution, an Example, and Some Policy Implications","year":2025,"lang":"en","type":"article","venue":"Educational Evaluation and Policy Analysis","topic":"Higher Education Learning Practices","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Wilfrid Laurier University","funders":"Canadian Institutes of Health Research; University of Alberta","keywords":"Grading (engineering); Academic standards; Mathematics education; Policy analysis; Higher education; Econometrics; Psychology; Political science; Economics; Economic growth; Public administration; Engineering","score_opus":0.2785140261361238,"score_gpt":0.5528497145082416,"score_spread":0.2743356883721178,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407384694","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6317106,0.0009176367,0.002753463,0.36064056,0.00017370899,0.0004460443,0.000055067707,0.00007038638,0.0032325222],"genre_scores_gemma":[0.9913523,0.00040647967,0.0025509784,0.001753022,0.0008895954,0.00019996772,0.000219902,0.0000069800976,0.0026207522],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.99496675,0.0030814058,0.00039015227,0.0004906548,0.000770348,0.00030071323],"domain_scores_gemma":[0.995616,0.0019428965,0.00028570456,0.0002911877,0.0015644864,0.00029969457],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.009614435,0.00014709306,0.00027749222,0.0011184288,0.002131317,0.00035953746,0.00018170798,0.00012985247,0.0007471397],"category_scores_gemma":[0.01825696,0.00015245655,0.00008135653,0.0031608914,0.00027099898,0.0007671216,0.00005666643,0.00018361173,0.0000102720005],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001649671,0.00008736865,0.08591696,0.00000659839,0.00036081276,2.2145196e-8,0.0073651243,0.00041069835,0.000103292,0.8992026,0.0011353976,0.0053946213],"study_design_scores_gemma":[0.00031556387,0.000015071688,0.77172226,0.000011247356,0.00086939795,0.0000011003076,0.0020335808,0.00038959974,0.000017846392,0.14875661,0.07569472,0.00017298867],"about_ca_topic_score_codex":0.030668754,"about_ca_topic_score_gemma":0.0036120876,"teacher_disagreement_score":0.750446,"about_ca_system_score_codex":0.0014405397,"about_ca_system_score_gemma":0.005138886,"threshold_uncertainty_score":0.9991678},"labels":[],"label_agreement":null}]}