{"meta":{"query_hash":"8b186b870421","filters":{"venue":"Educational Assessment Evaluation and Accountability"},"cohort_total":15,"direct_labels_cover":0,"predictions_cover":15,"exported":15,"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/8b186b870421","api":"https://metacan.xera.ac/api/v1/cohort?venue=Educational+Assessment+Evaluation+and+Accountability"},"results":[{"id":"W1566806182","doi":"10.1007/s11092-011-9132-4","title":"Learning as identity and practice through involvement in online moderation","year":2011,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Online and Blended Learning","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Queen's University","keywords":"Moderation; Context (archaeology); Psychology; Identity (music); Social psychology; Qualitative research; Element (criminal law); Pedagogy; Sociology; Political science; Social science","score_opus":0.15529069055994224,"score_gpt":0.507281857070541,"score_spread":0.3519911665105987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1566806182","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":[0.965082,0.00012943425,0.00024168572,0.012082399,0.00017168578,0.00044528028,0.0000019257886,0.00001773051,0.021827884],"genre_scores_gemma":[0.9834878,0.00024130553,0.015319306,0.00048826824,0.00017623778,0.000060554838,0.0000801264,0.0000044384137,0.00014195508],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9976226,0.0009960651,0.00025534275,0.00028719875,0.0006979289,0.00014082903],"domain_scores_gemma":[0.99878806,0.0004633625,0.00014326189,0.00008194482,0.0004720745,0.0000513219],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005582265,0.000083341074,0.000094540745,0.00005715161,0.00043503774,0.00013432966,0.00006267048,0.0000585022,0.0012534002],"category_scores_gemma":[0.0020473104,0.00008691743,0.000016044667,0.00019730942,0.00011446409,0.0025607804,0.000038637238,0.00021312616,0.0000068377235],"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":[0.00005661136,0.0019080933,0.6853884,0.00003589662,0.000026514745,2.822758e-7,0.06113346,0.00020935715,0.0001086956,0.1945454,0.00009688237,0.05649045],"study_design_scores_gemma":[0.0004244139,0.00006353082,0.7667501,0.000019460751,0.000039368366,9.295689e-7,0.09831007,0.0031242117,0.0000055050155,0.12418058,0.0069424193,0.0001394287],"about_ca_topic_score_codex":0.006586135,"about_ca_topic_score_gemma":0.008281832,"teacher_disagreement_score":0.08136173,"about_ca_system_score_codex":0.00032527643,"about_ca_system_score_gemma":0.0014794135,"threshold_uncertainty_score":0.9996596},"labels":[],"label_agreement":null},{"id":"W1948718713","doi":"10.1007/s11092-015-9231-8","title":"Towards a framework for the validation of early childhood assessment systems","year":2015,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"Fok Ying Tong Education Foundation; U.S. Department of Education","keywords":"Early childhood education; Early childhood; Accountability; Educational assessment; Conceptual framework; Argument (complex analysis); Field (mathematics); Work (physics); Psychology; Political science; Pedagogy; Developmental psychology; Sociology; Medicine; Social science; Engineering","score_opus":0.09283552875712486,"score_gpt":0.4581947012178403,"score_spread":0.3653591724607154,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1948718713","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":[0.9351956,0.0005047503,0.017193075,0.030798478,0.0033832635,0.0036228453,0.000039334045,0.000038091843,0.009224563],"genre_scores_gemma":[0.98434895,0.000036986774,0.0137839755,0.00018680385,0.00065032305,0.0007753148,0.000089513145,0.0000085235,0.0001195919],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9971274,0.0004937285,0.00044844783,0.00029306396,0.0014447367,0.00019262453],"domain_scores_gemma":[0.9962591,0.0011519096,0.00030776486,0.00025855005,0.0018697528,0.00015290627],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008678145,0.00012394947,0.00016564546,0.00007297563,0.0005459117,0.0002702371,0.00021568725,0.00009055461,0.00035034347],"category_scores_gemma":[0.001567636,0.000098434975,0.00005909672,0.00030321677,0.00017767951,0.0004028367,0.000033162443,0.00013616314,0.0000052129008],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015447955,0.00075941125,0.30118856,0.00003166212,0.0000937679,6.650843e-9,0.046543285,0.00041179036,0.000005081875,0.62678885,0.0024981438,0.021663997],"study_design_scores_gemma":[0.00037402025,0.00004655649,0.8803624,0.000024535473,0.000060802984,2.9681993e-7,0.02218005,0.00044436095,0.0000077376,0.092217624,0.0041662194,0.00011539153],"about_ca_topic_score_codex":0.0017003834,"about_ca_topic_score_gemma":0.00007317229,"teacher_disagreement_score":0.57917386,"about_ca_system_score_codex":0.0006816606,"about_ca_system_score_gemma":0.01149056,"threshold_uncertainty_score":0.9941134},"labels":[],"label_agreement":null},{"id":"W1983949565","doi":"10.1007/s11092-008-9063-x","title":"Using data to support educational improvement","year":2008,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":127,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Ministry of Education and Child Care","funders":"","keywords":"Data collection; Context (archaeology); Computer science; Scale (ratio); Data science; Knowledge management; Sociology","score_opus":0.5729116068459882,"score_gpt":0.5914743349389096,"score_spread":0.018562728092921366,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983949565","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":[0.9519051,0.00005192179,0.0014406836,0.038112428,0.001870631,0.0014572681,0.0001948144,0.000017745479,0.0049494603],"genre_scores_gemma":[0.9707683,0.00000981536,0.022749238,0.0024478883,0.00088550587,0.00029887745,0.00094215584,0.000016080727,0.0018821764],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99231845,0.0003259483,0.0012057356,0.0014066222,0.004377117,0.00036612447],"domain_scores_gemma":[0.9938186,0.0014609569,0.00041735588,0.0015718774,0.0023735226,0.00035771323],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009523775,0.00028225582,0.00030944397,0.00036684744,0.0007907636,0.00037986063,0.0011500191,0.00006906506,0.017165719],"category_scores_gemma":[0.002532561,0.00024379134,0.000070756694,0.0008388679,0.00017243854,0.0017738693,0.0006156723,0.00018427694,0.00027641258],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003351561,0.0023566938,0.83995044,0.000018285124,0.000070725335,1.7257126e-7,0.0010825718,0.00032082092,0.005371096,0.011993387,0.09438781,0.044414468],"study_design_scores_gemma":[0.0004879148,0.00012868525,0.9345005,0.000007925356,0.000044110857,0.00001434336,0.0011797006,0.011575847,0.00006393185,0.03753353,0.014149552,0.0003139248],"about_ca_topic_score_codex":0.0002441478,"about_ca_topic_score_gemma":0.0001137195,"teacher_disagreement_score":0.09455007,"about_ca_system_score_codex":0.00086365826,"about_ca_system_score_gemma":0.007289428,"threshold_uncertainty_score":0.99833834},"labels":[],"label_agreement":null},{"id":"W2000242929","doi":"10.1007/s11092-014-9195-0","title":"Navigating dilemmas in transforming assessment practices: experiences of mathematics teachers in Ontario, Canada","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":34,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Manitoba; University of Ottawa","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Situated; Accountability; Judgement; Pedagogy; Assessment for learning; Curriculum; Best practice; Mathematics education; Faculty development; Professional development; Psychology; Formative assessment; Political science; Computer science","score_opus":0.0639113308409666,"score_gpt":0.4511814897350099,"score_spread":0.38727015889404326,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2000242929","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":[0.9820801,0.000021264545,0.00006413774,0.0023525753,0.00034637155,0.0007457044,0.0000013863945,0.000005854846,0.014382607],"genre_scores_gemma":[0.9953084,0.000008397139,0.0041297656,0.00007143129,0.000065250635,0.00032869953,0.00003082077,0.000005344497,0.000051894196],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99689734,0.00050486514,0.0006190287,0.00030942092,0.0014279217,0.00024142735],"domain_scores_gemma":[0.99795455,0.0011103676,0.00049010187,0.00014311139,0.00023275484,0.00006909706],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007483715,0.00013237714,0.00023646276,0.000049487146,0.00021417985,0.000091235015,0.0001902647,0.000062207735,0.0016756454],"category_scores_gemma":[0.00062791025,0.00013230642,0.000030273017,0.0002912103,0.00014097209,0.0006708196,0.000026118303,0.00033522802,4.2397306e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000002206206,0.0003801483,0.8405382,0.000020336744,0.0000064913806,4.6397556e-8,0.15126696,0.00004699335,0.000036690657,0.0046778126,0.000021149406,0.0030029512],"study_design_scores_gemma":[0.0003766728,0.000015716629,0.7005319,0.00006494693,0.0000134469765,1.4524126e-7,0.29464707,0.0018410666,0.000006590236,0.0018271178,0.0005634249,0.00011188592],"about_ca_topic_score_codex":0.9363547,"about_ca_topic_score_gemma":0.99399436,"teacher_disagreement_score":0.1433801,"about_ca_system_score_codex":0.002408778,"about_ca_system_score_gemma":0.007218331,"threshold_uncertainty_score":0.99923694},"labels":[],"label_agreement":null},{"id":"W2040697633","doi":"10.1007/s11092-011-9117-3","title":"Consistency of report card grades and external assessments in a Canadian province","year":2011,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":21,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Trent University; University of Toronto","funders":"Ministère de l’Éducation, Gouvernement de l’Ontario","keywords":"Report card; Psychology; Consistency (knowledge bases); Reading (process); Medical education; Grade level; Variance (accounting); Mathematics education; Medicine; Pedagogy; Mathematics; Political science","score_opus":0.2160058372944961,"score_gpt":0.5025141811864667,"score_spread":0.28650834389197066,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2040697633","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":[0.98100805,0.00018135134,0.00008878965,0.0022007504,0.0004796952,0.000840222,0.000026966827,0.000004825616,0.015169348],"genre_scores_gemma":[0.9936734,0.000014621455,0.005470767,0.00017429405,0.000059681104,0.00018490128,0.00004496698,0.000006483035,0.00037088135],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99593204,0.00035574078,0.0010712494,0.0006503976,0.0017273503,0.00026324918],"domain_scores_gemma":[0.99709475,0.00060078665,0.00046815505,0.00042245074,0.0011741371,0.00023973631],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009342493,0.0001706397,0.0002825066,0.00040621607,0.00015924848,0.00014032872,0.00025048744,0.00006430442,0.0016967918],"category_scores_gemma":[0.0011764738,0.00014572436,0.00004921446,0.0003820764,0.00022272566,0.0007954681,0.000076026314,0.00014384386,0.0000053750014],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009294571,0.0003574523,0.9734425,0.000014331551,0.000011473763,6.8437527e-7,0.0005256787,0.0000013884121,0.00019285576,0.013902221,0.00018235957,0.01135979],"study_design_scores_gemma":[0.0003273549,0.00006852171,0.89883196,0.000023337569,0.000022580502,0.0000126233535,0.0022369719,0.0009575937,0.000062367646,0.096992865,0.00033259625,0.00013123699],"about_ca_topic_score_codex":0.08166616,"about_ca_topic_score_gemma":0.09856468,"teacher_disagreement_score":0.08309064,"about_ca_system_score_codex":0.00056451553,"about_ca_system_score_gemma":0.007392032,"threshold_uncertainty_score":0.9992158},"labels":[],"label_agreement":null},{"id":"W2043024755","doi":"10.1007/s11092-013-9184-8","title":"The journey from rhetoric to reality: participatory evaluation in a development context","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Citizen journalism; Context (archaeology); Situated; Stakeholder; Sociology; Warrant; Rhetoric; Participatory action research; International development; Participatory evaluation; Engineering ethics; Conceptual framework; Epistemology; Political science; Social science; Public relations; Computer science; Engineering; Geography; Business","score_opus":0.3829268271012413,"score_gpt":0.5696211298539817,"score_spread":0.1866943027527404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2043024755","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":[0.9576222,0.00016670157,0.0012117317,0.03237257,0.0012076437,0.001470756,0.000009139568,0.000014152411,0.0059251143],"genre_scores_gemma":[0.99576014,0.000011691701,0.0008188507,0.0015095009,0.00027354847,0.00131592,0.00009406466,0.000009037915,0.00020725321],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9888198,0.002964711,0.0014746445,0.00069775386,0.005707079,0.00033606304],"domain_scores_gemma":[0.99110264,0.004836776,0.0004337771,0.00076210406,0.0026408662,0.00022386728],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.06747412,0.00021229907,0.00027989378,0.00023177454,0.00062161084,0.00066653313,0.00046428415,0.00007634415,0.0042337417],"category_scores_gemma":[0.008169542,0.0001473888,0.000055764463,0.0007548869,0.000099476136,0.0006588169,0.00012311428,0.00021942226,0.0002591008],"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":[0.000040770276,0.000305767,0.40191945,0.0000035498238,0.000019876607,1.4093768e-8,0.004641605,0.0012461955,0.00007665333,0.0071702255,0.0038153937,0.5807605],"study_design_scores_gemma":[0.00071416755,0.000030691954,0.80325055,0.000017410002,0.000023568862,3.2552109e-7,0.003767718,0.07966953,0.000035998553,0.052238986,0.06009216,0.00015888282],"about_ca_topic_score_codex":0.00030530157,"about_ca_topic_score_gemma":0.006621468,"teacher_disagreement_score":0.58060163,"about_ca_system_score_codex":0.0013693512,"about_ca_system_score_gemma":0.003834401,"threshold_uncertainty_score":0.9966765},"labels":[],"label_agreement":null},{"id":"W2076487748","doi":"10.1007/s11092-014-9200-7","title":"Personality and student performance on evaluation methods used in business administration courses","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Management and Marketing Education","field":"Business, Management and Accounting","cited_by":14,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Université de Sherbrooke","funders":"","keywords":"Personality; Preference; Psychology; Sample (material); Multilevel model; Latent variable; Control (management); Applied psychology; Medical education; Social psychology; Computer science; Machine learning; Medicine; Artificial intelligence; Statistics; Mathematics","score_opus":0.06513401286280288,"score_gpt":0.446581531766921,"score_spread":0.3814475189041181,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2076487748","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":[0.9696934,0.00003893174,0.00022929834,0.009403312,0.0007499734,0.0012498586,0.0000013594188,0.000031990607,0.018601833],"genre_scores_gemma":[0.99651974,0.00001676392,0.0010431542,0.0008187844,0.0007036331,0.00050658116,0.00030258563,0.000013559732,0.00007520477],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9972629,0.0003872079,0.00044899687,0.00058712624,0.0011095167,0.00020423533],"domain_scores_gemma":[0.9979853,0.00047672878,0.0003169525,0.00026840964,0.0009282555,0.000024346604],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.015041634,0.0002242176,0.00021045041,0.00030168888,0.00028636467,0.0004485783,0.00012675564,0.000069158596,0.0009954715],"category_scores_gemma":[0.0010897984,0.00021769723,0.000028280103,0.00047879762,0.00008756985,0.0012977428,0.00006971425,0.00014333904,0.000015153023],"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":[0.000054154138,0.0007273735,0.8874009,0.00024160229,0.000014495151,1.4108005e-8,0.00017471262,0.00022550886,0.000076284705,0.021055741,0.00044839407,0.089580774],"study_design_scores_gemma":[0.00087855704,0.000022855798,0.9096548,0.000054964592,0.00010059365,4.601344e-7,0.00061586115,0.07890879,0.000006861623,0.007065797,0.0024777658,0.00021266926],"about_ca_topic_score_codex":0.00023439359,"about_ca_topic_score_gemma":0.00031615476,"teacher_disagreement_score":0.089368105,"about_ca_system_score_codex":0.00032192635,"about_ca_system_score_gemma":0.00033528332,"threshold_uncertainty_score":0.99991775},"labels":[],"label_agreement":null},{"id":"W2088210129","doi":"10.1007/s11092-013-9181-y","title":"Teacher supervision practices and characteristics of in-school supervisors in Uganda","year":2014,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":14,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Supervisor; Perception; Professional development; Psychology; Directive; Descriptive statistics; Medical education; Pedagogy; Medicine; Political science","score_opus":0.13688318478546668,"score_gpt":0.4800151790265387,"score_spread":0.343131994241072,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2088210129","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":[0.9512207,0.00016317857,0.0000059886793,0.045163106,0.00019957287,0.00036147184,0.000002230121,0.00000615359,0.0028776252],"genre_scores_gemma":[0.99897534,0.0001260281,0.00031850356,0.0002231881,0.00017480108,0.00007962661,0.000023449482,0.000004109285,0.00007493604],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9978762,0.0008931857,0.0003451729,0.00025676424,0.0004913215,0.00013734824],"domain_scores_gemma":[0.9985533,0.00078218314,0.00021204718,0.00011710737,0.0002656991,0.00006970895],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0066137984,0.000088006855,0.00016602814,0.00012211352,0.000112393725,0.00006489215,0.00007968751,0.000064683794,0.0011771447],"category_scores_gemma":[0.003912318,0.000086158994,0.00001657024,0.00021937901,0.0001852409,0.00045956546,0.000024400724,0.00016410659,0.0000031687066],"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":[0.000007350328,0.00035560306,0.94725376,0.000029278046,0.000003479975,5.9496976e-9,0.03878725,8.988889e-7,0.000032820248,0.0019642916,0.000060790957,0.011504457],"study_design_scores_gemma":[0.00031871788,0.000014722069,0.90526664,0.000023094231,0.000010998325,1.2155905e-7,0.08958882,0.0003726921,0.0000018705675,0.0025687038,0.0017540078,0.00007957889],"about_ca_topic_score_codex":0.0032899333,"about_ca_topic_score_gemma":0.0051791086,"teacher_disagreement_score":0.05080157,"about_ca_system_score_codex":0.00023877731,"about_ca_system_score_gemma":0.00076279143,"threshold_uncertainty_score":0.9997359},"labels":[],"label_agreement":null},{"id":"W2183435816","doi":"10.1007/s11092-015-9233-6","title":"Teacher assessment literacy: a review of international standards and measures","year":2015,"lang":"en","type":"review","venue":"Educational Assessment Evaluation and Accountability","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":253,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Literacy; Educational assessment; Mainland; Psychology; Pedagogy; Geography","score_opus":0.13742489297286617,"score_gpt":0.5795117938372106,"score_spread":0.44208690086434443,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2183435816","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000070910195,0.9466183,0.000028332346,0.0029385,0.0010372748,0.0023122719,0.00021035915,0.000021447728,0.04676258],"genre_scores_gemma":[0.00059036084,0.9954776,0.0012236641,0.00016560507,0.0007095266,0.00062387163,0.0007282111,0.000021239792,0.00045990988],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9920936,0.0017142249,0.0010788116,0.00064115244,0.00422735,0.0002448302],"domain_scores_gemma":[0.99441123,0.0006462519,0.00087784807,0.00033797073,0.0035319782,0.00019469377],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.019898808,0.0003648247,0.001121853,0.00020339819,0.00028392562,0.00029252144,0.00040092075,0.0002018631,0.0043696505],"category_scores_gemma":[0.0010661912,0.0003140791,0.00021191448,0.00038542596,0.00035246304,0.00066303986,0.00017300271,0.00037824683,0.000004934088],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000036664612,0.00041655597,0.008758956,0.009534821,0.00020637889,8.3757435e-8,0.0010178514,8.103374e-8,9.133324e-8,0.003787145,0.010471021,0.9658033],"study_design_scores_gemma":[0.00029457634,0.000026800499,0.0099680135,0.010690639,0.0008722066,0.0000017452187,0.00086513365,0.00002604948,9.680338e-9,0.0015047371,0.9754583,0.00029174858],"about_ca_topic_score_codex":0.0002951071,"about_ca_topic_score_gemma":0.00010423875,"teacher_disagreement_score":0.9655116,"about_ca_system_score_codex":0.0022665164,"about_ca_system_score_gemma":0.01331173,"threshold_uncertainty_score":0.99993116},"labels":[],"label_agreement":null},{"id":"W2531903003","doi":"10.1007/s11092-016-9247-8","title":"Two teacher quality measures and the role of context: evidence from Chile","year":2016,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"School Choice and Performance","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Christian Studies","funders":"","keywords":"Context (archaeology); Quality (philosophy); Teacher quality; Psychology; Geography; Political science; Business; Archaeology; Epistemology; Philosophy; Marketing","score_opus":0.09023241200854629,"score_gpt":0.47158391345562656,"score_spread":0.3813515014470803,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2531903003","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":[0.9710794,0.0020952486,0.00003375409,0.021918422,0.000152341,0.0003830745,0.000016413316,0.0000073614774,0.004313952],"genre_scores_gemma":[0.9988016,0.0002513962,0.0001457087,0.00017299734,0.00040165026,0.00010381558,0.0000056258154,0.0000022785466,0.00011496035],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9976555,0.000984271,0.00025322172,0.00021041888,0.0008000966,0.000096456046],"domain_scores_gemma":[0.9968752,0.002260586,0.00016230885,0.00017818697,0.00047296847,0.00005071356],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008321559,0.00006852349,0.00012219812,0.000018269375,0.00032309318,0.0000645763,0.00012620825,0.000033882552,0.002225778],"category_scores_gemma":[0.0019646601,0.000039125556,0.000030395373,0.00007947321,0.0005418906,0.00060254586,0.000028137032,0.00006793405,0.0000063540238],"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":[0.000040825675,0.000057370475,0.9004488,0.0000027963426,0.0000120002005,8.8755214e-10,0.0039895144,6.1465045e-7,0.00034801563,0.013218014,0.000110270594,0.081771806],"study_design_scores_gemma":[0.0006471403,0.000005868363,0.93767864,0.000033538472,0.00002663999,5.3963404e-8,0.0057693,0.00012855574,0.000062388644,0.05316682,0.002421165,0.000059875467],"about_ca_topic_score_codex":0.015597951,"about_ca_topic_score_gemma":0.007764817,"teacher_disagreement_score":0.08171193,"about_ca_system_score_codex":0.00013294007,"about_ca_system_score_gemma":0.00083212694,"threshold_uncertainty_score":0.9986863},"labels":[],"label_agreement":null},{"id":"W3044087203","doi":"10.1007/s11092-020-09329-5","title":"Is Canada really an education superpower? The impact of non-participation on results from PISA 2015","year":2020,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Youth Substance Use and School Attendance","field":"Social Sciences","cited_by":31,"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":"H2020 European Research Council","keywords":"Superpower; Political science; Inclusion (mineral); Scale (ratio); Test (biology); Sample (material); Population; Economic growth; Psychology; Geography; Mathematics education; Demography; Sociology; China; Social psychology; Economics","score_opus":0.07773513248884605,"score_gpt":0.4788059318311023,"score_spread":0.40107079934225626,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3044087203","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":[0.9503996,0.00004351757,0.000021088264,0.042059653,0.00028918142,0.00055910385,0.00024529846,0.000007970687,0.0063746143],"genre_scores_gemma":[0.9970747,0.000020871936,0.0002568468,0.0016238578,0.00050633674,0.000078564306,0.0003885721,0.0000054942075,0.000044756034],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9976058,0.00044373694,0.00038295114,0.00035335365,0.0010612156,0.0001529535],"domain_scores_gemma":[0.99800956,0.0004451429,0.00023336227,0.0002629363,0.0008731692,0.00017581033],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017670484,0.000114332004,0.0001334045,0.000022357097,0.00036297218,0.00009176704,0.00019528675,0.000048409274,0.0012353656],"category_scores_gemma":[0.00063221686,0.00008710341,0.000044222546,0.00025717495,0.00010233364,0.00048614692,0.000013533615,0.00013288476,0.0000065060913],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032259067,0.0010664762,0.8137749,0.000014227352,0.000075342854,5.0299228e-8,0.077044874,0.00095296034,0.000537086,0.0029965704,0.08162513,0.021589786],"study_design_scores_gemma":[0.00030489944,0.00006389511,0.9830563,0.0000136291155,0.000027131317,2.7555929e-8,0.008178005,0.0041829906,0.000032862918,0.0031034867,0.0009371904,0.000099592806],"about_ca_topic_score_codex":0.63213134,"about_ca_topic_score_gemma":0.3527885,"teacher_disagreement_score":0.27934283,"about_ca_system_score_codex":0.0007392292,"about_ca_system_score_gemma":0.012741003,"threshold_uncertainty_score":0.99967766},"labels":[],"label_agreement":null},{"id":"W3188740232","doi":"10.1007/s11092-021-09368-6","title":"“I am here for the students”: principals’ perception of accountability amid work intensification","year":2021,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Teacher Education and Leadership Studies","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; University of Manitoba; University of British Columbia","funders":"","keywords":"Accountability; Perception; Work (physics); Context (archaeology); Public relations; Psychology; Political science; Focus group; Task (project management); Pedagogy; Sociology; Management; Law","score_opus":0.2359750747103962,"score_gpt":0.5289763041093049,"score_spread":0.29300122939890866,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3188740232","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":[0.9000383,0.0004638981,0.00058654585,0.0954692,0.000991618,0.0014336236,0.00001671868,0.000024245706,0.000975861],"genre_scores_gemma":[0.99640805,0.00012536338,0.001203966,0.00050578266,0.00045404347,0.0007395534,0.00012327635,0.000008060887,0.00043192855],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.997046,0.0007439008,0.00045842683,0.00042169588,0.001130741,0.00019925929],"domain_scores_gemma":[0.9949026,0.0016136246,0.0002535675,0.00036416957,0.0027971144,0.00006893422],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0071011977,0.00013249069,0.0001831941,0.00004347351,0.0007054121,0.00017884026,0.00023752045,0.00008528513,0.001734282],"category_scores_gemma":[0.0021915238,0.00011112062,0.0000967566,0.00040993156,0.00047716906,0.0003376066,0.000052192732,0.00015885445,0.000007368194],"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":[0.000024836472,0.0008587654,0.91327494,0.000050851846,0.000059976184,4.852417e-9,0.054341406,0.00001674193,0.00011625643,0.007547404,0.0011062282,0.022602556],"study_design_scores_gemma":[0.00022708344,0.000009088024,0.711866,0.000014467535,0.000076084485,1.8008484e-7,0.27907258,0.00007901698,0.000009363783,0.0027199243,0.0058420664,0.00008411897],"about_ca_topic_score_codex":0.00050230575,"about_ca_topic_score_gemma":0.0010909438,"teacher_disagreement_score":0.22473118,"about_ca_system_score_codex":0.00070978055,"about_ca_system_score_gemma":0.0018768104,"threshold_uncertainty_score":0.9991783},"labels":[],"label_agreement":null},{"id":"W4283698056","doi":"10.1007/s11092-022-09389-9","title":"Mapping the constellation of assessment discourses: a scoping review study on assessment competence, literacy, capability, and identity","year":2022,"lang":"en","type":"review","venue":"Educational Assessment Evaluation and Accountability","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":31,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University; Memorial University of Newfoundland","funders":"","keywords":"Competence (human resources); Operationalization; Construct (python library); Authentic assessment; Literacy; Rubric; Pedagogy; Identity (music); Psychology; Engineering ethics; Sociology; Epistemology; Social psychology; Computer science; Engineering","score_opus":0.19602477827829456,"score_gpt":0.5663204934521253,"score_spread":0.37029571517383075,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283698056","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006654717,0.9474011,0.0000959048,0.0027519553,0.00170517,0.022965105,0.000119422024,0.000047145077,0.018259477],"genre_scores_gemma":[0.04627483,0.9500688,0.00048941374,0.00018947564,0.00032326902,0.0021338286,0.00043012755,0.000029187437,0.00006109835],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98495024,0.0075763287,0.0018803665,0.0011431129,0.0040557683,0.00039418635],"domain_scores_gemma":[0.991549,0.005041269,0.001760233,0.00080450735,0.00069364806,0.00015131474],"candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.026448311,0.00055633433,0.0015899268,0.00029970732,0.0018484618,0.0006231524,0.0007120628,0.00013223491,0.0047709853],"category_scores_gemma":[0.00051713263,0.0004339578,0.00032415526,0.001160552,0.0006946883,0.0010531697,0.0004963881,0.00080059166,0.0000039112506],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000073141982,0.005064021,0.22775,0.038322926,0.0005448615,6.639996e-7,0.009985577,0.000009465075,4.66841e-7,0.020005343,0.00045767863,0.69785166],"study_design_scores_gemma":[0.0014302853,0.00046287652,0.7079127,0.07314509,0.003811967,0.0000073525034,0.0634001,0.00031660774,2.6675508e-8,0.0049321265,0.1431074,0.0014734727],"about_ca_topic_score_codex":0.00063796237,"about_ca_topic_score_gemma":0.00046514627,"teacher_disagreement_score":0.69637823,"about_ca_system_score_codex":0.0019554193,"about_ca_system_score_gemma":0.008670215,"threshold_uncertainty_score":0.99981123},"labels":[],"label_agreement":null},{"id":"W4417293613","doi":"10.1007/s11092-025-09471-y","title":"Investigating the relationship between educational inequity and teacher participation in professional development: A cross-national and quasi-experimental approach using TIMSS","year":2025,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Early Childhood Education and Development","field":"Social Sciences","cited_by":0,"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":"Göteborgs Universitet; Vetenskapsrådet","keywords":"Socioeconomic status; Student achievement; Academic achievement; Professional development; Higher education; Social class","score_opus":0.18038049535966852,"score_gpt":0.5374982171340111,"score_spread":0.35711772177434264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417293613","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":[0.9836693,0.00027994928,0.00015107077,0.010808772,0.00040769053,0.0009405888,0.0000056165554,0.000014041911,0.0037230093],"genre_scores_gemma":[0.9933163,0.0000033508209,0.0043343683,0.00038506338,0.00023606696,0.0004216654,0.00018766442,0.0000065034546,0.0011090196],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99689704,0.00090099155,0.0005634958,0.00045292068,0.0009594042,0.00022612227],"domain_scores_gemma":[0.9973276,0.0017720449,0.00019834071,0.00011020504,0.00047342601,0.00011835717],"candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.007677886,0.00016030941,0.00015192915,0.00018496341,0.001636193,0.00033574205,0.000117916905,0.00011455773,0.0002810466],"category_scores_gemma":[0.002693487,0.00014432742,0.000019649708,0.0004878996,0.00048775232,0.0007069399,0.000095711344,0.00027419138,0.00000292274],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000033635679,0.000552112,0.9329837,0.000025098563,0.000012629458,2.5480522e-9,0.01356413,0.0000319031,0.0000079838965,0.05105527,0.00018748498,0.0015763296],"study_design_scores_gemma":[0.00041935794,0.0000061755823,0.96706414,0.000049183305,0.000016501568,4.3096716e-7,0.005094888,0.0022745254,0.0000053939802,0.024385177,0.0005369765,0.00014722854],"about_ca_topic_score_codex":0.0004698342,"about_ca_topic_score_gemma":0.00026019945,"teacher_disagreement_score":0.03408047,"about_ca_system_score_codex":0.001046116,"about_ca_system_score_gemma":0.008588704,"threshold_uncertainty_score":0.99966353},"labels":[],"label_agreement":null},{"id":"W944434266","doi":"10.1007/s11092-015-9224-7","title":"Juggling multiple accountability systems: how three principals manage these tensions in Ontario, Canada","year":2015,"lang":"en","type":"article","venue":"Educational Assessment Evaluation and Accountability","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":31,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"York University; Western University","funders":"","keywords":"Accountability; Negotiation; Mandate; Variety (cybernetics); Public relations; Work (physics); Public administration; Political science; Sociology; Engineering; Law","score_opus":0.2865166200978165,"score_gpt":0.45404375441143446,"score_spread":0.167527134313618,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W944434266","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":[0.97628397,0.00015705409,0.00025388936,0.015524504,0.0023494223,0.0019091271,0.0000524029,0.00001834938,0.003451269],"genre_scores_gemma":[0.9959942,0.000002620113,0.0012567984,0.00046723292,0.00022036862,0.0007176913,0.00020514733,0.000013984283,0.0011219258],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99176556,0.0006977224,0.0011763151,0.0011094229,0.0048167906,0.0004342116],"domain_scores_gemma":[0.99185085,0.0030406602,0.00053876307,0.000952447,0.0032631294,0.0003541247],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01659711,0.00033233617,0.0004661094,0.00027416126,0.0002651783,0.0008296712,0.0005953899,0.000102623024,0.0014626089],"category_scores_gemma":[0.005165966,0.00026764694,0.000075664524,0.00067591015,0.000131463,0.001453105,0.00024002718,0.0004005366,0.000019655587],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.000021520145,0.00051289884,0.9897052,0.000012483294,0.000018995828,2.3583546e-7,0.0008886769,0.001094104,0.00003720897,0.002048631,0.0041546156,0.0015054509],"study_design_scores_gemma":[0.00084066764,0.00003442592,0.91862047,0.000025152423,0.000028741637,0.000002577681,0.012741779,0.030554049,0.000005045766,0.023116719,0.013750481,0.00027991727],"about_ca_topic_score_codex":0.8374575,"about_ca_topic_score_gemma":0.9880023,"teacher_disagreement_score":0.15054484,"about_ca_system_score_codex":0.006059891,"about_ca_system_score_gemma":0.017533507,"threshold_uncertainty_score":0.9999776},"labels":[],"label_agreement":null}]}