{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":233,"total_is_capped":false,"direct_labels_cover":233,"predictions_cover":233,"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":"4d5d67e28eb3","filters":{"category":"metaepi_broad"}},"results":[{"id":"W2068264290","doi":"10.1002/jrsm.1123","title":"A scoping review of scoping reviews: advancing the approach and enhancing the consistency","year":2014,"lang":"en","type":"review","venue":"Research Synthesis Methods","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3082,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Guelph; Public Health Agency of Canada","funders":"Ontario Ministry of Agriculture, Food and Rural Affairs","keywords":"Consistency (knowledge bases); Computer science; Management science; Data science; Artificial intelligence; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.923851977769396,"gpt":0.7277039479738097,"spread":0.1961480297955863,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8674546,0.001038748,0.03233648,0.001053986,0.001231888,0.001237589,0.007614791,0.0003332912,0.001921087],"category_scores_gemma":[0.7785432,0.0003363213,0.00714746,0.006299767,0.00103737,0.0002395186,0.00170997,0.001707473,0.0004517362],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001302364,"about_ca_system_score_gemma":0.001512423,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001762829,"about_ca_topic_score_gemma":0.00001686503,"domain_scores_codex":[0.3442413,0.6025283,0.03582132,0.003359887,0.01253693,0.001512236],"domain_scores_gemma":[0.1675581,0.7719051,0.0312148,0.02374607,0.00490973,0.0006662289],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[4.055199e-7,0.000009484009,5.320543e-7,0.3897209,0.0001654179,4.844143e-7,0.00005685872,1.866977e-7,0.00000125018,0.0003091281,0.001830547,0.6079048],"study_design_scores_gemma":[0.00001265201,0.00001099199,2.307412e-7,0.5386342,0.001300408,0.00003874799,0.0001469155,0.00007943699,0.000005345767,0.0002387714,0.4593396,0.0001926586],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[7.881358e-8,0.8912809,0.08701174,0.0003627573,0.0001505349,0.0130855,0.000007171405,0.000006399373,0.008094846],"genre_scores_gemma":[2.586249e-7,0.8728453,0.1229208,0.0001712135,0.0002188215,0.003064823,0.000002458247,0.00006128768,0.0007150216],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6077121,"threshold_uncertainty_score":0.9999089,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2612870289","doi":"10.15173/ijsap.v1i1.3119","title":"A Systematic Literature Review of Students as Partners in Higher Education","year":2017,"lang":"en","type":"article","venue":"International Journal for Students as Partners","topic":"Higher Education Practises and Engagement","field":"Social Sciences","cited_by":597,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"General partnership; Reciprocity (cultural anthropology); Systematic review; Higher education; Space (punctuation); Scale (ratio); Pedagogy; Psychology; Medical education; Sociology; Political science; Medicine; MEDLINE; Computer science; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.1026937746921305,"gpt":0.6131118316714596,"spread":0.5104180569793291,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.00267441,0.0001433916,0.0003060173,0.0001892955,0.0005447002,0.001082958,0.00196155,0.00007147741,0.0006139565],"category_scores_gemma":[0.001414871,0.0001253805,0.0001832566,0.0001109105,0.00008290327,0.0007727669,0.0001271166,0.000235543,0.00006191908],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004063376,"about_ca_system_score_gemma":0.0004708322,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002300902,"about_ca_topic_score_gemma":0.00007343676,"domain_scores_codex":[0.9966361,0.0002940873,0.0007119327,0.0002032014,0.00191645,0.0002382582],"domain_scores_gemma":[0.9975121,0.0001489402,0.001033972,0.0002651644,0.0008606718,0.0001791403],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0004669352,0.006706655,0.3956757,0.02772327,0.003528673,0.000200871,0.05761908,0.00003708476,0.00005897284,0.299568,0.2017248,0.006689937],"study_design_scores_gemma":[0.002292349,0.0001743444,0.088883,0.1478004,0.0002676437,0.00004801164,0.009142586,0.000001597093,0.00001770954,0.004704247,0.7461888,0.0004794053],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5250037,0.05662496,0.00004124166,0.1057296,0.06610364,0.006988013,0.00005126692,0.00007907859,0.2393784],"genre_scores_gemma":[0.8852469,0.04298728,0.0001284705,0.002593906,0.0009054852,0.000279552,0.00001913421,0.0000191876,0.06782003],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5444639,"threshold_uncertainty_score":0.999954,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3209288188","doi":"10.2196/28102","title":"The Application of Human-Centered Design Approaches in Health Research and Innovation: A Narrative Review of Current Practices","year":2021,"lang":"en","type":"review","venue":"JMIR mhealth and uhealth","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":357,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"Radboud Universitair Medisch Centrum; Radboud Universiteit","keywords":"CINAHL; PsycINFO; Participatory design; User-centered design; Context (archaeology); Computer science; Knowledge management; Research design; Health care; Management science; MEDLINE; Medicine; Psychological intervention; Sociology; Engineering; Nursing; Social science; Human–computer interaction","retraction":null,"screen_n_in":null,"score":{"opus":0.963688333660892,"gpt":0.7975905777322816,"spread":0.1660977559286104,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.04647373,0.0003579445,0.002474478,0.001038669,0.001999144,0.00002096343,0.0004578835,0.0002576748,0.0000166948],"category_scores_gemma":[0.004600867,0.0002554073,0.00006289753,0.005582773,0.0005486501,0.0001971044,0.0002886175,0.002377139,0.000004387242],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007642948,"about_ca_system_score_gemma":0.01877363,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005995181,"about_ca_topic_score_gemma":0.0004260968,"domain_scores_codex":[0.9717125,0.01879405,0.006206401,0.0008782478,0.001071252,0.001337555],"domain_scores_gemma":[0.9754317,0.0106706,0.0113487,0.0008200931,0.001257014,0.0004718754],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001373666,0.00008445941,0.0001022061,0.5004687,0.00000843753,1.696306e-7,0.006490256,1.740318e-8,6.54307e-8,0.01818507,0.004712044,0.4699348],"study_design_scores_gemma":[0.0004463387,0.0004160137,0.0002420843,0.1268355,0.00002849053,0.000004918217,0.005180509,0.000009739907,9.653961e-8,0.000369656,0.866329,0.0001376664],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00009078334,0.9750678,0.0002295452,0.01189185,0.00015549,0.01231812,0.00009511812,0.00001519026,0.0001361162],"genre_scores_gemma":[0.0001947159,0.9931436,0.0004823693,0.001560068,0.0001650227,0.004218901,0.000134658,0.00003428314,0.00006638929],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.861617,"threshold_uncertainty_score":0.9999898,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"opus","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2007317588","doi":"10.1007/s13187-010-0183-x","title":"Effective Teaching Strategies and Methods of Delivery for Patient Education: A Systematic Review and Practice Guideline Recommendations","year":2010,"lang":"en","type":"review","venue":"Journal of Cancer Education","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":334,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"London Health Sciences Centre; McMaster University; Grand River Hospital; Sudbury Regional Hospital; Ontario Clinical Oncology Group; University Health Network","funders":"Ontario Ministry of Health and Long-Term Care; University Health Network; Cancer Care Ontario","keywords":"Medicine; Guideline; Medical education; Teaching method; Anxiety; Recall; Multimedia; Psychology; Pedagogy; Computer science; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.2071370985328947,"gpt":0.6437488558854468,"spread":0.4366117573525521,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004217071,0.0002926648,0.002119932,0.000381056,0.0004618234,0.00003541703,0.00029403,0.0003014927,0.00002484458],"category_scores_gemma":[0.01175137,0.0002260799,0.000204304,0.0002322201,0.00005688201,0.0007881903,0.0001016117,0.001573176,8.696946e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006367931,"about_ca_system_score_gemma":0.01439292,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003267706,"about_ca_topic_score_gemma":0.00005178377,"domain_scores_codex":[0.9876434,0.008012318,0.00370116,0.000232353,0.0002234039,0.0001873151],"domain_scores_gemma":[0.9729791,0.01095341,0.01092058,0.0005659101,0.004411796,0.0001691809],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000008846054,0.00008795445,0.00000243373,0.4528334,0.0001189032,1.313494e-8,0.00229466,8.086494e-8,4.294907e-7,0.0004484499,0.005348551,0.5388563],"study_design_scores_gemma":[0.00008039662,0.0001144684,0.000001976298,0.3788385,0.004338831,0.00004103794,0.01540907,0.000002108148,1.861744e-7,0.0002678575,0.6007901,0.0001153768],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000003902049,0.9799297,0.0008183412,0.007005887,0.002747225,0.009154704,0.00005230776,0.000008578358,0.000279308],"genre_scores_gemma":[0.00001084163,0.86879,0.1238175,0.001848888,0.0003733793,0.00499999,0.00007170882,0.00003838088,0.00004926429],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.5954416,"threshold_uncertainty_score":0.9965731,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2103669116","doi":"10.1186/1748-5908-9-54","title":"Applications of social constructivist learning theories in knowledge translation for healthcare professionals: a scoping review","year":2014,"lang":"en","type":"review","venue":"Implementation Science","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":326,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre for Interdisciplinary Research in Rehabilitation; McGill University","funders":"McGill University","keywords":"CINAHL; Constructivism (international relations); Social constructivism; Knowledge translation; PsycINFO; Health care; Constructivist teaching methods; Psychological intervention; Context (archaeology); Epistemology; MEDLINE; Medicine; Psychology; Sociology; Knowledge management; Social science; Nursing; Computer science; Pedagogy; Politics; Teaching method; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.7546677192685569,"gpt":0.793139881474038,"spread":0.03847216220548111,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.01644733,0.0003463759,0.001773033,0.0009886165,0.002353709,0.00002910132,0.0007322467,0.0001823857,0.00029982],"category_scores_gemma":[0.002112428,0.0003119932,0.0002059938,0.00383981,0.0006618915,0.0005256571,0.0001496448,0.0006453292,0.00004957858],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000710431,"about_ca_system_score_gemma":0.01433348,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008417993,"about_ca_topic_score_gemma":0.0008170929,"domain_scores_codex":[0.9902505,0.003031649,0.004003828,0.0008307148,0.0008522787,0.001031082],"domain_scores_gemma":[0.986463,0.008374379,0.003492392,0.0003421881,0.001125848,0.0002021731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003005647,0.00001287423,0.00006678832,0.2298289,0.000004304825,2.967514e-8,0.005938835,6.805174e-8,0.00000136916,0.01525436,0.0001896872,0.7486998],"study_design_scores_gemma":[0.0007997851,0.00006354207,0.00003893891,0.1474262,0.00009105034,0.000001783144,0.00564789,0.00001020417,0.000002719405,0.0006067548,0.8449916,0.0003194692],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000009136227,0.9428148,0.02196604,0.002329511,0.0004731959,0.03102084,0.0004743219,0.00008063608,0.0008314871],"genre_scores_gemma":[0.0004901838,0.9724782,0.002692529,0.001496888,0.0002795062,0.02213664,0.0002858266,0.00004919127,0.0000910153],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.844802,"threshold_uncertainty_score":0.9999332,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2150288182","doi":"10.1016/s0140-6736(14)61133-8","title":"Health interventions for people who are homeless","year":2014,"lang":"en","type":"review","venue":"The Lancet","topic":"Homelessness and Social Issues","field":"Health Professions","cited_by":314,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Respite care; Psychological intervention; Assertive community treatment; Mental health; Medicine; Intervention (counseling); Nursing; Case management; Health care; Mental illness; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.2372561639754704,"gpt":0.5367720883114878,"spread":0.2995159243360174,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002073772,0.0002862691,0.009537031,0.00007081761,0.001300102,0.00002875251,0.0007892638,0.0003303427,0.0001473052],"category_scores_gemma":[0.0002274564,0.0001699108,0.0008335217,0.0002384359,0.00004365573,0.00003147717,0.0001714464,0.0008823267,0.0003566426],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001797395,"about_ca_system_score_gemma":0.0003626789,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000483911,"about_ca_topic_score_gemma":0.01728343,"domain_scores_codex":[0.9962716,0.001617264,0.00094076,0.0002825887,0.0001499113,0.0007379496],"domain_scores_gemma":[0.9964438,0.001301963,0.00124521,0.000790848,0.0001117734,0.0001063867],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000005766733,0.00002756501,0.0001397839,0.3443899,0.0001041707,2.119003e-7,0.007132868,8.689199e-8,4.632972e-10,0.01240928,0.02950389,0.6062865],"study_design_scores_gemma":[0.0002942823,0.00002821873,0.0002073076,0.1101814,0.0001469301,8.038227e-8,0.004827224,0.000001741041,1.594894e-9,0.0006398016,0.8835134,0.00015964],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001428188,0.9914964,0.0006828133,0.002025531,0.001178768,0.002721892,0.0008141689,0.0001227475,0.0009433565],"genre_scores_gemma":[0.0003535554,0.9887779,0.00004780275,0.0003272831,0.005338464,0.002111166,0.0002694221,0.00008149357,0.002692886],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8540095,"threshold_uncertainty_score":0.9999468,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3178875310","doi":"10.1186/s13012-021-01140-0","title":"Barriers and facilitators to implementing evidence-based guidelines in long-term care: a qualitative evidence synthesis","year":2021,"lang":"en","type":"review","venue":"Implementation Science","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":291,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Research Institute for Aging; University of Waterloo; McMaster University; Canada Research Chairs; Centre for Excellence in Mining Innovation; University of Toronto; Dalhousie University","funders":"Norwegian Institute of Public Health; Canadian Institutes of Health Research; McMaster University","keywords":"CINAHL; Knowledge translation; Medicine; Nursing; MEDLINE; Staffing; Long-term care; Critical appraisal; Implementation research; Health administration; Health services research; Inclusion (mineral); Teamwork; Qualitative research; Health care; Nursing research; Public health; Psychological intervention; Psychology; Alternative medicine; Knowledge management","retraction":null,"screen_n_in":null,"score":{"opus":0.8042982745295546,"gpt":0.7805708647425855,"spread":0.02372740978696908,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03234766,0.0007824203,0.00201825,0.002511799,0.002631062,0.0003584662,0.001604683,0.0002230999,0.002186972],"category_scores_gemma":[0.07791011,0.0007242605,0.000236378,0.007652044,0.0006734891,0.002593469,0.001145337,0.0007554796,0.0002048864],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003706767,"about_ca_system_score_gemma":0.02555759,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00346789,"about_ca_topic_score_gemma":0.006321081,"domain_scores_codex":[0.9793125,0.007494509,0.005386986,0.002354356,0.002692187,0.002759431],"domain_scores_gemma":[0.9671842,0.0254735,0.002022417,0.001084483,0.002759002,0.001476398],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001018871,0.000004086058,0.01342178,0.01441574,0.00001853412,0.00001375972,0.05596231,0.000002356002,0.0000272496,0.0002003159,0.001511877,0.9144118],"study_design_scores_gemma":[0.0009781811,0.0002370393,0.005680111,0.1137033,0.0003720573,0.000009634366,0.2171958,0.0000156326,0.00009621408,0.00002122942,0.6600307,0.001660108],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.007346295,0.9702848,0.002757855,0.006726724,0.0009968373,0.01059284,0.001090683,0.0001351227,0.00006889201],"genre_scores_gemma":[0.002938198,0.9693328,0.00929221,0.009307458,0.0002284686,0.008664646,0.00004762102,0.00008401048,0.0001046043],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9127517,"threshold_uncertainty_score":0.9995208,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2974839292","doi":"10.1136/bmj.l5221","title":"Design characteristics, risk of bias, and reporting of randomised controlled trials supporting approvals of cancer drugs by European Medicines Agency, 2014-16: cross sectional analysis","year":2019,"lang":"en","type":"article","venue":"BMJ","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":244,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"Medical Research Council; Economic and Social Research Council; U.S. Department of the Treasury; National Institute for Health and Care Research; Bill and Melinda Gates Foundation; Harvard University; National Institutes of Health; National Science Foundation","keywords":"Medicine; Clinical trial; Randomized controlled trial; Reporting bias; Research design; Clinical endpoint; Publication bias; MEDLINE; Family medicine; Meta-analysis; Internal medicine; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.4469013499055617,"gpt":0.5635049915159488,"spread":0.1166036416103871,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1663085,0.0002378437,0.005939792,0.000176304,0.00004319032,0.00002023777,0.0001865249,0.0001362806,0.0009731987],"category_scores_gemma":[0.6872705,0.0001606954,0.0009832832,0.0002804372,0.0002629868,0.00004142207,0.00007482371,0.0001846762,0.000001676016],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002159463,"about_ca_system_score_gemma":0.0001266,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001539499,"about_ca_topic_score_gemma":0.00000197087,"domain_scores_codex":[0.9723406,0.01112377,0.01503004,0.0004699758,0.0007687268,0.0002668923],"domain_scores_gemma":[0.7908946,0.1643804,0.04320049,0.0005565056,0.000851522,0.0001165057],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.02712101,0.0004545119,0.8876359,0.001962558,0.01576965,0.000008532779,0.001055802,0.0001426707,0.04289712,0.00046169,0.005372381,0.01711813],"study_design_scores_gemma":[0.2687288,0.001145574,0.4423322,0.001504433,0.04181717,0.00001126469,0.0006337537,0.01554816,0.09109859,0.1356896,0.00006498877,0.00142549],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8103694,0.0002386376,0.1845033,0.00005685972,0.0004933461,0.00312679,0.0005046466,0.00002384977,0.0006831415],"genre_scores_gemma":[0.8246808,0.0002006947,0.1742592,0.00002292837,0.000314357,0.00007849102,0.00001247642,0.00003724221,0.0003938378],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5209619,"threshold_uncertainty_score":0.99994,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3182386410","doi":"10.1097/ccm.0000000000005195","title":"Coronavirus Disease 2019 as Cause of Viral Sepsis: A Systematic Review and Meta-Analysis*","year":2021,"lang":"en","type":"review","venue":"Critical Care Medicine","topic":"COVID-19 Clinical Research Studies","field":"Medicine","cited_by":179,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Medicine; Meta-analysis; Coronavirus disease 2019 (COVID-19); Sepsis; Coronavirus; 2019-20 coronavirus outbreak; Disease; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Intensive care medicine; Virology; Pandemic; Immunology; Infectious disease (medical specialty); Internal medicine; Outbreak","retraction":null,"screen_n_in":null,"score":{"opus":0.361154865756216,"gpt":0.5950817646424796,"spread":0.2339268988862636,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001862929,0.0008360071,0.02576994,0.0004492202,0.00008531653,0.00001795982,0.0003607837,0.000274998,0.002130377],"category_scores_gemma":[0.2283967,0.0004565077,0.004625535,0.001374277,0.001507645,0.00005046993,0.0005446617,0.001042954,0.00006409405],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002271487,"about_ca_system_score_gemma":0.001218122,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001166055,"about_ca_topic_score_gemma":0.000007496073,"domain_scores_codex":[0.9918069,0.001581803,0.002930617,0.001167923,0.001929525,0.0005831971],"domain_scores_gemma":[0.9735381,0.02068001,0.0004262036,0.002015483,0.001820046,0.001520152],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.00001143357,0.00009187194,0.000004487819,0.8919568,0.1037702,0.001206841,0.00004119031,3.107174e-8,2.513717e-8,0.0005537621,0.0005949275,0.001768424],"study_design_scores_gemma":[0.000237128,0.0003714779,0.000004212136,0.2284967,0.7142612,0.00001665073,0.00009864169,0.000001012138,3.298707e-8,0.00002725263,0.05625442,0.0002312979],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[1.702675e-7,0.9857618,0.00004169402,0.009499367,0.00007081233,0.003891943,0.0005527296,0.0000408328,0.0001406006],"genre_scores_gemma":[0.0001785261,0.9902437,0.00004074348,0.006543085,0.000143717,0.0009198753,0.0004916005,0.00006600439,0.001372751],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6634601,"threshold_uncertainty_score":0.9997886,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W1561379059","doi":"10.1371/journal.pone.0132634","title":"PEDro or Cochrane to Assess the Quality of Clinical Trials? A Meta-Epidemiological Study","year":2015,"lang":"en","type":"review","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":173,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Fondation pour la Recherche Médicale; Physiotherapy Foundation of Canada; Alberta Innovates; Killam Trusts; National Institutes of Health; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; University of Alberta; Alberta Innovates - Health Solutions; Women and Children's Health Research Institute; Children's Health Research Institute; Bill and Melinda Gates Foundation; Canadian Institutes of Health Research; National Science Foundation","keywords":"Blinding; Medicine; Meta-analysis; Systematic review; Clinical trial; MEDLINE; Physical therapy; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9978832001452151,"gpt":0.8031065913405215,"spread":0.1947766088046936,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.9341789,0.001104338,0.1499178,0.0003863922,0.0001198605,0.0005167568,0.007663617,0.0005806173,0.01905786],"category_scores_gemma":[0.9231139,0.0002626773,0.02818557,0.002980128,0.0001806183,0.0001010436,0.0009175164,0.001075259,0.005094427],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000042313,"about_ca_system_score_gemma":0.0007532155,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002597151,"about_ca_topic_score_gemma":0.00009582097,"domain_scores_codex":[0.1056691,0.7451407,0.121547,0.003305061,0.02376516,0.0005729304],"domain_scores_gemma":[0.1135538,0.7561658,0.09788492,0.02273848,0.008476942,0.001180119],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001988807,0.01895623,0.001917213,0.01861689,0.1459693,0.00002723639,0.0003440042,0.000002567477,2.793002e-7,0.000930585,0.1064532,0.7065836],"study_design_scores_gemma":[0.0003326504,0.0007960787,0.0002471897,0.001921868,0.1447193,0.000003288578,0.0003832311,0.00003202651,6.239053e-8,0.0006348988,0.8505335,0.0003959252],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00049421,0.984337,0.0007223492,0.0004572561,0.0002733341,0.01202155,0.0002125917,0.00001025036,0.001471453],"genre_scores_gemma":[0.000144131,0.9763132,0.005798894,0.0003487881,0.0007326962,0.001820064,0.00002176389,0.00003707811,0.01478335],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7440803,"threshold_uncertainty_score":0.9999825,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"evaluation","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3116112122","doi":"10.1111/cdoe.12616","title":"Socioeconomic status and oral health‐related quality of life: A systematic review and meta‐analysis","year":2020,"lang":"en","type":"review","venue":"Community Dentistry And Oral Epidemiology","topic":"Dental Health and Care Utilization","field":"Dentistry","cited_by":162,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"Fundação de Amparo à Pesquisa do Estado do Rio Grande do Sul; Conselho Nacional de Desenvolvimento Científico e Tecnológico","keywords":"Socioeconomic status; Medicine; Observational study; Meta-analysis; Confidence interval; Scopus; Quality of life (healthcare); Critical appraisal; Demography; Study heterogeneity; Systematic review; MEDLINE; Gerontology; Environmental health; Population; Internal medicine; Pathology; Alternative medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.2491112332091381,"gpt":0.4792272198465582,"spread":0.2301159866374201,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.009377636,0.0006025328,0.01732453,0.0001829232,0.0004122664,0.00003356852,0.0004094018,0.000662695,0.0002265082],"category_scores_gemma":[0.004793245,0.0004996361,0.001204726,0.0004342712,0.0005378407,0.0001304274,0.0006120438,0.001474586,0.00004199493],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009689516,"about_ca_system_score_gemma":0.0002513393,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004865827,"about_ca_topic_score_gemma":0.003972693,"domain_scores_codex":[0.9641778,0.0302444,0.004405275,0.0005354722,0.0001159152,0.0005210894],"domain_scores_gemma":[0.9875475,0.00692046,0.003818915,0.0009153652,0.00009244619,0.0007053617],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.0000103875,0.00004695257,0.001245906,0.9631197,0.02864005,0.000008696706,0.00004691869,2.02197e-7,5.60908e-9,0.0006619045,0.000528502,0.005690703],"study_design_scores_gemma":[0.0009741846,0.0003577039,0.009674294,0.0314838,0.9283656,0.001122309,0.000702457,0.00006300378,2.856132e-8,0.002085832,0.02386362,0.001307112],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0009043856,0.9962272,0.00009358903,0.0001831767,0.0001413575,0.001583768,0.0007931364,0.00004971183,0.00002367954],"genre_scores_gemma":[0.002755582,0.9927312,0.000141397,0.002554914,0.00001016005,0.0001242753,0.001487153,0.00003267139,0.0001626752],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.931636,"threshold_uncertainty_score":0.9997455,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2890588440","doi":"10.1016/j.jclinepi.2018.08.019","title":"Lack of transparency in reporting narrative synthesis of quantitative data: a methodological assessment of systematic reviews","year":2018,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":162,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Medical Research Council; McMaster University","keywords":"Transparency (behavior); Systematic review; Narrative review; Medicine; Narrative; MEDLINE; Accounting; Computer science; Political science; Business; Intensive care medicine; Computer security","retraction":null,"screen_n_in":null,"score":{"opus":0.9948432497923828,"gpt":0.8785193252618085,"spread":0.1163239245305743,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4736593,0.0001549085,0.009008804,0.0002630843,0.00008387026,6.985662e-7,0.0008284361,0.0003353667,0.0002601877],"category_scores_gemma":[0.8964843,0.00009375927,0.000407168,0.000539617,0.0008895487,0.0002113446,0.0001484574,0.00113627,0.000005182551],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007504012,"about_ca_system_score_gemma":0.00189176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002003271,"about_ca_topic_score_gemma":0.0002231817,"domain_scores_codex":[0.7380742,0.1857912,0.0743195,0.0005258028,0.0006056252,0.0006835792],"domain_scores_gemma":[0.1182316,0.7336804,0.1441222,0.001218594,0.002357449,0.0003896804],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003443423,0.0003094539,0.9365078,0.0344054,0.000181029,0.000009182215,0.01352254,0.00002812253,0.0003332899,0.01007386,0.002972702,0.001312244],"study_design_scores_gemma":[0.001677448,0.00413722,0.9271244,0.03388321,0.0002781693,0.00002363091,0.01900024,0.002508201,0.00006804107,0.01030531,0.0008106056,0.0001834953],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8232793,0.0006339999,0.1525219,0.01785057,0.001695943,0.002818283,0.0001006825,0.000006738707,0.001092589],"genre_scores_gemma":[0.7302659,0.001204803,0.2659662,0.002317288,0.0001862506,0.00004048781,0.000001150894,0.000006997129,0.00001091917],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6198426,"threshold_uncertainty_score":0.6158239,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"systematic_review","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2026016649","doi":"10.1016/s2352-3026(15)00003-4","title":"Rituximab plus standard of care for treatment of primary immune thrombocytopenia: a systematic review and meta-analysis","year":2015,"lang":"en","type":"review","venue":"The Lancet Haematology","topic":"Platelet Disorders and Treatments","field":"Medicine","cited_by":138,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Canadian Blood Services; McMaster University","funders":"","keywords":"Medicine; Rituximab; Internal medicine; Splenectomy; Meta-analysis; Immune thrombocytopenia; Relative risk; Randomized controlled trial; Confidence interval; Gastroenterology; Platelet; Lymphoma","retraction":null,"screen_n_in":null,"score":{"opus":0.1531214375545641,"gpt":0.4016144854047747,"spread":0.2484930478502107,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.0004472278,0.0004753962,0.06881621,0.0001469546,0.00004113507,0.000005164859,0.0002196839,0.0002007491,0.00003669221],"category_scores_gemma":[0.00007515257,0.0002087818,0.002626941,0.0003253439,0.0001611368,0.00001995667,0.000061211,0.00007580683,0.000005004505],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002073611,"about_ca_system_score_gemma":0.0004023529,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000423223,"about_ca_topic_score_gemma":0.00004665502,"domain_scores_codex":[0.9974514,0.0003938148,0.001333805,0.0003402701,0.0002207651,0.0002599224],"domain_scores_gemma":[0.9968106,0.0003976911,0.001303953,0.001160197,0.0002684369,0.00005913483],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.00008751555,0.0001129439,5.867447e-7,0.8225245,0.1755735,0.0001176688,0.0001790985,6.771768e-8,1.842292e-8,0.0001707335,0.00009871233,0.001134634],"study_design_scores_gemma":[0.002600069,0.0008441381,2.768843e-7,0.01898276,0.951282,0.001249359,0.0000525433,4.66579e-7,0.00000143838,0.0000594873,0.02479347,0.0001340322],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001479725,0.9896358,0.00001869634,0.0003517141,0.00001940376,0.008310744,0.001371601,0.00001440313,0.0002628662],"genre_scores_gemma":[0.0001117655,0.9970886,0.0004254553,0.00007830674,0.000009270857,0.001454008,0.0006440685,0.00003402597,0.0001545258],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8035418,"threshold_uncertainty_score":0.9450082,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2528175253","doi":"10.7717/peerj.2429","title":"Prevention of necrotizing enterocolitis with probiotics: a systematic review and meta-analysis","year":2016,"lang":"en","type":"review","venue":"PeerJ","topic":"Infant Nutrition and Health","field":"Nursing","cited_by":134,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; London Health Sciences Centre","funders":"London Health Sciences Centre","keywords":"Necrotizing enterocolitis; Medicine; Meta-analysis; Randomized controlled trial; Internal medicine; Enterocolitis; Incidence (geometry); Clinical trial; Probiotic; Relative risk; Cochrane Library; Sepsis; Gestational age; MEDLINE; Systematic review; Pediatrics; Intensive care medicine; Confidence interval; Pregnancy","retraction":null,"screen_n_in":null,"score":{"opus":0.09907042979141645,"gpt":0.3923774784407081,"spread":0.2933070486492917,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008400878,0.000334878,0.00870449,0.0003406281,0.00006486702,0.00003329901,0.0001236588,0.0001319437,0.0002534549],"category_scores_gemma":[0.00007827172,0.0001797427,0.00191018,0.0004488688,0.00004351397,0.00007970217,0.00002916596,0.0001554518,0.00002575837],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008230333,"about_ca_system_score_gemma":0.00004917669,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001457714,"about_ca_topic_score_gemma":0.00002977311,"domain_scores_codex":[0.9972028,0.0006560177,0.001285351,0.0003580788,0.0002699203,0.0002278925],"domain_scores_gemma":[0.9975829,0.0001582225,0.001481723,0.0005048648,0.0001871135,0.00008516331],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.000003411945,0.00003062544,2.434097e-7,0.9649931,0.03313961,0.000001745466,0.00002725493,7.332232e-9,2.509077e-8,0.0001108705,0.000171704,0.001521438],"study_design_scores_gemma":[0.00008877714,0.0001044509,2.595236e-7,0.2740932,0.6809193,0.00006843026,0.000004481376,3.619107e-7,4.195168e-7,0.0000195818,0.04455177,0.000148939],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[7.559348e-8,0.9928886,0.0001707027,0.0004341574,0.00002776017,0.00608649,0.000180569,0.0000321346,0.0001795355],"genre_scores_gemma":[0.000005072873,0.9973624,0.0005982268,0.0002396856,0.00002150006,0.0006353726,0.00007902697,0.00003484138,0.001023825],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6908998,"threshold_uncertainty_score":0.7329692,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2081069584","doi":"10.1097/01.sla.0000154258.30305.df","title":"Evaluating Meta-analyses in the General Surgical Literature","year":2005,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":131,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; University of Calgary","funders":"","keywords":"Medicine; Meta-analysis; MEDLINE; Concordance; Critical appraisal; Systematic review; Evidence-based medicine; Family medicine; Alternative medicine; Internal medicine; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9842320020590651,"gpt":0.6970802859876226,"spread":0.2871517160714425,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2731131,0.0002597481,0.004710898,0.0006318925,0.00007430485,0.000606087,0.00126476,0.00008586672,0.009595282],"category_scores_gemma":[0.03606727,0.00008852242,0.006882219,0.002865985,0.00005551543,0.0003757047,0.00006852684,0.0001925113,0.0004994694],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00000392645,"about_ca_system_score_gemma":0.00005679932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001708199,"about_ca_topic_score_gemma":0.00001659714,"domain_scores_codex":[0.9572694,0.02380348,0.0105428,0.0007513728,0.007278417,0.0003544836],"domain_scores_gemma":[0.9675193,0.02418369,0.003779335,0.002940191,0.001489401,0.00008811528],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003859975,0.0004027984,0.02967573,0.00008537232,0.005848264,0.0001345957,0.002938794,0.006120134,0.0003831092,0.004073814,0.8113729,0.1389259],"study_design_scores_gemma":[0.0002695895,0.00006099486,0.05701112,0.0001320574,0.005283858,0.0001597964,0.00108577,0.02720219,0.00138117,0.01486777,0.8917685,0.0007772485],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9401754,0.03870529,0.00007147048,0.01403335,0.0001072789,0.000283581,0.00002297956,0.000003053748,0.006597555],"genre_scores_gemma":[0.9916838,0.0002390217,0.001405058,0.002077,0.0002809549,0.00003586739,0.00001115346,0.000007374644,0.004259737],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2370458,"threshold_uncertainty_score":0.9913101,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2769378808","doi":"10.1016/j.jclinepi.2017.10.014","title":"Same family, different species: methodological conduct and quality varies according to purpose for five types of knowledge synthesis","year":2017,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":126,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Canada Research Chairs; Ottawa Hospital; University of Toronto; University of Alberta; St. Michael's Hospital","funders":"National Health and Medical Research Council; Medical Research Council; Canadian Institutes of Health Research; University of Ottawa","keywords":"Systematic review; Medicine; Protocol (science); Citation; Missing data; MEDLINE; Medical physics; Statistics; Alternative medicine; Computer science; Mathematics; Pathology; Library science","retraction":null,"screen_n_in":null,"score":{"opus":0.9944644201276043,"gpt":0.7932841591772162,"spread":0.2011802609503881,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.7123786,0.0007356831,0.09167862,0.0006064063,0.0001795932,0.0001762633,0.004084386,0.001111367,0.0006405377],"category_scores_gemma":[0.9657862,0.0002763334,0.01804075,0.0003215564,0.0007635977,0.0001424662,0.0007125076,0.001144075,0.0001131641],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004838176,"about_ca_system_score_gemma":0.0004640268,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008558354,"about_ca_topic_score_gemma":0.00001785692,"domain_scores_codex":[0.5642409,0.3439709,0.08740374,0.001889666,0.001838421,0.0006563144],"domain_scores_gemma":[0.04566747,0.8528041,0.09474733,0.003322334,0.002762126,0.0006966966],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001026237,0.0001158902,0.00287877,0.004096586,0.001859094,0.000003180125,0.0000473515,0.000002119314,5.162671e-7,0.007376181,0.01489479,0.9686229],"study_design_scores_gemma":[0.0001784747,0.0005622373,0.006920699,0.003376279,0.00383424,0.00003683878,0.0002166232,0.000040805,2.62536e-7,0.02207723,0.9624972,0.0002590806],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0008879572,0.9360163,0.05775418,0.001110762,0.002202947,0.001234976,0.0001352347,0.000001994798,0.0006556509],"genre_scores_gemma":[0.0002693204,0.936711,0.05899245,0.000325115,0.001139401,0.00005603066,0.000002912277,0.00002095898,0.00248282],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9683638,"threshold_uncertainty_score":0.9999689,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3163788934","doi":"10.1186/s12874-021-01295-w","title":"Quality assessment tools used in systematic reviews of in vitro studies: A systematic review","year":2021,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":126,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Al-Azhar University; Zagazig University; McMaster University","keywords":"Systematic review; Medical physics; Protocol (science); Guideline; Scopus; Quality assessment; Medicine; Computer science; MEDLINE; Pathology; External quality assessment; Alternative medicine; Biology","retraction":null,"screen_n_in":null,"score":{"opus":0.9945041192763169,"gpt":0.8198052753117314,"spread":0.1746988439645855,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"category_scores_codex":[0.9613585,0.001428907,0.1668841,0.004059894,0.0001106042,0.000367389,0.01065668,0.001320562,0.006820119],"category_scores_gemma":[0.988098,0.0006134547,0.01392104,0.01525444,0.0007777135,0.000231529,0.002333131,0.003926601,0.00145252],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001177859,"about_ca_system_score_gemma":0.007898412,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008380972,"about_ca_topic_score_gemma":0.0008124691,"domain_scores_codex":[0.0218038,0.845766,0.09581819,0.003519665,0.03160351,0.001488823],"domain_scores_gemma":[0.02482185,0.9162979,0.035824,0.01642033,0.005632799,0.001003111],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000003671342,0.0001842113,0.00001551367,0.9400328,0.000779301,0.0001313315,0.0002087565,1.82326e-7,6.114476e-7,0.001526983,0.002251803,0.05486484],"study_design_scores_gemma":[0.0002536521,0.0000563287,0.000003369886,0.9377954,0.00243173,0.0001026809,0.001728421,0.00007081704,2.547394e-7,0.0008316234,0.05631294,0.0004127887],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000002628009,0.9513879,0.01517963,0.0002512967,0.0004593434,0.03220964,0.00003257297,0.000005706168,0.000471251],"genre_scores_gemma":[0.000001815443,0.947226,0.03569809,0.0002341912,0.00008908712,0.0152133,0.00005116969,0.000055516,0.001430815],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1529631,"threshold_uncertainty_score":0.9999759,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2888025021","doi":"10.1093/scan/nsy072","title":"Meta-analysis of ERP investigations of pain empathy underlines methodological issues in ERP research","year":2018,"lang":"en","type":"review","venue":"Social Cognitive and Affective Neuroscience","topic":"Neural and Behavioral Psychology Studies","field":"Neuroscience","cited_by":121,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research","keywords":"Empathy; Psychology; Flexibility (engineering); Cognitive psychology; Reliability (semiconductor); Cognition; Meta-analysis; Social psychology; Neuroscience","retraction":null,"screen_n_in":null,"score":{"opus":0.9030527721176843,"gpt":0.6320929554922415,"spread":0.2709598166254428,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.004363484,0.0004083711,0.00353485,0.001156654,0.0005071107,0.00003908695,0.0004697868,0.0002551637,0.00002117963],"category_scores_gemma":[0.013094,0.0002746512,0.00123055,0.006802971,0.007985782,0.0001696165,0.0004118291,0.000734369,0.000003059045],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002645806,"about_ca_system_score_gemma":0.0001070849,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008377873,"about_ca_topic_score_gemma":0.0001062429,"domain_scores_codex":[0.9829683,0.01383457,0.0007164394,0.001351031,0.0006592292,0.0004703951],"domain_scores_gemma":[0.98671,0.01204172,0.0005803596,0.0001808742,0.0004015197,0.00008549687],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.0004883475,0.0073039,0.00499258,0.01359799,0.02662113,0.0007472624,0.02695662,0.000002798192,0.06927546,0.008839609,0.001437406,0.8397369],"study_design_scores_gemma":[0.002431825,0.0162682,0.03957107,0.004798084,0.7593344,0.00009406896,0.01163558,0.0001125605,0.05116991,0.05080828,0.05761595,0.006160058],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.05476884,0.9400136,0.0001471691,0.0005953364,0.0003065459,0.002548386,0.001060677,0.00004967932,0.0005097297],"genre_scores_gemma":[0.1336536,0.8652106,0.00006191264,0.0004103021,0.00008570189,0.0003226151,0.000008609257,0.00002584182,0.0002208101],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8335769,"threshold_uncertainty_score":0.9999706,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W1966512523","doi":"10.1186/1472-6963-13-528","title":"A governance model for integrated primary/secondary care for the health-reforming first world – results of a systematic review","year":2013,"lang":"en","type":"review","venue":"BMC Health Services Research","topic":"Interprofessional Education and Collaboration","field":"Health Professions","cited_by":116,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Medicine; Health informatics; Integrated care; Health administration; Nursing research; Nursing; Health care; Grey literature; CINAHL; Corporate governance; Population health; Health services research; Health policy; Inclusion (mineral); Public health; MEDLINE; Business; Political science; Psychological intervention; Psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.1817677203764707,"gpt":0.5574195498135738,"spread":0.3756518294371031,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.01262837,0.0005386411,0.004081724,0.0004177638,0.003205302,0.00005601508,0.001627094,0.0004042177,0.00005890478],"category_scores_gemma":[0.001073216,0.0003068627,0.0005335677,0.002024434,0.00007933193,0.0003128232,0.000317123,0.002048654,0.0001014163],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.004125531,"about_ca_system_score_gemma":0.05970972,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003698027,"about_ca_topic_score_gemma":0.03770177,"domain_scores_codex":[0.9872846,0.003985578,0.005298996,0.0008119093,0.001172371,0.001446531],"domain_scores_gemma":[0.9773355,0.01061564,0.005030931,0.001584547,0.005063305,0.0003700765],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001872452,0.00005055057,5.759693e-7,0.9255725,0.00005479447,3.901028e-8,0.007313227,0.000002004551,1.338157e-8,0.0006639108,0.02296491,0.04319027],"study_design_scores_gemma":[0.000380469,0.0001107335,0.000002111778,0.5287135,0.00008033971,4.519552e-7,0.006041608,0.002199636,2.180859e-8,0.00002394333,0.4623188,0.0001282849],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[4.895675e-7,0.9227347,0.0006592696,0.003950446,0.0008146758,0.06771231,0.003749741,0.00004421624,0.0003341245],"genre_scores_gemma":[0.000005571857,0.9022596,0.002155571,0.005473968,0.0003061649,0.06574684,0.002704848,0.0001190291,0.02122837],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4393539,"threshold_uncertainty_score":0.9999384,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W1821441292","doi":"10.1186/s13012-015-0286-1","title":"The effectiveness of knowledge translation interventions for promoting evidence-informed decision-making among nurses in tertiary care: a systematic review and meta-analysis","year":2015,"lang":"en","type":"review","venue":"Implementation Science","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":103,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"St. Peter's Hospital; Hamilton Health Sciences; Toronto Metropolitan University; Hamilton General Hospital; Lakehead University; McMaster Children's Hospital; McMaster University","funders":"McMaster University; London Health Sciences Centre; Canadian Institutes of Health Research; Hamilton Health Sciences","keywords":"CINAHL; Psychological intervention; Knowledge translation; Medicine; Data extraction; MEDLINE; Meta-analysis; Nursing research; Systematic review; Health care; Health services research; Nursing; Health administration; Public health; Knowledge management; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.5148330095571396,"gpt":0.701256680182869,"spread":0.1864236706257294,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04311352,0.0002173914,0.002248049,0.001002996,0.001148898,0.00008820398,0.0006892913,0.0000785122,0.00003849755],"category_scores_gemma":[0.01503733,0.0001215306,0.0007016441,0.004247657,0.0003243516,0.0009474166,0.0001133466,0.0002515221,0.000005520234],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000808387,"about_ca_system_score_gemma":0.005315223,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001610344,"about_ca_topic_score_gemma":0.003564139,"domain_scores_codex":[0.9915485,0.003851152,0.002734925,0.0004894719,0.0007915596,0.0005843462],"domain_scores_gemma":[0.9650102,0.03135872,0.001663698,0.0004432595,0.001346561,0.0001775892],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000003779154,0.00001190791,0.0003398158,0.9255968,0.0008040965,6.070458e-8,0.004870404,5.235003e-7,9.879551e-8,0.00003851539,0.00004154517,0.06829242],"study_design_scores_gemma":[0.0003223562,0.0001532991,0.00274855,0.7928192,0.1857684,0.000001759456,0.01345091,0.0001944847,7.472447e-7,0.0002989391,0.003908349,0.0003330547],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00008306305,0.977909,0.001015013,0.00005892743,0.00026989,0.02054562,0.00007743399,0.00001027579,0.0000308015],"genre_scores_gemma":[0.008553582,0.9754289,0.0005452594,0.00002378714,0.00001814478,0.0153446,0.00004994764,0.00001227658,0.00002346509],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1849643,"threshold_uncertainty_score":0.9932594,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2018803201","doi":"10.1016/s0895-4356(99)00160-2","title":"Randomized trials versus observational studies in adolescent pregnancy prevention","year":2000,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Adolescent Sexual and Reproductive Health","field":"Health Professions","cited_by":94,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Observational study; Randomized controlled trial; Medicine; Psychological intervention; CINAHL; PsycINFO; Pregnancy; MEDLINE; Family medicine; Psychiatry; Surgery; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9380215926437274,"gpt":0.7360098352874899,"spread":0.2020117573562376,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1770054,0.000191645,0.007367685,0.0001289413,0.0001721783,9.952893e-7,0.0002116023,0.0004799194,0.0008701438],"category_scores_gemma":[0.4180903,0.0001115708,0.00108259,0.0001747701,0.0004860715,0.0001510685,0.00004963954,0.002277213,0.0001145771],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001482299,"about_ca_system_score_gemma":0.0007974164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001204573,"about_ca_topic_score_gemma":0.00002022483,"domain_scores_codex":[0.8795796,0.1023109,0.01685805,0.000401983,0.0002808241,0.0005686444],"domain_scores_gemma":[0.6621632,0.3215996,0.01422607,0.0004697287,0.001097191,0.0004442509],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.4769093,0.001945216,0.3827553,0.000774741,0.0008516642,0.00003201828,0.001006764,0.0004091212,0.000002331863,0.009457876,0.02192391,0.1039318],"study_design_scores_gemma":[0.3896682,0.001267016,0.5075793,0.005719026,0.0002546151,0.00001017788,0.0008609321,0.0001647545,6.55908e-7,0.07298058,0.02128543,0.0002093402],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9095352,0.02140174,0.001525789,0.04481123,0.01838167,0.002795219,0.000006712345,0.00002970165,0.00151268],"genre_scores_gemma":[0.7869185,0.1487844,0.01304187,0.01558077,0.02078501,0.0002140157,0.00002558165,0.00006459543,0.01458533],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2410849,"threshold_uncertainty_score":0.9893484,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2763139081","doi":"10.1186/s13012-017-0648-y","title":"Organizational participatory research: a systematic mixed studies review exposing its extra benefits and the key factors associated with them","year":2017,"lang":"en","type":"review","venue":"Implementation Science","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":94,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Canadian Pharmacists Association; Université de Montréal; Université de Sherbrooke; Douglas Mental Health University Institute; McGill University","funders":"Canadian Institutes of Health Research; McGill University","keywords":"Health services research; Health administration; Medicine; Health informatics; Citizen journalism; Participatory action research; Nursing research; Value (mathematics); Public health; Key (lock); Public relations; Knowledge management; Sociology; Nursing; Political science; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.9823586787688844,"gpt":0.7849291663275012,"spread":0.1974295124413832,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05217391,0.0004641393,0.00215683,0.0004779398,0.01001207,0.0002922344,0.001559575,0.0001198737,0.0003757374],"category_scores_gemma":[0.05128305,0.0002469239,0.0001015245,0.002960734,0.002385402,0.000991381,0.0007331822,0.0008147011,0.0001539589],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0013025,"about_ca_system_score_gemma":0.007202043,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005325453,"about_ca_topic_score_gemma":0.0004120421,"domain_scores_codex":[0.9831378,0.008313889,0.002872336,0.0009147985,0.003249524,0.0015116],"domain_scores_gemma":[0.9676277,0.02328549,0.004040933,0.000995833,0.003617914,0.0004321533],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00001491461,0.00005256139,0.001494788,0.7761645,0.0005624246,0.000007431705,0.09985963,4.715914e-7,0.000002359594,0.04085775,0.005769622,0.07521354],"study_design_scores_gemma":[0.003723311,0.0001647328,0.001493684,0.7553376,0.002059659,0.00002685357,0.05805571,0.00001592424,0.00001089118,0.0004323917,0.1775367,0.00114257],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0009596061,0.9834267,0.00001327383,0.002003642,0.0004765522,0.01256376,0.0002630456,0.0000679618,0.0002254241],"genre_scores_gemma":[0.003024305,0.9912536,0.00006850123,0.001594738,0.00009030815,0.003736133,0.00004850378,0.0000491004,0.0001348525],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.171767,"threshold_uncertainty_score":0.9999983,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3087727097","doi":"10.1186/s43058-020-00068-8","title":"Advancing the pragmatic measurement of sustainment: a narrative review of measures","year":2020,"lang":"en","type":"review","venue":"Implementation Science Communications","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":86,"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":"National Institute on Drug Abuse; National Institute of Mental Health; U.S. Department of Health and Human Services","keywords":"Narrative; History; Political science; Linguistics; Philosophy","retraction":null,"screen_n_in":null,"score":{"opus":0.6523923922329052,"gpt":0.7300703640853611,"spread":0.07767797185245584,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.02734807,0.0003781345,0.001714931,0.0005180779,0.002631937,0.0000256911,0.004865682,0.00008473107,0.00051249],"category_scores_gemma":[0.01332375,0.0002632376,0.0002992298,0.005465651,0.001652332,0.0005772028,0.001514186,0.0008365759,0.0000598466],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001794711,"about_ca_system_score_gemma":0.02025371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000304634,"about_ca_topic_score_gemma":0.0005698143,"domain_scores_codex":[0.9832963,0.006714205,0.005247871,0.0005405816,0.003308028,0.0008929524],"domain_scores_gemma":[0.9809501,0.005128141,0.006627364,0.003083899,0.003896017,0.0003144801],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003076713,0.00006184613,0.00004245415,0.1284549,0.0001198769,2.749521e-7,0.05115687,4.371158e-7,0.00004956443,0.01038356,0.02503673,0.7846904],"study_design_scores_gemma":[0.0002399493,0.00006249928,0.00002482003,0.04398291,0.0003315641,0.000002303327,0.03132344,0.000007838077,0.000007578526,0.0001213154,0.9237159,0.0001798374],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00000109464,0.9613265,0.00115994,0.02600402,0.0002419149,0.009090727,0.0002947012,0.00003961586,0.001841469],"genre_scores_gemma":[0.0002501173,0.9907613,0.002527168,0.003439692,0.0000420263,0.00285581,0.00006794074,0.00003023098,0.00002569422],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8986792,"threshold_uncertainty_score":0.999982,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"opus","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3110084249","doi":"10.1016/j.joca.2020.10.004","title":"Visual Analogue Scale has higher assay sensitivity than WOMAC pain in detecting between-group differences in treatment effects: a meta-epidemiological study","year":2020,"lang":"en","type":"review","venue":"Osteoarthritis and Cartilage","topic":"Osteoarthritis Treatment and Mechanisms","field":"Medicine","cited_by":85,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Impact; Dalhousie University; McMaster University; University of Toronto; St. Michael's Hospital","funders":"Arthritis Society","keywords":"WOMAC; Visual analogue scale; Medicine; Osteoarthritis; Meta-analysis; Randomized controlled trial; Placebo; Physical therapy; Internal medicine; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.08019099511676102,"gpt":0.321414104342709,"spread":0.2412231092259479,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.003201714,0.001304107,0.008256466,0.0003837108,0.0001810684,0.0001267445,0.0001003894,0.000657927,0.00004408307],"category_scores_gemma":[0.0004299374,0.0008770716,0.001139031,0.0006258423,0.0001662061,0.0001181884,0.000161213,0.0007736042,0.0000218449],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003119498,"about_ca_system_score_gemma":0.0001157848,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002059016,"about_ca_topic_score_gemma":0.007864297,"domain_scores_codex":[0.9889678,0.006916886,0.001321366,0.001506055,0.0004081302,0.0008797527],"domain_scores_gemma":[0.995715,0.003015121,0.0003726061,0.0004011873,0.00002983059,0.0004662345],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"nonrandomized_trial","study_design_scores_codex":[0.00007027166,0.000726229,0.02863054,0.00128948,0.0002192847,0.00653936,0.0002900121,2.04711e-8,0.0001253394,0.000005871539,0.00000298585,0.9621006],"study_design_scores_gemma":[0.07494086,0.3626202,0.04163206,0.08879831,0.08005293,0.001853316,0.004789604,0.00003422134,0.001202415,0.0003673017,0.3341056,0.009603172],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.3400274,0.6559978,0.000009322458,0.00005713804,0.0001258486,0.003526066,0.0000963381,0.00008752849,0.00007257099],"genre_scores_gemma":[0.6172987,0.3805401,0.00008721308,0.00005505942,0.0002719334,0.001380742,0.0002279221,0.0000735585,0.00006479561],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9524974,"threshold_uncertainty_score":0.999971,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2772564313","doi":"10.1016/j.jclinepi.2017.11.022","title":"Flaws in the application and interpretation of statistical analyses in systematic reviews of therapeutic interventions were common: a cross-sectional analysis","year":2017,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":85,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto; St. Michael's Hospital; Ottawa Hospital; University of Ottawa","funders":"National Health and Medical Research Council; Generalitat Valenciana; University of Bristol; National Institute for Health and Care Research; Canada Research Chairs; Cancer Research UK; Medical Research Council; University of Ottawa","keywords":"Meta-analysis; Funnel plot; Interquartile range; Medicine; Random effects model; Sample size determination; Statistics; Publication bias; Subgroup analysis; Confounding; Psychological intervention; Study heterogeneity; MEDLINE; Mathematics; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9443845471738417,"gpt":0.7567789989945132,"spread":0.1876055481793285,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5300859,0.000166384,0.01414527,0.0006695689,0.00006750179,0.0001115742,0.001813132,0.0001736524,0.0003442761],"category_scores_gemma":[0.5043805,0.00006769574,0.005122395,0.000648729,0.000498697,0.0002141318,0.00009111772,0.0004874904,0.00001601196],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001861304,"about_ca_system_score_gemma":0.00005199093,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006916752,"about_ca_topic_score_gemma":0.000654816,"domain_scores_codex":[0.7754369,0.1462049,0.07578049,0.0006241291,0.001734559,0.0002190429],"domain_scores_gemma":[0.6667369,0.2409101,0.08787781,0.003051428,0.00129523,0.0001285429],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000527737,0.0001561296,0.9849139,0.001045866,0.0006627754,0.000001524713,0.0001192677,0.0008620954,0.000004365626,0.005908117,0.00009256916,0.006180624],"study_design_scores_gemma":[0.0002425098,0.0001433815,0.8928452,0.0005356895,0.0009087538,0.00001477175,0.0001520589,0.07045032,2.521856e-7,0.03457281,0.00009083082,0.00004341524],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4675383,0.005976547,0.5241485,0.001008422,0.0002050745,0.0008474623,0.00001358677,4.60939e-7,0.0002616517],"genre_scores_gemma":[0.9953935,0.0007093733,0.003607383,0.0001827692,0.00004537512,0.00002669692,0.000003076019,0.000002905182,0.00002899544],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5278551,"threshold_uncertainty_score":0.9669423,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2287844929","doi":"10.1136/bmjopen-2015-008562","title":"What is the influence of randomisation sequence generation and allocation concealment on treatment effects of physical therapy trials? A meta-epidemiological study","year":2015,"lang":"en","type":"review","venue":"BMJ Open","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":81,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Alberta Health; University of Alberta","funders":"Canadian Institutes of Health Research; Alberta Innovates; University of Alberta; Physiotherapy Foundation of Canada","keywords":"Medicine; Epidemiology; Meta-analysis; Alternative medicine; Sequence (biology); Public health; MEDLINE; Physical therapy; Gerontology; Internal medicine; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9578471903528902,"gpt":0.6968038048601565,"spread":0.2610433854927338,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2597782,0.0006103638,0.0228565,0.0001473997,0.00008529533,0.0007136922,0.001684872,0.0001291728,0.0001144386],"category_scores_gemma":[0.04089486,0.0001620402,0.003546937,0.0005939548,0.0001101464,0.0004019874,0.0001864226,0.0001053055,0.00006227122],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009638568,"about_ca_system_score_gemma":0.0003140209,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008946395,"about_ca_topic_score_gemma":0.00001204875,"domain_scores_codex":[0.8752403,0.1083557,0.0116893,0.00113067,0.003440675,0.0001433316],"domain_scores_gemma":[0.9373533,0.0403589,0.01786323,0.00338965,0.0009448658,0.00009009578],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001262602,0.0004347092,0.00002290528,0.0007320002,0.00606437,0.000001217195,0.002651113,0.0002392935,0.00001898614,0.0004433747,0.0009003665,0.9883654],"study_design_scores_gemma":[0.01500105,0.01221674,0.0005663764,0.009909393,0.0917971,0.00002279456,0.002132072,0.005955318,0.0002177866,0.004592742,0.8561955,0.001393102],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.01744215,0.9030035,0.0001322808,0.0003005661,0.00014601,0.07889667,0.00002191834,0.000001786729,0.00005516427],"genre_scores_gemma":[0.01200136,0.9725461,0.0001832829,0.0001878162,0.0001232537,0.01447628,0.00002994931,0.00001249472,0.0004394458],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9869723,"threshold_uncertainty_score":0.9916502,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2111376665","doi":"10.1177/0306624x12447771","title":"A Meta-Analysis of Moral Reconation Therapy","year":2012,"lang":"en","type":"review","venue":"International Journal of Offender Therapy and Comparative Criminology","topic":"Psychology of Moral and Emotional Judgment","field":"Neuroscience","cited_by":76,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Psychology; Medicine; Psychotherapist","retraction":null,"screen_n_in":null,"score":{"opus":0.9301869343671447,"gpt":0.4773109657371682,"spread":0.4528759686299765,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004255987,0.000356944,0.002895234,0.001160796,0.00004985245,0.0000279123,0.0007305106,0.0002108196,0.0008785339],"category_scores_gemma":[0.00002050247,0.0002267919,0.001956008,0.0002308444,0.0003480701,0.0002387958,0.00005559848,0.0004270689,0.000009801402],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005117994,"about_ca_system_score_gemma":0.00008674535,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005479511,"about_ca_topic_score_gemma":0.000001321657,"domain_scores_codex":[0.9970163,0.000815378,0.001147601,0.000314353,0.0005065094,0.0001998567],"domain_scores_gemma":[0.9967557,0.0006872258,0.001925314,0.0001637413,0.0003837558,0.00008430172],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0005264998,0.0009601979,0.00002140243,0.00007955974,0.2451028,0.00003822367,0.001119522,0.00002350269,0.0002597783,0.01910421,0.0002833051,0.7324811],"study_design_scores_gemma":[0.001344087,0.0007326173,0.0005744123,0.0001024657,0.09277081,0.0008899009,0.00008974843,0.00001908826,0.000917912,0.003676353,0.8984349,0.0004477268],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.004499001,0.9928672,0.00012986,0.0005068771,0.001099209,0.0001964001,0.00008434778,0.000006672552,0.000610405],"genre_scores_gemma":[0.04249522,0.9563383,0.0001507079,0.0006325035,0.00018698,0.0000190431,0.0000209884,0.00001053667,0.0001457138],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8981516,"threshold_uncertainty_score":0.961933,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaepi_narrow","metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2885318744","doi":"10.1016/j.jval.2018.06.010","title":"Valuation of Health and Nonhealth Outcomes from Next-Generation Sequencing: Approaches, Challenges, and Solutions","year":2018,"lang":"en","type":"review","venue":"Value in Health","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":76,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary; Canadian Centre for Applied Research in Cancer Control; University of British Columbia","funders":"Canadian Cancer Society Research Institute; Canadian Centre for Applied Research in Cancer Control; Cancer Research Institute","keywords":"Valuation (finance); MEDLINE; Health care; Preference; Actuarial science; Computer science; Data science; Management science; Medicine; Economics; Biology; Accounting; Microeconomics","retraction":null,"screen_n_in":null,"score":{"opus":0.9440262427380309,"gpt":0.4852302784682474,"spread":0.4587959642697835,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0460607,0.0005223203,0.005310892,0.0009645778,0.0003584034,0.00006723393,0.0002635083,0.0004886098,0.00002403015],"category_scores_gemma":[0.002612828,0.0006066962,0.0001945215,0.000294065,0.0001615512,0.0004383488,0.0001185587,0.0004761441,0.00008332008],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00301948,"about_ca_system_score_gemma":0.003898478,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02650858,"about_ca_topic_score_gemma":0.001249632,"domain_scores_codex":[0.9827529,0.004263157,0.01057703,0.001387181,0.0002073924,0.0008123894],"domain_scores_gemma":[0.9878461,0.001906651,0.008886935,0.0008310794,0.00005796552,0.0004712584],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000006912223,0.0002769816,0.0009999434,0.1163957,0.0002891525,5.392517e-7,0.01984154,0.0001358651,2.368892e-8,0.2388109,0.002690635,0.6205518],"study_design_scores_gemma":[0.001906591,0.0009829312,0.002498009,0.02540615,0.00008040183,0.00002796094,0.003658836,0.05219728,3.990441e-8,0.03944486,0.8721809,0.001616036],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0003219891,0.9785725,0.0009599542,0.01558664,0.0006829968,0.002984064,0.0007020283,0.00003949021,0.0001502849],"genre_scores_gemma":[0.0005200302,0.9902185,0.005912815,0.001830478,0.0004868432,0.0002581892,0.0006584154,0.00008113693,0.00003364932],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8694903,"threshold_uncertainty_score":0.9996384,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2103936380","doi":"10.1146/annurev-publhealth-031811-124540","title":"Advertising of Prescription-Only Medicines to the Public: Does Evidence of Benefit Counterbalance Harm?","year":2012,"lang":"en","type":"review","venue":"Annual Review of Public Health","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":76,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Harm; Scrutiny; Medical prescription; Direct-to-consumer advertising; Causality (physics); Medicine; Public health; Public economics; Health care; Business; Family medicine; Psychology; Nursing; Economics; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.5564649027992554,"gpt":0.6081295504974032,"spread":0.05166464769814783,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0167235,0.0006518276,0.003964593,0.0002941393,0.0002582763,0.00001630797,0.00177647,0.0007778445,0.001293245],"category_scores_gemma":[0.003481545,0.0003760459,0.0006499304,0.001518642,0.0005552868,0.0006352362,0.0004046248,0.002518232,0.00005580261],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003103074,"about_ca_system_score_gemma":0.004319293,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001567505,"about_ca_topic_score_gemma":0.00008014984,"domain_scores_codex":[0.9884761,0.004598947,0.003962794,0.0006283673,0.0009240379,0.001409775],"domain_scores_gemma":[0.9902022,0.002537369,0.003077781,0.0009777085,0.001608833,0.001596118],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001239677,0.0002117642,0.0002317294,0.2990224,0.0002061703,0.000001549959,0.0004789441,1.181911e-7,0.000001043347,0.001802729,0.0177492,0.6802819],"study_design_scores_gemma":[0.0001485765,0.0002584293,0.00007225872,0.1265122,0.0006690765,0.00005408155,0.0001576168,0.000003235934,0.000004893191,0.0000174045,0.8718514,0.000250778],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00002370879,0.934309,0.00004893305,0.05842712,0.001960936,0.003226054,0.001622494,0.00003612078,0.0003456417],"genre_scores_gemma":[0.001483379,0.977273,0.00006572052,0.01957374,0.0008367622,0.0002337234,0.0001281826,0.00005184171,0.0003536431],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8541023,"threshold_uncertainty_score":0.9998692,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2332511819","doi":"10.1016/j.jclinepi.2016.03.025","title":"A third of systematic reviews changed or did not specify the primary outcome: a PROSPERO register study","year":2016,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":75,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; Canadian Agency for Drugs and Technologies in Health; Canada Research Chairs; University of Ottawa; University of Toronto; St. Michael's Hospital","funders":"National Institute for Health and Care Research; Canadian Institutes of Health Research; University of Ottawa","keywords":"Register (sociolinguistics); Outcome (game theory); Medicine; Family medicine; Linguistics; Mathematics; Philosophy","retraction":null,"screen_n_in":null,"score":{"opus":0.973059586795305,"gpt":0.7017350610403976,"spread":0.2713245257549074,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.8325769,0.0004366665,0.04107178,0.0002997698,0.00009139514,0.00007756902,0.004467438,0.0002855428,0.002340397],"category_scores_gemma":[0.908933,0.00009707595,0.01089545,0.0007690291,0.0004284462,0.0002365648,0.0003465381,0.0007659407,0.001217539],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004986512,"about_ca_system_score_gemma":0.0001547574,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004040135,"about_ca_topic_score_gemma":0.00004251374,"domain_scores_codex":[0.3617352,0.410121,0.2192847,0.001596392,0.00649312,0.0007696474],"domain_scores_gemma":[0.2482034,0.5566531,0.1829386,0.008427211,0.003244152,0.0005334485],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005755729,0.0007402094,0.80445,0.003688452,0.001341695,0.00004772398,0.000472847,0.000002772354,0.00002276059,0.00125819,0.1722466,0.01515317],"study_design_scores_gemma":[0.003753336,0.00439577,0.8143153,0.008494236,0.003692647,0.0005016407,0.000981987,0.0001576029,0.00000291238,0.0100538,0.1531809,0.000469964],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6607638,0.02269896,0.08550344,0.1899854,0.01188887,0.0246473,0.00004520236,0.00001544425,0.004451561],"genre_scores_gemma":[0.9257931,0.003987357,0.01307175,0.01792451,0.002385559,0.0002111706,4.345765e-7,0.00003949221,0.03658666],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2650293,"threshold_uncertainty_score":0.9995601,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W1985235854","doi":"10.1177/135245850000600509","title":"Multiple sclerosis, disease modifying treatments and depression: a critical methodological review","year":2000,"lang":"en","type":"review","venue":"Multiple Sclerosis Journal","topic":"Multiple Sclerosis Research Studies","field":"Medicine","cited_by":68,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Sunnybrook Health Science Centre","funders":"","keywords":"Depression (economics); Mood; Psychiatry; Disease; Psychology; Clinical psychology; Mood disorders; Multiple sclerosis; Major depressive disorder; Placebo; Medicine; Anxiety; Alternative medicine; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.5577539936427163,"gpt":0.4657759437016936,"spread":0.09197804994102271,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.003086173,0.001785494,0.006588359,0.000724408,0.001538322,0.000445955,0.0007810935,0.0007608926,0.001762324],"category_scores_gemma":[0.02601716,0.001222213,0.002678138,0.0008681442,0.0009201184,0.0004780455,0.000812281,0.003288505,0.0002245859],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006734593,"about_ca_system_score_gemma":0.0007180591,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003334294,"about_ca_topic_score_gemma":0.000005791297,"domain_scores_codex":[0.9876187,0.003535761,0.002769746,0.0019553,0.00215721,0.001963219],"domain_scores_gemma":[0.9867304,0.006939636,0.0008240127,0.001433756,0.0005849706,0.003487239],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003931083,0.0008134869,0.00207462,0.04081832,0.001091076,0.0003384638,0.00004843816,0.000001430295,0.00002803099,0.000001442325,0.007466489,0.9469251],"study_design_scores_gemma":[0.005331594,0.0003555125,0.01981232,0.4101314,0.008978278,0.001117923,0.00003914726,0.0002620415,0.000008255905,0.00001678609,0.5525624,0.001384424],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001492544,0.9941769,0.0002295937,0.0007041913,0.0005010896,0.003347132,0.0004532216,0.0001994036,0.0002392368],"genre_scores_gemma":[0.0007153794,0.9838033,0.01151226,0.0006930046,0.001467486,0.001135557,0.0001938043,0.0002723772,0.000206808],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9455407,"threshold_uncertainty_score":0.9997615,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2988645757","doi":"10.1186/s40814-019-0500-z","title":"The reporting of progression criteria in protocols of pilot trials designed to assess the feasibility of main trials is insufficient: a meta-epidemiological study","year":2019,"lang":"en","type":"article","venue":"Pilot and Feasibility Studies","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":65,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Population Health Research Institute; McMaster University; Hamilton Health Sciences; University Health Network; Toronto General Hospital; St. Joseph’s Healthcare Hamilton; Impact","funders":"","keywords":"Medicine; Odds ratio; Confidence interval; Clinical trial; Sample size determination; Randomized controlled trial; Meta-analysis; Intervention (counseling); Odds; MEDLINE; Family medicine; Internal medicine; Logistic regression; Psychiatry; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.9811647581245485,"gpt":0.7154181071887392,"spread":0.2657466509358093,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7953476,0.0004671345,0.01875847,0.0002124814,0.0002346377,0.0001731682,0.001399502,0.00006416024,0.0003083206],"category_scores_gemma":[0.7542502,0.0001324096,0.002293143,0.001565592,0.0004596126,0.0001294341,0.0007253569,0.0001859142,0.000007329747],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005056179,"about_ca_system_score_gemma":0.0001172453,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005594342,"about_ca_topic_score_gemma":0.0001511957,"domain_scores_codex":[0.5595531,0.3050646,0.1205141,0.004061662,0.009803853,0.001002604],"domain_scores_gemma":[0.6402913,0.2480947,0.09410888,0.01095538,0.006299954,0.0002497457],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002115525,0.00228944,0.9803224,0.0004946621,0.001980735,9.654694e-7,0.004705342,0.00001264417,0.004015759,0.0002597393,0.0009673829,0.002835428],"study_design_scores_gemma":[0.002109079,0.007646909,0.9513968,0.0002770677,0.001766958,0.000002027847,0.02072252,0.0001823671,0.001128101,0.01437247,0.0001676099,0.0002281099],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9175283,0.002525234,0.0003141157,0.0008402827,0.0001149341,0.07852174,0.0000401719,0.00000388568,0.0001113115],"genre_scores_gemma":[0.9920878,0.00003143152,0.001367283,0.00005319232,0.00002358615,0.006293298,4.999337e-7,0.000006963239,0.000135928],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08073819,"threshold_uncertainty_score":0.9958091,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W2081631580","doi":"10.1016/j.jmpt.2013.04.001","title":"Manipulative and Multimodal Therapy for Upper Extremity and Temporomandibular Disorders: A Systematic Review","year":2013,"lang":"en","type":"review","venue":"Journal of Manipulative and Physiological Therapeutics","topic":"Temporomandibular Joint Disorders","field":"Health Professions","cited_by":62,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Canadian Memorial Chiropractic College","funders":"","keywords":"Medicine; Chiropractic; Manual therapy; Physical therapy; Randomized controlled trial; MEDLINE; Orthopedic surgery; Physical medicine and rehabilitation; Alternative medicine; Surgery; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.4647909135108516,"gpt":0.4799292772683735,"spread":0.01513836375752187,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001588043,0.0008670094,0.005190421,0.0001797837,0.000544134,0.00004104407,0.0003042293,0.0007334409,0.00007644657],"category_scores_gemma":[0.0003175588,0.0004752851,0.0006630417,0.0001737407,0.0003036534,0.0002435859,0.0001985494,0.001381964,0.000007699328],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009229973,"about_ca_system_score_gemma":0.0001297915,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000134338,"about_ca_topic_score_gemma":0.000003461991,"domain_scores_codex":[0.9934662,0.002693267,0.002404506,0.0005656417,0.0003310291,0.0005393341],"domain_scores_gemma":[0.9929553,0.002850186,0.003265834,0.0003665408,0.0003237831,0.0002383791],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001762357,0.000344983,0.0005070722,0.9185761,0.002997726,0.000008942685,0.001530262,3.74522e-7,0.00001593901,0.001280868,0.000843073,0.07371843],"study_design_scores_gemma":[0.003817464,0.002630477,0.002726982,0.5739197,0.00689745,0.0001172719,0.001472603,0.0002134006,3.084699e-7,0.02421808,0.3823624,0.001623828],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0009865562,0.9894511,0.0003961855,0.0003187429,0.0001649955,0.008550769,0.00005406127,0.00002624683,0.00005139806],"genre_scores_gemma":[0.001158372,0.9958806,0.001104295,0.0009238336,0.0001989563,0.0004819075,0.00004437237,0.00008577803,0.0001218724],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.3815193,"threshold_uncertainty_score":0.9997699,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2028540474","doi":"10.1371/journal.pone.0003684","title":"Following 411 Cochrane Protocols to Completion: A Retrospective Cohort Study","year":2008,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":62,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; Institute of Population and Public Health; Children's Hospital of Eastern Ontario; University of Ottawa","funders":"Canadian Institutes of Health Research; University of Ottawa","keywords":"Cochrane Library; Medicine; Systematic review; MEDLINE; Protocol (science); Confidence interval; Cohort study; Hazard ratio; Internal medicine; Alternative medicine; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.7904036982622812,"gpt":0.5202770186509911,"spread":0.2701266796112901,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.07070532,0.0002638444,0.006709467,0.0002862359,0.0003448655,0.0003699085,0.001317556,0.0000461411,0.007965324],"category_scores_gemma":[0.03317454,0.0001450175,0.001000988,0.001911366,0.00003211441,0.0001747634,0.0001625433,0.0001535946,0.008924771],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009200547,"about_ca_system_score_gemma":0.00004494792,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002046622,"about_ca_topic_score_gemma":0.00004016539,"domain_scores_codex":[0.9708427,0.004354732,0.005365117,0.001229004,0.01791808,0.0002903355],"domain_scores_gemma":[0.9917055,0.001007061,0.001441121,0.003546787,0.002073678,0.0002258208],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000009725714,0.002021579,0.9854829,0.00001949408,0.0008847918,0.00003574531,0.001007303,0.00000367557,0.0003397694,0.00008142435,0.01007235,0.00004124028],"study_design_scores_gemma":[0.0004123346,0.0004937664,0.9952177,0.0001576004,0.0004467361,0.000005350201,0.0003405093,0.0003595254,0.0001180621,0.0003555614,0.001878212,0.0002146689],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9384473,0.00004355062,0.001464926,0.000415949,0.00004636367,0.04695478,0.000007919869,0.00001918139,0.01260001],"genre_scores_gemma":[0.9660405,0.000001353528,0.005945849,0.0002354055,0.0001182825,0.01538027,0.000002714668,0.00001515433,0.01226049],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03753077,"threshold_uncertainty_score":0.9929416,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3193216869","doi":"10.1161/jaha.120.020513","title":"Representation of Women in Randomized Trials in Cardiac Surgery: A Meta‐Analysis","year":2021,"lang":"en","type":"review","venue":"Journal of the American Heart Association","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":61,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Randomized controlled trial; Cochrane Library; Meta-analysis; Internal medicine; Clinical trial; Sample size determination; MEDLINE; Psychological intervention; Inclusion and exclusion criteria; Surgery; Alternative medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.1037101779545427,"gpt":0.4127714158278478,"spread":0.3090612378733051,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.07294405,0.0002169393,0.05914109,0.001153159,0.00001520553,0.00002211385,0.00009178826,0.0001362962,0.0000461776],"category_scores_gemma":[0.06330025,0.0001193823,0.04653271,0.004180687,0.0000533556,0.0000615554,0.00002465828,0.0004409421,0.00000109167],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001683802,"about_ca_system_score_gemma":0.0008402865,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005107905,"about_ca_topic_score_gemma":0.000009927875,"domain_scores_codex":[0.9612514,0.03368088,0.003677586,0.0001681952,0.0009414284,0.0002805167],"domain_scores_gemma":[0.9242707,0.06818768,0.006856356,0.000293602,0.0002883832,0.0001032108],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.005652695,0.0003248929,0.2237921,0.001586726,0.7086925,0.00009648912,0.000928974,0.00005992368,0.000002817112,0.0000260879,0.001178222,0.05765856],"study_design_scores_gemma":[0.008672792,0.00005080054,0.02138448,0.00166073,0.7252405,0.00004213269,0.0003853615,0.000004356738,0.000005299865,0.00001836911,0.2423101,0.0002249956],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.007509391,0.9900238,0.000004184131,0.001351854,0.00020988,0.0008130564,0.00001260115,0.000002846977,0.00007242078],"genre_scores_gemma":[0.002251868,0.9968945,0.00008622003,0.0001420606,0.0001653317,0.00006741659,0.00001211983,0.00001787884,0.0003626264],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.2411319,"threshold_uncertainty_score":0.963016,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2059874779","doi":"10.1371/journal.pone.0020185","title":"Epidemiology, Quality and Reporting Characteristics of Systematic Reviews of Traditional Chinese Medicine Interventions Published in Chinese Journals","year":2011,"lang":"en","type":"review","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":58,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"University of Alberta; Lanzhou University","keywords":"Systematic review; Medicine; Psychological intervention; Epidemiology; MEDLINE; Alternative medicine; Specialty; Meta-analysis; Data extraction; Family medicine; Publication bias; China; Quality (philosophy); Pathology; Psychiatry","retraction":null,"screen_n_in":null,"score":{"opus":0.9767140384802047,"gpt":0.6506685977850202,"spread":0.3260454406951845,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5611646,0.0007313179,0.06898637,0.001701637,0.00005841389,0.00007341123,0.00193064,0.0003291661,0.009331256],"category_scores_gemma":[0.8737319,0.0002836093,0.006891168,0.00260309,0.0002712169,0.0003279322,0.0002190402,0.0006599831,0.00007341059],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005214258,"about_ca_system_score_gemma":0.0001175548,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003279348,"about_ca_topic_score_gemma":0.00003403057,"domain_scores_codex":[0.6565627,0.1070304,0.228545,0.0019561,0.005404152,0.0005016261],"domain_scores_gemma":[0.4298179,0.09357694,0.462624,0.009010531,0.004447906,0.0005228071],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000002628632,0.000912645,0.04875186,0.9295033,0.001838176,0.000004291636,0.0002566543,3.074829e-8,0.000001797455,0.0008226486,0.0004897143,0.01741631],"study_design_scores_gemma":[0.0002360089,0.0001307731,0.0806347,0.8883978,0.01023995,0.0001056689,0.00006191574,0.00007244158,5.93374e-8,0.01641998,0.003245089,0.0004556089],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001874358,0.9923156,0.0004669695,0.0001469081,0.0001724586,0.004075701,0.0001373444,0.000003198992,0.0008074709],"genre_scores_gemma":[0.001403369,0.9943838,0.002813042,0.00003665581,0.0002116099,0.0004254863,0.0001177551,0.00002387138,0.0005844333],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.3125673,"threshold_uncertainty_score":0.9999616,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2125820775","doi":"10.1002/j.0022-0337.2004.68.9.tb03849.x","title":"Implementing Evidence‐Based Practice in Undergraduate Teaching Clinics: A Systematic Review and Recommendations","year":2004,"lang":"en","type":"review","venue":"Journal of Dental Education","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":55,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Evidence-based dentistry; Medical education; Curriculum; Systematic review; Evidence-based practice; Psychological intervention; Evidence-based medicine; MEDLINE; Quality (philosophy); Psychology; Health care; Medicine; Alternative medicine; Nursing; Pedagogy","retraction":null,"screen_n_in":null,"score":{"opus":0.2586363994870742,"gpt":0.6457552192344075,"spread":0.3871188197473333,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02100097,0.0002207106,0.001519392,0.0007684047,0.0006256977,0.0000671592,0.0003664017,0.0001655372,0.0001029121],"category_scores_gemma":[0.02993057,0.0001648554,0.0001976351,0.0006842649,0.00003236859,0.001306531,0.00008269364,0.002224437,0.00009824663],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002184842,"about_ca_system_score_gemma":0.02133907,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000350841,"about_ca_topic_score_gemma":0.0001154169,"domain_scores_codex":[0.9895358,0.005481896,0.003631878,0.0002520231,0.0005442132,0.0005541756],"domain_scores_gemma":[0.9868894,0.006664294,0.005322963,0.0002686436,0.0005122795,0.0003424225],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000003119212,0.0002025939,0.00006138883,0.9088467,0.00003228274,0.000001618203,0.0006869414,1.342619e-7,3.430456e-8,0.0001429357,0.005242442,0.08477981],"study_design_scores_gemma":[0.0001177251,0.00007625294,0.000009883246,0.7336255,0.0009068667,0.00008952656,0.006285298,0.000002380886,7.134076e-9,0.00008707478,0.2586994,0.0001000717],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00000830787,0.9835666,0.00005669593,0.01081197,0.001796506,0.00345871,0.000005830049,0.000006376008,0.0002889537],"genre_scores_gemma":[0.00005425723,0.9937181,0.00274762,0.002280057,0.0003813384,0.0003871871,0.00003229975,0.00002220202,0.0003769376],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.2534569,"threshold_uncertainty_score":0.984209,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2165271154","doi":"10.1093/pubmed/fdu050","title":"The benefits and challenges of conducting an overview of systematic reviews in public health: a focus on physical activity","year":2014,"lang":"en","type":"article","venue":"Journal of Public Health","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Systematic review; Public health; MEDLINE; Alternative medicine; Medicine; Scientific literature; Cochrane collaboration; Management science; Psychology; Engineering ethics; Political science; Nursing; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.9583424756036388,"gpt":0.5984104860565596,"spread":0.3599319895470793,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4623645,0.0002447742,0.007540371,0.0005945234,0.0001613081,0.0003217645,0.00129937,0.00005769725,0.00003306778],"category_scores_gemma":[0.09675643,0.00009700352,0.0009262749,0.00114215,0.00008130257,0.000657846,0.0000840046,0.0003710371,0.00001106354],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001187963,"about_ca_system_score_gemma":0.0006282942,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005174861,"about_ca_topic_score_gemma":0.0003008042,"domain_scores_codex":[0.8804961,0.09447116,0.01786717,0.0006474854,0.005822699,0.0006954208],"domain_scores_gemma":[0.9514273,0.01320735,0.03118193,0.002113076,0.001409494,0.0006608768],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004217012,0.0004782214,0.0006567952,0.01716316,0.0001003203,4.403184e-7,0.004046605,0.000005669845,0.000007566894,0.06426157,0.0005200322,0.9127554],"study_design_scores_gemma":[0.008639709,0.02679637,0.2713501,0.1448995,0.0008029549,0.001363383,0.07190019,0.04878376,0.0000871388,0.06993276,0.3528382,0.002605909],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6281241,0.2443006,0.000821167,0.1240275,0.0002959909,0.002055907,0.000006506275,0.000002363668,0.0003659051],"genre_scores_gemma":[0.9645457,0.03463785,0.0003675172,0.0002909546,0.0001152605,0.00001531357,1.696156e-7,0.000009321876,0.00001789684],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9101495,"threshold_uncertainty_score":0.910852,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2766184587","doi":"10.1186/s40510-017-0196-7","title":"Effectiveness of en masse versus two-step retraction: a systematic review and meta-analysis","year":2017,"lang":"en","type":"review","venue":"Progress in Orthodontics","topic":"Orthodontics and Dentofacial Orthopedics","field":"Dentistry","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Medicine; Meta-analysis; Dentistry; Randomized controlled trial; Orthodontics; Surgery; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.1865449401167475,"gpt":0.4686407095626851,"spread":0.2820957694459376,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.00898363,0.001013629,0.01703977,0.0007510525,0.0001414306,0.000298963,0.0009248139,0.0008251872,0.00008103392],"category_scores_gemma":[0.002495539,0.0007675448,0.004772052,0.001183769,0.0002733378,0.0002515344,0.0005040009,0.001098623,0.00003192488],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001531028,"about_ca_system_score_gemma":0.0002083872,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004509601,"about_ca_topic_score_gemma":0.00028263,"domain_scores_codex":[0.9899106,0.004641711,0.002831376,0.001059718,0.00103722,0.0005193493],"domain_scores_gemma":[0.9900385,0.002984572,0.003902062,0.00245925,0.0004080636,0.0002075382],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.00002789033,0.0001173109,0.004886171,0.8754627,0.1115774,0.0003554928,0.000009958277,4.472391e-7,3.030312e-9,0.0005300533,0.000008788596,0.007023798],"study_design_scores_gemma":[0.0004719409,0.00006864161,0.0002436886,0.07067263,0.8995192,0.00008805483,0.000004795082,0.00001431119,1.764444e-7,0.000003485295,0.02825827,0.0006548142],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000008786069,0.9933023,0.0004452432,0.00001042968,0.0007690649,0.004751027,0.0003124353,0.00005313996,0.000347554],"genre_scores_gemma":[0.0001547628,0.995366,0.002562627,0.000007044553,0.00006480063,0.001022137,0.0001545738,0.0001199755,0.0005480474],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.80479,"threshold_uncertainty_score":0.9994776,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"evaluation","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity","scholarly_communication"],"domain":"evaluation","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3172845662","doi":"10.1186/s12916-021-02008-2","title":"Utilization of the evidence from studies with no events in meta-analyses of adverse events: an empirical investigation","year":2021,"lang":"en","type":"article","venue":"BMC Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Medical Research Council; National Health and Medical Research Council; Qatar National Library","keywords":"Systematic review; Medicine; Meta-analysis; Adverse effect; MEDLINE; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.951712016036363,"gpt":0.6398872374389903,"spread":0.3118247785973727,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03723853,0.0001921969,0.003425883,0.0001945337,0.00003790624,0.000004260979,0.0007599278,0.000043115,0.002704557],"category_scores_gemma":[0.1431453,0.0000610181,0.0006530058,0.002445743,0.000171923,0.0002729145,0.0001057921,0.00007761607,0.000034273],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002526644,"about_ca_system_score_gemma":0.0001896791,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001632446,"about_ca_topic_score_gemma":0.00139813,"domain_scores_codex":[0.9757277,0.01329116,0.005685157,0.0006231412,0.004574462,0.00009840299],"domain_scores_gemma":[0.9833187,0.006911695,0.004364193,0.002618571,0.002708128,0.000078731],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001749747,0.00004992611,0.9927999,0.00009534943,0.001282055,0.000001121695,0.002172557,0.0007645584,0.001464318,0.00002571529,0.001187529,0.0001394199],"study_design_scores_gemma":[0.0004963183,0.0001315343,0.9685357,0.0010076,0.00902125,0.000002390137,0.007297488,0.005882036,0.002920698,0.004396257,0.0002022134,0.0001065316],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9843494,0.007303016,0.006665473,0.001046759,0.0001398708,0.0003977926,0.000006446047,0.000001536853,0.00008973976],"genre_scores_gemma":[0.9937397,0.0001107135,0.004849961,0.0002520579,0.0000423824,0.00002182352,0.000009947655,0.000005014155,0.0009684507],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1059068,"threshold_uncertainty_score":0.9982071,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W1980348059","doi":"10.1177/000841740807500513","title":"Qualitative Meta-Synthesis: Reflections on the Utility and Challenges in Occupational Therapy","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Occupational Therapy","topic":"Occupational Therapy Practice and Research","field":"Health Professions","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"McMaster University; Toronto Rehabilitation Institute; University of Toronto","funders":"","keywords":"Qualitative research; Phenomenon; Interpretation (philosophy); Computer science; Management science; Engineering ethics; Knowledge management; Epistemology; Psychology; Sociology; Engineering; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.901061825158874,"gpt":0.6193784157795532,"spread":0.2816834093793208,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004650114,0.0002342142,0.0004398062,0.0006647949,0.00155196,0.00002270559,0.0003297954,0.0001892067,0.004308974],"category_scores_gemma":[0.001456887,0.0001427847,0.0002023774,0.000449015,0.0003151564,0.000393436,0.00001080296,0.001084379,0.00004798061],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000248412,"about_ca_system_score_gemma":0.004833943,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00226737,"about_ca_topic_score_gemma":0.009824029,"domain_scores_codex":[0.9942885,0.003482508,0.0008194561,0.0002574804,0.0006988176,0.0004532593],"domain_scores_gemma":[0.9893129,0.008625752,0.0005116457,0.000266804,0.0009245629,0.0003583211],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0321193,0.001894849,0.346025,0.0001419682,0.009567751,0.0004008897,0.3066263,0.0002584395,0.0006285435,0.1612338,0.06781656,0.07328654],"study_design_scores_gemma":[0.001650565,0.0004658282,0.8131975,0.00008186389,0.00003274695,0.00006121584,0.01336198,0.00002451609,0.000109237,0.007297666,0.1634707,0.0002461593],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9310579,0.02257048,0.0001092613,0.03838911,0.0005302974,0.0009495767,0.0002033161,0.00001446702,0.006175545],"genre_scores_gemma":[0.968277,0.02741594,0.000425163,0.002856347,0.0004240076,0.0003196782,0.000008738921,0.0000270907,0.000246007],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4671725,"threshold_uncertainty_score":0.9997479,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4224442502","doi":"10.1007/s10459-022-10103-z","title":"Exploring professional identity in rehabilitation professions: a scoping review","year":2022,"lang":"en","type":"review","venue":"Advances in Health Sciences Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":51,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; McGill University Health Centre; Centre for Interdisciplinary Research in Rehabilitation","funders":"Fonds de Recherche du Québec - Santé","keywords":"Identity (music); Rehabilitation; Medical education; Health professions; Medicine; Psychology; Physical therapy; Health care; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.2165631896154671,"gpt":0.5923900897870247,"spread":0.3758269001715576,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.008543826,0.0002846372,0.001339861,0.001648151,0.0004495371,0.0000203356,0.0004056372,0.00009991315,0.0003570271],"category_scores_gemma":[0.008185394,0.0002462531,0.0001190074,0.008474057,0.0002437511,0.002340696,0.0001191871,0.001169195,0.00001866801],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002607675,"about_ca_system_score_gemma":0.01817216,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001194785,"about_ca_topic_score_gemma":0.0001777476,"domain_scores_codex":[0.9941354,0.0009651221,0.00244636,0.0008369341,0.001116859,0.0004993339],"domain_scores_gemma":[0.9970818,0.0009026515,0.001325647,0.0004720775,0.0001301861,0.00008768046],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000001971237,0.000327998,0.0003187638,0.2861909,9.605654e-7,3.800827e-7,0.0004251803,0.00000283745,4.71894e-9,0.0009060245,0.0002654038,0.7115596],"study_design_scores_gemma":[0.00007454513,0.0001106639,0.0001767158,0.5540082,0.00001620953,0.00001523774,0.001486335,0.000007283882,1.134324e-8,0.0003828486,0.4435907,0.0001311781],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00007681346,0.9810153,0.00001969678,0.004270406,0.006580373,0.007616986,0.000001896776,0.00003682376,0.000381679],"genre_scores_gemma":[0.000006549911,0.9740974,0.008507319,0.002259152,0.000233522,0.01455166,0.0002321003,0.00002475853,0.00008753435],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7114284,"threshold_uncertainty_score":0.999999,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2802073505","doi":"10.1183/13993003.00376-2018","title":"The effect of anti-acid therapy on survival in idiopathic pulmonary fibrosis: a methodological review of observational studies","year":2018,"lang":"en","type":"review","venue":"European Respiratory Journal","topic":"Interstitial Lung Diseases and Idiopathic Pulmonary Fibrosis","field":"Medicine","cited_by":50,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; Jewish General Hospital","funders":"","keywords":"Medicine; Observational study; Pulmonary fibrosis; Idiopathic pulmonary fibrosis; Intensive care medicine; Internal medicine; Fibrosis; Lung","retraction":null,"screen_n_in":null,"score":{"opus":0.2886020444561501,"gpt":0.4418224539358594,"spread":0.1532204094797093,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.01655984,0.0005982317,0.003737854,0.0002527365,0.0002052364,0.00002358278,0.0006151007,0.0001580423,0.00004274313],"category_scores_gemma":[0.003589423,0.0002979955,0.001759242,0.0004939828,0.0005351189,0.00007560585,0.0001670747,0.001239795,0.00002977589],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001551596,"about_ca_system_score_gemma":0.0004540628,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":4.856948e-7,"about_ca_topic_score_gemma":1.225161e-7,"domain_scores_codex":[0.9822217,0.01361535,0.002453263,0.0004500392,0.0008870597,0.0003726119],"domain_scores_gemma":[0.9951721,0.001634465,0.001902992,0.0006512205,0.0004630026,0.0001761895],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.005455421,0.0003663014,0.0007200238,0.2255891,0.001417437,0.001847278,0.0000513023,7.565509e-7,0.0002616068,0.0001160183,0.003426924,0.7607478],"study_design_scores_gemma":[0.0005292386,0.01197743,0.001455397,0.4056505,0.0007482783,0.0003594836,0.00001255881,5.33522e-7,0.00002879275,0.000007635885,0.5789949,0.0002352533],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001163295,0.9961259,0.000004197668,0.00008129908,0.0007622684,0.0009602903,0.00005086165,0.0000155628,0.0008363171],"genre_scores_gemma":[0.0005326872,0.9975029,0.00004789231,0.0003539831,0.001398185,0.00003082334,0.00001400371,0.00008287852,0.00003663695],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7605125,"threshold_uncertainty_score":0.9999472,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2167473079","doi":"10.1093/humrep/des342","title":"Methodological quality of systematic reviews in subfertility: a comparison of Cochrane and non-Cochrane systematic reviews in assisted reproductive technologies","year":2012,"lang":"en","type":"article","venue":"Human Reproduction","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":50,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"University of Auckland","keywords":"Systematic review; Data extraction; Cochrane Library; Medicine; MEDLINE; Meta-analysis; Biology; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.9080448907257459,"gpt":0.6256176361119402,"spread":0.2824272546138057,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.529749,0.0005405755,0.02146034,0.001155773,0.0000942375,0.00009707025,0.001143423,0.0002742012,0.0001519573],"category_scores_gemma":[0.5073236,0.0002733316,0.001328468,0.003285545,0.0003571647,0.0004221121,0.0002423907,0.0004386265,0.00009277538],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001276221,"about_ca_system_score_gemma":0.00002269128,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009784201,"about_ca_topic_score_gemma":0.0001748104,"domain_scores_codex":[0.8330001,0.1117539,0.04701823,0.003119941,0.004481306,0.0006264911],"domain_scores_gemma":[0.9275293,0.01549056,0.03872442,0.0161095,0.001996997,0.0001492032],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.00008140031,0.001280119,0.1723811,0.789965,0.0002019026,0.000001462913,0.007529092,0.00002007861,0.02264925,0.001564568,0.0005579229,0.003768123],"study_design_scores_gemma":[0.001238497,0.0004649157,0.7729962,0.1761565,0.001947873,0.0001265594,0.02702739,0.0005953891,0.009247483,0.008652696,0.0003183368,0.001228148],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7702892,0.2138243,0.002323935,0.0002397529,0.0003388362,0.01255691,0.000003039584,0.00002256495,0.0004015049],"genre_scores_gemma":[0.9909869,0.001504577,0.006042673,0.000008289104,0.0000597392,0.0009544534,0.00000572763,0.0000140447,0.0004236368],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6138085,"threshold_uncertainty_score":0.9999719,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2146200566","doi":"10.1016/j.annemergmed.2008.05.036","title":"The Relative Efficacy of Meperidine for the Treatment of Acute Migraine: A Meta-analysis of Randomized Controlled Trials","year":2008,"lang":"en","type":"review","venue":"Annals of Emergency Medicine","topic":"Migraine and Headache Studies","field":"Medicine","cited_by":49,"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":"National Institute of Neurological Disorders and Stroke","keywords":"Medicine; Dihydroergotamine; Migraine; Randomized controlled trial; Odds ratio; Relative risk; Anesthesia; Adverse effect; Clinical trial; Antiemetic; Confidence interval; Meta-analysis; Ketorolac; Triptans; Nausea; Internal medicine; Analgesic","retraction":null,"screen_n_in":null,"score":{"opus":0.5200886805233103,"gpt":0.5377539528548487,"spread":0.01766527233153847,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaepi_broad"],"category_scores_codex":[0.01122458,0.0008365526,0.05874243,0.0007994483,0.0001380784,7.637189e-7,0.0003662771,0.0002335428,0.001129319],"category_scores_gemma":[0.03524367,0.0002596979,0.06485598,0.001696741,0.001027831,0.00002414049,0.00003449143,0.00017641,6.132933e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001617587,"about_ca_system_score_gemma":0.0002994754,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004408542,"about_ca_topic_score_gemma":0.00006451159,"domain_scores_codex":[0.985832,0.00217325,0.01020317,0.0004298284,0.001002242,0.00035949],"domain_scores_gemma":[0.9246871,0.06133031,0.01028062,0.001236397,0.002353151,0.0001123904],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.04515186,0.0006094893,0.000001067711,0.002430367,0.9015099,0.000001715044,0.002337713,0.000001540445,0.00001055233,0.000720111,0.04074556,0.006480114],"study_design_scores_gemma":[0.06258278,0.001889214,0.000003340699,0.0006723785,0.6301617,0.000001556888,0.0001307372,0.00001995285,0.00002936561,0.0000611423,0.3043439,0.0001039507],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00003038502,0.9800074,0.0001258166,0.005936956,0.0001566329,0.01240017,0.0005616825,0.000007527545,0.0007734288],"genre_scores_gemma":[0.0001673455,0.9862375,0.00006964398,0.00003139868,0.0003545099,0.002102603,0.0002263764,0.00004198113,0.01076862],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.2713482,"threshold_uncertainty_score":0.9999855,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2928825510","doi":"10.1097/corr.0000000000000690","title":"Is the Risk of Infection Lower with Sutures than with Staples for Skin Closure After Orthopaedic Surgery? A Meta-analysis of Randomized Trials","year":2019,"lang":"en","type":"review","venue":"Clinical Orthopaedics and Related Research","topic":"Surgical Sutures and Adhesives","field":"Medicine","cited_by":48,"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":"Medicine; Orthopedic surgery; Surgery; Randomized controlled trial; Meta-analysis; Sports medicine; Closure (psychology); MEDLINE; Physical therapy; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.3512710627236364,"gpt":0.5029919533418621,"spread":0.1517208906182257,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch","metaepi_broad","research_integrity"],"category_scores_codex":[0.05682863,0.0006424731,0.02095769,0.001158004,0.0001747622,0.00005568963,0.0002042885,0.00136425,0.0006938052],"category_scores_gemma":[0.0175786,0.0002186453,0.04781064,0.002954787,0.002382931,0.00005019161,0.0001390543,0.002689558,0.000003706357],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002091596,"about_ca_system_score_gemma":0.001020764,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001833401,"about_ca_topic_score_gemma":0.00009294641,"domain_scores_codex":[0.9804581,0.01156429,0.004681721,0.001014155,0.001710686,0.0005710666],"domain_scores_gemma":[0.8985732,0.09556212,0.002984956,0.001052015,0.00148625,0.0003414619],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.07510038,0.0006565718,0.01841469,0.01061512,0.8342543,0.00008074508,0.0001040968,0.000008875119,2.939243e-8,0.0001455042,0.000700456,0.05991919],"study_design_scores_gemma":[0.01901652,0.001485971,0.00249825,0.003076311,0.7834014,0.00003116117,0.0000356259,0.00005341282,5.400891e-7,0.00008150093,0.1900306,0.0002886974],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.01092695,0.9813284,0.00004125224,0.0005627912,0.0002347989,0.006262312,0.0005215906,0.00001659116,0.0001052873],"genre_scores_gemma":[0.01018163,0.9870537,0.00007609058,0.00004974808,0.000256763,0.0006406572,0.0002131552,0.00008754162,0.001440708],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1893301,"threshold_uncertainty_score":0.9999322,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4387695005","doi":"10.3346/jkms.2023.38.e333","title":"Causes for Retraction in the Biomedical Literature: A Systematic Review of Studies of Retraction Notices","year":2023,"lang":"en","type":"review","venue":"Journal of Korean Medical Science","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":47,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"H. Lundbeck A/S","keywords":"Misconduct; Subspecialty; Medicine; Meta-analysis; MEDLINE; Systematic review; Confidence interval; Scientific misconduct; Publication bias; Family medicine; Internal medicine; Alternative medicine; Pathology; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.1576615077346172,"gpt":0.4965828687945747,"spread":0.3389213610599575,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07524016,0.0001774101,0.002194985,0.0005384238,0.0002266647,0.00004269029,0.001732692,0.0009991542,0.00001794118],"category_scores_gemma":[0.1599115,0.00008625112,0.0005195613,0.003561051,0.002197417,0.0003990277,0.00005407096,0.002895503,0.000002412797],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002682695,"about_ca_system_score_gemma":0.003098489,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007951219,"about_ca_topic_score_gemma":0.00007764728,"domain_scores_codex":[0.9903781,0.001699074,0.002586064,0.0002070006,0.004814689,0.0003150437],"domain_scores_gemma":[0.9866428,0.008053438,0.003849352,0.0001965193,0.001052255,0.0002056703],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000006181649,0.00007762997,8.937152e-7,0.9633203,0.0001017964,0.00003844143,0.006356533,3.859787e-8,5.953146e-7,0.002823784,0.00545821,0.02181561],"study_design_scores_gemma":[0.00006278503,0.0001320234,0.000001936129,0.8793256,0.0008167957,0.0001055797,0.003858644,0.000005145892,8.278895e-7,0.0005004917,0.1151074,0.00008276542],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001025965,0.992979,0.00004498718,0.004193753,0.001667675,0.0009479235,0.00001687713,0.000005398183,0.0001340594],"genre_scores_gemma":[0.00008979032,0.9987404,0.0001368793,0.0002068002,0.0007451501,0.00002044621,0.000003917121,0.000006686558,0.00004996469],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1096492,"threshold_uncertainty_score":0.9994048,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_broad","research_integrity"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3004444768","doi":"10.1089/neu.2019.6844","title":"Interventions to Optimize Spinal Cord Perfusion in Patients with Acute Traumatic Spinal Cord Injuries: A Systematic Review","year":2020,"lang":"en","type":"review","venue":"Journal of Neurotrauma","topic":"Spinal Cord Injury Research","field":"Medicine","cited_by":44,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Population Health Research Institute; McMaster University; University of Calgary","funders":"","keywords":"Medicine; Spinal cord; Adverse effect; Anesthesia; Spinal cord injury; Randomized controlled trial; Perfusion; Meta-analysis; Surgery; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.1808361664395247,"gpt":0.4725873115280886,"spread":0.2917511450885639,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.001216693,0.0008733917,0.007559903,0.001455446,0.00009147299,0.0001176178,0.001048723,0.0002116443,0.000200427],"category_scores_gemma":[0.00172295,0.0005679808,0.002011143,0.002135973,0.00009998462,0.0002783232,0.000253183,0.002650506,0.0001714879],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006829452,"about_ca_system_score_gemma":0.000928058,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004574508,"about_ca_topic_score_gemma":0.000003473008,"domain_scores_codex":[0.9911241,0.0009602114,0.004560312,0.0006169647,0.002115265,0.0006231783],"domain_scores_gemma":[0.9943968,0.0001457538,0.002798771,0.0008357107,0.0009567569,0.0008661922],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.01502838,0.0005866962,0.00001624463,0.8140001,0.0003813514,0.001764557,0.00001409047,1.556371e-7,4.493838e-7,0.000003328906,0.0005956315,0.167609],"study_design_scores_gemma":[0.001840048,0.1295604,0.0006609507,0.8450918,0.009674549,0.002890193,0.00002757319,0.000001869869,3.150189e-7,0.0000018708,0.009866568,0.0003838406],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.004548846,0.9842112,0.0001217029,0.000739739,0.0003346912,0.009890714,0.00004310894,0.00002891303,0.00008106964],"genre_scores_gemma":[0.0006793863,0.9962925,0.001538914,0.0006746023,0.0001919726,0.0003073178,0.00002331182,0.0001685169,0.0001235191],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1672252,"threshold_uncertainty_score":0.9996772,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaepi_broad"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2096833901","doi":"10.1371/journal.pone.0088008","title":"Quantifying Bias in Randomized Controlled Trials in Child Health: A Meta-Epidemiological Study","year":2014,"lang":"en","type":"review","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Epidemiology; Randomized controlled trial; Meta-analysis; Medicine; MEDLINE; Environmental health; Biology; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9817470593674227,"gpt":0.6519590330899844,"spread":0.3297880262774383,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.9492888,0.001662827,0.4463966,0.003379816,0.0001339287,0.0008348197,0.003374443,0.0005250056,0.007294501],"category_scores_gemma":[0.9237,0.0005016659,0.0609404,0.003533772,0.000098009,0.0001198615,0.000298566,0.001320991,0.002405271],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001366804,"about_ca_system_score_gemma":0.0003052994,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001237182,"about_ca_topic_score_gemma":0.0005874893,"domain_scores_codex":[0.04541874,0.8011624,0.1395652,0.00308304,0.009936431,0.0008341788],"domain_scores_gemma":[0.08946429,0.7860366,0.1148095,0.008275912,0.0009261728,0.0004875707],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.01791234,0.03241106,0.001923733,0.05848724,0.4068525,0.0001533917,0.001592524,0.00005472443,1.170478e-7,0.004295456,0.007902191,0.4684147],"study_design_scores_gemma":[0.3408185,0.0005750929,0.00005245396,0.05057988,0.3075927,0.00001591912,0.0003522638,0.003479317,3.592372e-8,0.002379395,0.2924202,0.001734246],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001497156,0.9530569,0.0003556894,0.0008172194,0.0001473438,0.04447864,0.00002128772,0.0000115455,0.0009617034],"genre_scores_gemma":[0.0007163549,0.9849271,0.001863931,0.0003703247,0.0002098683,0.0107133,0.00002520355,0.00004051879,0.001133401],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4666805,"threshold_uncertainty_score":0.9997435,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2022418062","doi":"10.1186/1471-2288-14-127","title":"Methodological and ethical issues in research using social media: a metamethod of Human Papillomavirus vaccine studies","year":2014,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Social Media in Health Education","field":"Social Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Social media; Credibility; Research ethics; Transparency (behavior); Generalizability theory; CLARITY; Psychology; Internet privacy; Informed consent; Public relations; Medicine; Computer science; World Wide Web; Political science; Alternative medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.9767535309614988,"gpt":0.79382938185026,"spread":0.1829241491112388,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.5161341,0.0002277571,0.002084801,0.001357449,0.001565848,0.00003885429,0.001018174,0.002205285,0.0004886158],"category_scores_gemma":[0.9159092,0.000195572,0.0001300568,0.002940164,0.009052991,0.00009632319,0.000831946,0.005544376,0.00001063762],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005388667,"about_ca_system_score_gemma":0.002754661,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01913805,"about_ca_topic_score_gemma":0.03974092,"domain_scores_codex":[0.621241,0.3691493,0.001422427,0.0009548473,0.005272823,0.00195958],"domain_scores_gemma":[0.1296396,0.8657362,0.0003272894,0.0004981786,0.002636355,0.001162377],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004984317,0.0002816212,0.02466813,0.0008862897,0.00008374181,0.00003014453,0.198372,0.000001016954,0.00469054,0.5731964,0.001060655,0.196231],"study_design_scores_gemma":[0.002337631,0.0009198562,0.07340495,0.000588691,0.00005690474,0.00001158009,0.1527998,0.0002502733,0.001504204,0.7537033,0.01395341,0.0004693937],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.9436165,0.007402748,0.006665877,0.03815499,0.001871896,0.001572098,0.000002578682,0.00006593098,0.0006473701],"genre_scores_gemma":[0.4804631,0.009616658,0.5015453,0.0004737773,0.006709131,0.001073583,0.000005361878,0.00006191145,0.00005120682],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4965869,"threshold_uncertainty_score":0.999734,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"}]}