{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":29,"total_is_capped":false,"direct_labels_cover":18,"predictions_cover":29,"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":"8ffaf2cf1999","filters":{"venue":"Research Integrity and Peer Review"}},"results":[{"id":"W2552820166","doi":"10.1186/s41073-016-0024-5","title":"Ranking major and minor research misbehaviors: results from a survey among participants of four World Conferences on Research Integrity","year":2016,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":219,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Women's College Research Institute","keywords":"Minor (academic); Misconduct; Psychology; Quality (philosophy); Product (mathematics); Social psychology; Public relations; Political science; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.6723211565749861,"gpt":0.5540285380960718,"spread":0.1182926184789144,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.2420829,0.0003166862,0.0009932108,0.0008757914,0.001823557,0.0003834807,0.00146841,0.001081742,0.0009356458],"category_scores_gemma":[0.1464778,0.0002072087,0.0001387389,0.002220894,0.006884288,0.0005498516,0.0006547113,0.01367506,0.0001619349],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003213708,"about_ca_system_score_gemma":0.00180898,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.3128915,"about_ca_topic_score_gemma":0.1856039,"domain_scores_codex":[0.9537946,0.03361335,0.001260175,0.001293313,0.007798052,0.002240512],"domain_scores_gemma":[0.9459236,0.04295914,0.000241303,0.0008526239,0.009016134,0.001007254],"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":[0.002750484,0.000780058,0.3320645,0.00212436,0.0002465387,0.0001408125,0.02231093,3.409883e-8,0.0006158635,0.08584691,0.3289609,0.2241586],"study_design_scores_gemma":[0.002466422,0.000965386,0.3054943,0.05715803,0.0000895669,0.000005518131,0.00989569,0.00002760351,0.001556453,0.05778453,0.5636376,0.0009189649],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.861551,0.02081162,0.00001795834,0.08644508,0.0002821679,0.002670167,0.001594898,0.00005837515,0.02656874],"genre_scores_gemma":[0.9092004,0.07678433,0.0001227906,0.0001139955,0.0003443041,0.0001329459,0.00005268015,0.00002195618,0.01322662],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2346767,"threshold_uncertainty_score":0.9999776,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2592437185","doi":"10.1186/s41073-017-0027-x","title":"Recruitment of reviewers is becoming harder at some journals: a test of the influence of reviewer fatigue at six journals in ecology and evolution","year":2017,"lang":"en","type":"editorial","venue":"Research Integrity and Peer Review","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":111,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"3v Geomatics (Canada); Women's Health Research Institute; B.C. Women's Hospital & Health Centre","funders":"British Ecological Society; Ecological Society of America; Society for the Study of Evolution","keywords":"Demography; Medicine; Gerontology; Ecology; Psychology; Biology; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.8090483276171163,"gpt":0.6622627539633377,"spread":0.1467855736537786,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","research_integrity"],"consensus_categories":["metaresearch","bibliometrics"],"category_scores_codex":[0.2427814,0.000348452,0.00266476,0.01369966,0.0005505351,0.0005201535,0.004191347,0.0007656552,0.0005209294],"category_scores_gemma":[0.7379476,0.0001931855,0.0005250705,0.02569423,0.001628204,0.0006622484,0.004254799,0.004103616,0.00002311342],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007623481,"about_ca_system_score_gemma":0.001776566,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007214471,"about_ca_topic_score_gemma":0.000524207,"domain_scores_codex":[0.9617069,0.005320957,0.004153596,0.001156592,0.02669067,0.0009713225],"domain_scores_gemma":[0.8245476,0.06756917,0.007961532,0.004452072,0.09441885,0.001050716],"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":[0.00002461804,0.0001295267,0.06711761,0.007131722,0.00004445216,0.000004584794,0.0001075729,7.488179e-7,0.0001393645,0.0000377978,0.8967075,0.02855455],"study_design_scores_gemma":[0.0003790355,0.0002695878,0.07963856,0.03925018,0.00005257817,0.00001221406,0.00003116018,0.00001675768,0.00008337678,0.002011024,0.8780813,0.0001741856],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.02577245,0.9318089,0.000002106391,0.01103628,0.02758501,0.002798155,0.0006991639,0.000001918601,0.0002959875],"genre_scores_gemma":[0.00654475,0.9825783,0.00008259507,0.00008254914,0.00630599,0.00008736342,0.000007382978,0.00001665685,0.004294385],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4951662,"threshold_uncertainty_score":0.998194,"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"],"domain":"evaluation","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2889155419","doi":"10.1186/s41073-018-0050-6","title":"Beyond sex and gender difference in funding and reporting of health research","year":2018,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Sex and Gender in Healthcare","field":"Medicine","cited_by":98,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"Institute of Gender and Health; Secretaría de Estado de Investigación, Desarrollo e Innovación; Narodowe Centrum Badań i Rozwoju; Medical Research Council; National Institutes of Health; Latvijas Zinātnes Padome; Vlaamse regering; Norges Forskningsråd; Agentúra na Podporu Výskumu a Vývoja; Danmarks Frie Forskningsfond; Centre National de la Recherche Scientifique; Lietuvos Mokslo Taryba; Hrvatska Zaklada za Znanost; Vetenskapsrådet; Canadian Institutes of Health Research; Fonds Wetenschappelijk Onderzoek; Ministero dell’Istruzione, dell’Università e della Ricerca; Academy of Finland; Eesti Teadusagentuur; Rannís; Unitatea Executiva pentru Finantarea Invatamantului Superior, a Cercetarii, Dezvoltarii si Inovarii; Irish Research Council; Malta Council for Science and Technology; Icelandic Centre for Research; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; Javna Agencija za Raziskovalno Dejavnost RS; Hungarian Scientific Research Fund; Fonds De La Recherche Scientifique - FNRS; Research Councils UK; Grantová Agentura České Republiky; Israel Science Foundation; Research Promotion Foundation; Fonds National de la Recherche Luxembourg; Deutsche Forschungsgemeinschaft; National Science Foundation; Science Foundation Ireland; Fundação para a Ciência e a Tecnologia; ZonMw; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Türkiye Bilimsel ve Teknolojik Araştırma Kurumu; Agence Nationale de la Recherche","keywords":"Inclusion (mineral); Scholarship; Gender analysis; Publishing; Consistency (knowledge bases); Political science; Inclusion–exclusion principle; Public relations; Psychology; Sociology; Gender studies; Politics","retraction":null,"screen_n_in":null,"score":{"opus":0.839745605306376,"gpt":0.6353264997642469,"spread":0.2044191055421291,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06375729,0.000113042,0.0007420391,0.0003128755,0.0003027917,0.00003595084,0.00008876848,0.0001207574,0.0000626972],"category_scores_gemma":[0.01714023,0.00008364515,0.00002977636,0.0006598042,0.0006964489,0.00005799414,0.0002445354,0.00206242,0.000003011215],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000867024,"about_ca_system_score_gemma":0.0005411853,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00226506,"about_ca_topic_score_gemma":0.0003476905,"domain_scores_codex":[0.9945359,0.001708518,0.001156333,0.0005180872,0.001400259,0.0006808722],"domain_scores_gemma":[0.9964603,0.000915144,0.0002084289,0.0003835001,0.001646646,0.000386004],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00006602052,0.00009412259,0.3369202,0.05190286,0.00002941747,0.00004876747,0.006643769,2.133809e-9,0.0002144499,0.005000188,0.005637007,0.5934432],"study_design_scores_gemma":[0.0009281524,0.002079568,0.8698893,0.02704053,0.00002009586,0.0007152119,0.007311414,0.0001369185,0.0002912221,0.03162391,0.0597176,0.000246102],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5703246,0.3700323,0.00001179087,0.05245512,0.00004816234,0.001345429,0.0000102876,0.00001277493,0.005759608],"genre_scores_gemma":[0.8548428,0.142864,0.0003169177,0.0005024335,0.0001102901,0.0000285949,0.000009847171,0.00001062913,0.001314487],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5931971,"threshold_uncertainty_score":0.9911388,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4283789592","doi":"10.1186/s41073-022-00123-z","title":"Improving equity, diversity, and inclusion in academia","year":2022,"lang":"en","type":"letter","venue":"Research Integrity and Peer Review","topic":"Diversity and Career in Medicine","field":"Social Sciences","cited_by":98,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Bruyère; University of Ottawa","funders":"","keywords":"Excellence; Inclusion (mineral); Publishing; Publication; Equity (law); Public relations; Diversity (politics); Library science; Political science; Set (abstract data type); Engineering ethics; Computer science; Sociology; Engineering; Social science; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.3620357640097279,"gpt":0.4985177753239914,"spread":0.1364820113142635,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","open_science","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.05148603,0.0001758485,0.0005361291,0.0003245676,0.008655653,0.0001462947,0.001866262,0.001453226,0.001550744],"category_scores_gemma":[0.005824875,0.0001629174,0.00008126438,0.0007032714,0.001028662,0.0002526062,0.1349752,0.0272518,0.00001192437],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005429029,"about_ca_system_score_gemma":0.0004520242,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03927825,"about_ca_topic_score_gemma":0.00191704,"domain_scores_codex":[0.9874558,0.002813791,0.0002651658,0.0005359044,0.008176466,0.0007529107],"domain_scores_gemma":[0.9980367,0.0008949487,0.00008827759,0.0002268105,0.0005390298,0.0002142472],"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":[0.000006976873,0.00001218274,0.000650209,0.004465299,0.000009803784,0.0002869066,0.009184249,4.76639e-9,3.608709e-7,0.0009314275,0.9010337,0.08341888],"study_design_scores_gemma":[0.0001411609,0.00007921789,0.00006288401,0.003241434,0.00003870123,0.000005342418,0.001949686,0.000001734608,1.788186e-7,0.007475632,0.9868372,0.0001668021],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0003192937,0.09211731,0.000001180661,0.8791443,0.0001913434,0.0006881209,0.00005353005,0.00002139613,0.02746356],"genre_scores_gemma":[0.001462839,0.4533747,0.00001995682,0.5287266,0.001121735,0.00002670131,0.0001310897,0.00001233396,0.01512401],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.3612574,"threshold_uncertainty_score":0.9998431,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2613383989","doi":"10.1186/s41073-017-0031-1","title":"Retractions in cancer research: a systematic survey","year":2017,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":83,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Juravinski Hospital; Hamilton Health Sciences; McMaster University","funders":"","keywords":"Impact factor; MEDLINE; Medicine; Descriptive statistics; Family medicine; Library science; Computer science; Political science; Law; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.7171569621897869,"gpt":0.63058816574139,"spread":0.08656879644839688,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2059369,0.0001182045,0.0005420686,0.0002662322,0.003627497,0.0006126357,0.001249594,0.0006824849,0.0006708574],"category_scores_gemma":[0.1821889,0.00009390079,0.00007283771,0.0006902989,0.001585539,0.0005877044,0.0002651946,0.01205828,0.0002349904],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003865078,"about_ca_system_score_gemma":0.001134578,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.3674237,"about_ca_topic_score_gemma":0.2502567,"domain_scores_codex":[0.9799778,0.01458181,0.0004964932,0.0004221336,0.003575036,0.0009467166],"domain_scores_gemma":[0.9898578,0.005102193,0.0001353502,0.0006634606,0.003905108,0.0003360832],"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":[0.00007756393,0.000330847,0.0455764,0.1011841,0.0001023349,0.0001300756,0.01608712,1.434924e-7,0.0000240887,0.2321558,0.586793,0.01753849],"study_design_scores_gemma":[0.0002271179,0.00006871347,0.04408685,0.1381963,0.0000253786,0.00001184519,0.004483663,0.00003720691,0.00001172123,0.019518,0.7930055,0.0003277923],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.02544775,0.278621,0.00001998801,0.4800517,0.0008498012,0.005526694,0.0002424716,0.00006567856,0.2091749],"genre_scores_gemma":[0.4126217,0.5549334,0.00003882146,0.0003154541,0.0003169548,0.0003800094,0.00001276629,0.00001250483,0.03136841],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.4797363,"threshold_uncertainty_score":0.9976696,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2416590355","doi":"10.1186/s41073-016-0014-7","title":"Updating standards for reporting diagnostic accuracy: the development of STARD 2015","year":2016,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":82,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Public Health; Ottawa Hospital; University of Ottawa","funders":"Medical Research Council","keywords":"Diagnostic accuracy; Medical physics; Computer science; Medicine; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.921923900201108,"gpt":0.6904396986495264,"spread":0.2314842015515817,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7030503,0.0001771525,0.002350253,0.0001515219,0.0004113115,0.0003869653,0.001340684,0.00006061136,0.003105818],"category_scores_gemma":[0.8594182,0.00005345941,0.0006952247,0.0008808497,0.0001786964,0.0002088626,0.0003470797,0.0004063577,0.000162091],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007748703,"about_ca_system_score_gemma":0.0009292584,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002027093,"about_ca_topic_score_gemma":0.00008255867,"domain_scores_codex":[0.9380088,0.01692302,0.02155951,0.001174835,0.02164701,0.000686777],"domain_scores_gemma":[0.4127584,0.405696,0.04783198,0.01553233,0.1168609,0.00132034],"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":[0.000005248071,0.00001820079,0.002403483,0.00226387,0.00009250893,0.000001399177,0.0003170558,6.131273e-8,0.00005933386,0.002608916,0.3710184,0.6212115],"study_design_scores_gemma":[0.00008311812,0.00003336536,0.001335169,0.005665701,0.00005539453,0.000007189746,0.0004254535,0.00002320726,0.0001649517,0.007002165,0.9851204,0.0000838566],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.03479385,0.4025634,0.1584483,0.3686638,0.0005632074,0.01490931,0.000753727,0.00001753511,0.01928686],"genre_scores_gemma":[0.7998797,0.08169883,0.07034227,0.00129729,0.0005597874,0.002213559,0.00005788506,0.00007153595,0.04387917],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7650858,"threshold_uncertainty_score":0.9978055,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2736138949","doi":"10.1186/s41073-017-0039-6","title":"Reporting of sex and gender in randomized controlled trials in Canada: a cross-sectional methods study","year":2017,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Sex and Gender in Healthcare","field":"Medicine","cited_by":77,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Queen's University; Institute of Gender and Health; Canadian Agency for Drugs and Technologies in Health; Ottawa Hospital; University of British Columbia; Ottawa Public Health; Cochrane; B.C. Women's Hospital & Health Centre; University of Waterloo; University of British Columbia Hospital; Bruyère; University of Ottawa","funders":"Institute of Gender and Health; Canadian Institutes of Health Research","keywords":"Psychological intervention; Randomized controlled trial; Medicine; Data extraction; MEDLINE; Family medicine; Psychology; Political science; Nursing","retraction":null,"screen_n_in":null,"score":{"opus":0.7495869905793017,"gpt":0.6726107220083017,"spread":0.07697626857099993,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.384031,0.0001285727,0.00527195,0.0001840872,0.0001655567,0.00006268892,0.000118818,0.00009434937,0.000104796],"category_scores_gemma":[0.4430262,0.00007700134,0.0002244266,0.0001328051,0.0002015744,0.00006052039,0.0001146373,0.00137964,2.543924e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001817957,"about_ca_system_score_gemma":0.002632795,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.7315086,"about_ca_topic_score_gemma":0.4053375,"domain_scores_codex":[0.9705798,0.01982415,0.006599001,0.00061892,0.001881868,0.0004962764],"domain_scores_gemma":[0.9801442,0.01541005,0.001839131,0.000665034,0.001696763,0.0002448148],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01444988,0.0001074927,0.9397709,0.006168585,0.0002258398,0.0001260807,0.0008480554,1.10069e-7,0.00001312765,0.0001217788,0.0001945306,0.03797365],"study_design_scores_gemma":[0.1344504,0.0001238295,0.8577126,0.00262792,0.00009969009,0.00009029535,0.00131708,0.0001856676,0.00002260645,0.002003679,0.001268667,0.00009761847],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8241727,0.1594068,0.00001503606,0.006890452,0.0001291369,0.007574274,0.00001137946,0.000003844159,0.001796379],"genre_scores_gemma":[0.9684279,0.02992473,0.0004598932,0.0001545043,0.00007047928,0.0004132728,0.000005347282,0.000007260106,0.0005366348],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3261711,"threshold_uncertainty_score":0.6342697,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3045465353","doi":"10.1186/s41073-020-00096-x","title":"Quantifying professionalism in peer review","year":2020,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":69,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Environment and Climate Change Canada; Memorial University of Newfoundland; Library and Archives Canada; University of Victoria","funders":"","keywords":"Rubric; Criticism; Psychology; Peer review; Distress; Quality (philosophy); Medical education; Applied psychology; Medicine; Clinical psychology; Epistemology; Law; Mathematics education; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.9673817885818834,"gpt":0.7461647558699385,"spread":0.2212170327119449,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","scholarly_communication","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.250138,0.0002024955,0.0009250742,0.008662999,0.0003279313,0.001243421,0.002727813,0.0001696129,0.003137276],"category_scores_gemma":[0.5492743,0.0001223404,0.0002098221,0.1314059,0.0003035171,0.0006489505,0.001693203,0.003785944,0.001475851],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009824391,"about_ca_system_score_gemma":0.0006506505,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002902282,"about_ca_topic_score_gemma":0.00004969623,"domain_scores_codex":[0.9498593,0.003966381,0.00153105,0.001298473,0.04222193,0.001122816],"domain_scores_gemma":[0.9106461,0.01938592,0.0004689363,0.002071377,0.06508356,0.002344133],"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":[0.00001103454,0.0000715701,0.007678232,0.004393061,0.000005967569,0.00005068172,0.0001123092,6.537331e-8,0.00002645376,0.00346336,0.7612672,0.22292],"study_design_scores_gemma":[0.0001782897,0.00009213049,0.007159459,0.008996517,0.000004969657,0.00001190461,0.0001413097,0.0003186585,0.00001641887,0.002666689,0.9802539,0.0001597929],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.001488528,0.4935625,0.00005045821,0.497716,0.0001010269,0.0009503905,0.00003834334,0.00001349769,0.00607924],"genre_scores_gemma":[0.05893954,0.9100099,0.0004632967,0.01505616,0.0001351557,0.0001230978,0.00003902708,0.000019501,0.01521438],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.4826598,"threshold_uncertainty_score":0.9997934,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2676139175","doi":"10.1186/s41073-017-0038-7","title":"Improving the process of research ethics review","year":2017,"lang":"en","type":"editorial","venue":"Research Integrity and Peer Review","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":67,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Research ethics; Process (computing); Workload; Workflow; Set (abstract data type); Engineering ethics; Process management; Business; Management science; Knowledge management; Political science; Computer science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.8275826729916408,"gpt":0.742965650036921,"spread":0.08461702295471973,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.5042065,0.0004856568,0.002692649,0.0004066156,0.001720292,0.0003361157,0.004046653,0.004673383,0.0003616287],"category_scores_gemma":[0.9291375,0.0002801434,0.0005790418,0.001337897,0.007985244,0.0001459931,0.002698332,0.1663236,0.0001798792],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003440129,"about_ca_system_score_gemma":0.03212958,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003034524,"about_ca_topic_score_gemma":0.001033622,"domain_scores_codex":[0.9271649,0.01427011,0.002802964,0.002148805,0.05135246,0.002260749],"domain_scores_gemma":[0.1275634,0.5622894,0.002156147,0.01485291,0.2906373,0.002500933],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00006920005,0.00009735388,0.000009219068,0.3927084,0.00007877245,0.0000450379,0.0001044299,1.179365e-9,0.000003080968,0.001058724,0.583779,0.02204669],"study_design_scores_gemma":[0.0002178357,0.0005055036,0.000006632868,0.3646941,0.0001638731,0.00001504201,0.00006593389,0.000002218457,0.00001815662,0.007360802,0.6268029,0.0001470495],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000002727934,0.6917337,0.000003532253,0.2292703,0.06094135,0.004588868,0.0002029712,0.00002457411,0.01323197],"genre_scores_gemma":[0.0000312921,0.7616955,0.00006594628,0.0007331114,0.1977559,0.0004991588,0.0001981963,0.00007485547,0.03894599],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7996016,"threshold_uncertainty_score":0.9999651,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2895514839","doi":"10.1186/s41073-018-0053-3","title":"Protocol for the development of a CONSORT extension for RCTs using cohorts and routinely collected health data","year":2018,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Multiple Sclerosis Society of Canada; Jewish General Hospital; University of Ottawa; Dalhousie University; McMaster University; Western University; Children's Hospital of Eastern Ontario; Impact; Ottawa Hospital","funders":"Medical Research Council; Canadian Institutes of Health Research; Wellcome Trust","keywords":"Consolidated Standards of Reporting Trials; Checklist; Observational study; Generalizability theory; Guideline; Protocol (science); Randomized controlled trial; Medicine; Systematic review; Strengthening the reporting of observational studies in epidemiology; Research design; MEDLINE; Family medicine; Alternative medicine; Psychology; Statistics; Political science; Surgery","retraction":null,"screen_n_in":null,"score":{"opus":0.9111252944079001,"gpt":0.6940410504859482,"spread":0.2170842439219519,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01561267,0.0001140584,0.0004502636,0.00005104553,0.0004434962,0.00003481643,0.0002652231,0.00006105638,0.00001869654],"category_scores_gemma":[0.0156113,0.00006730441,0.00002370974,0.0002179659,0.000318458,0.0001003949,0.0003323881,0.0003186277,3.379484e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006200175,"about_ca_system_score_gemma":0.0008026611,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006470225,"about_ca_topic_score_gemma":0.0002916973,"domain_scores_codex":[0.9981105,0.0002660592,0.0005622095,0.0002941402,0.0004705448,0.0002964861],"domain_scores_gemma":[0.9944521,0.002219833,0.0002058533,0.0005344789,0.002496239,0.00009151414],"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":[0.000864742,0.0005681352,0.0004525841,0.1187219,0.0002278434,0.000001678993,0.002241359,3.442658e-8,0.00171712,0.07689913,0.4504331,0.3478723],"study_design_scores_gemma":[0.0008019062,0.000825253,0.0002530557,0.02502936,0.00004484283,0.000034734,0.0001647791,0.002758604,0.00318171,0.0810976,0.8855781,0.0002300832],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"protocol","genre_gemma":"methods","genre_scores_codex":[0.001077239,0.005198083,0.4002696,0.01196465,0.00003192783,0.5808006,0.0003071782,0.0001033786,0.0002472988],"genre_scores_gemma":[0.001593626,0.0024175,0.8560075,0.0003719965,0.00008686716,0.1389914,0.0000587365,0.00003877646,0.0004335867],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.4557379,"threshold_uncertainty_score":0.9926806,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4296369701","doi":"10.1186/s41073-022-00125-x","title":"Reducing the Inadvertent Spread of Retracted Science: recommendations from the RISRS report","year":2022,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Lethbridge","funders":"Alfred P. Sloan Foundation","keywords":"Stakeholder; Public relations; Publishing; Scholarship; Stakeholder engagement; Political science; Process (computing); Business; Computer science; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.219800253043156,"gpt":0.4836677666894115,"spread":0.2638675136462555,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0924817,0.00008778732,0.0002318003,0.0000798831,0.005359447,0.0001226527,0.001383012,0.000145607,0.002433873],"category_scores_gemma":[0.03381363,0.00005082109,0.0000908931,0.001864448,0.002209218,0.0002501645,0.0005500783,0.008083642,0.00001049331],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002524123,"about_ca_system_score_gemma":0.001473989,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.08705031,"about_ca_topic_score_gemma":0.001403148,"domain_scores_codex":[0.9910982,0.00384307,0.0005063628,0.000355029,0.003730877,0.0004664793],"domain_scores_gemma":[0.9937873,0.003409265,0.0002261981,0.0005654085,0.001861577,0.0001502265],"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":[0.00001753413,0.00008717434,0.001160153,0.0001055307,0.0000387376,0.00001700385,0.02160889,7.2351e-7,0.0001115624,0.06213555,0.799471,0.1152462],"study_design_scores_gemma":[0.00004095125,0.00002921123,0.001333947,0.000651827,0.00002288944,0.00002445144,0.01469023,0.00003592361,0.00003869072,0.007895649,0.9751678,0.00006840542],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.01994389,0.03922785,0.00003564593,0.8866392,0.0006741802,0.001314559,0.0002258845,0.0000294896,0.05190926],"genre_scores_gemma":[0.7855555,0.1947933,0.0003754806,0.004096272,0.0006412353,0.0002655303,0.0001762487,0.00001504484,0.01408143],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.882543,"threshold_uncertainty_score":0.9984781,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2531842628","doi":"10.1186/s41073-016-0022-7","title":"Is it becoming harder to secure reviewers for peer review? A test with data from five ecology journals","year":2016,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"World Wildlife Fund Canada; University of British Columbia; Women's Health Research Institute","funders":"","keywords":"Ecology; Community; Publishing; Functional ecology; Evolutionary ecology; Rhetorical question; Biology; Habitat; Political science; Ecosystem; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.9221730707033521,"gpt":0.7096993867356763,"spread":0.2124736839676757,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","scholarly_communication","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.2290588,0.0003113286,0.001402878,0.008199491,0.0005751274,0.001818981,0.005888099,0.0002072314,0.009945498],"category_scores_gemma":[0.6551285,0.0001458161,0.0002353563,0.04588677,0.0003807069,0.001072985,0.002725561,0.001550264,0.001713916],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001878836,"about_ca_system_score_gemma":0.0009277498,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003106674,"about_ca_topic_score_gemma":0.0004576737,"domain_scores_codex":[0.9717272,0.001787334,0.001553273,0.001922306,0.02165881,0.001351069],"domain_scores_gemma":[0.6633531,0.1357204,0.001503326,0.01104961,0.1831023,0.005271216],"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":[0.00002262712,0.00007156259,0.004205429,0.0009697998,0.00003650592,0.00001250371,0.00005983887,1.213001e-8,0.00002450694,0.00005459077,0.8221684,0.1723743],"study_design_scores_gemma":[0.0004255371,0.0004206226,0.003107212,0.01907452,0.000047578,0.00002809512,0.0001308452,0.00005704773,0.00002755196,0.002012026,0.9744233,0.0002456388],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.0005110418,0.3405317,0.001447545,0.650457,0.0001540503,0.002460596,0.003103672,0.000009509171,0.001324894],"genre_scores_gemma":[0.001295789,0.9054024,0.004976423,0.02056897,0.0002999484,0.0002700661,0.00009847949,0.0000304212,0.06705754],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.629888,"threshold_uncertainty_score":0.9994905,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","bibliometrics"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2947766279","doi":"10.1186/s41073-019-0068-4","title":"Measuring the data gap: inclusion of sex and gender reporting in diabetes research","year":2019,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Sex and Gender in Healthcare","field":"Medicine","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Canada Research Chairs; Toronto Public Health; Public Health Ontario; Women's College Hospital","funders":"Diabetes Action Canada; University of Toronto; Ontario Ministry of Health and Long-Term Care; Diabetes Action Research and Education Foundation","keywords":"Observational study; Diabetes mellitus; Medicine; Type 2 diabetes; Inclusion (mineral); Research design; Randomized controlled trial; Sex characteristics; Family medicine; Psychology; Social psychology; Internal medicine; Social science; Endocrinology","retraction":null,"screen_n_in":null,"score":{"opus":0.8197748069475616,"gpt":0.5915464012853751,"spread":0.2282284056621865,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09723032,0.00009767333,0.0004914823,0.0001923562,0.0002931586,0.00003062424,0.0003717261,0.0001146181,0.00009647296],"category_scores_gemma":[0.0246959,0.00005823615,0.00003419258,0.0006637338,0.0002463775,0.0001104305,0.002599691,0.002767559,0.00001041584],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005997218,"about_ca_system_score_gemma":0.0003725204,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001151962,"about_ca_topic_score_gemma":0.0001404993,"domain_scores_codex":[0.9936758,0.00189302,0.0008730603,0.0005590454,0.00245282,0.0005463086],"domain_scores_gemma":[0.9952435,0.001719384,0.0001399339,0.001275815,0.001449484,0.0001718646],"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":[0.00005137367,0.0001033205,0.6566786,0.05599866,0.00005107445,0.00004144869,0.004466078,8.350322e-8,0.001058199,0.0006241662,0.00708623,0.2738407],"study_design_scores_gemma":[0.001683487,0.0009603894,0.3770079,0.06098404,0.00008231059,0.0003076058,0.01115272,0.002060258,0.001877452,0.01928906,0.5241333,0.0004615864],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6584089,0.3016471,7.325793e-7,0.03374708,0.00003838572,0.001493681,0.00001386517,0.00000816163,0.004642054],"genre_scores_gemma":[0.9094144,0.0892553,0.00009899204,0.0003379301,0.00007927672,0.00002621878,0.0000426539,0.00001194299,0.0007333388],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.517047,"threshold_uncertainty_score":0.9995331,"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"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4366783310","doi":"10.1186/s41073-023-00128-2","title":"Scientific sinkhole: estimating the cost of peer review based on survey data with snowball sampling","year":2023,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Prince Edward Island; Agricultural Research Institute of Ontario","funders":"","keywords":"Snowball sampling; Scopus; Peer review; Psychology; Demography; Medicine; Actuarial science; Geography; Statistics; MEDLINE; Political science; Business; Sociology; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.952932832349722,"gpt":0.7128552951760203,"spread":0.2400775371737017,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","scholarly_communication","open_science"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5547979,0.000215187,0.0007798712,0.0109458,0.001065403,0.003532759,0.006520464,0.00009799213,0.0006333195],"category_scores_gemma":[0.6385837,0.0001025496,0.0001148882,0.1926906,0.001200418,0.0005447088,0.002348496,0.002160765,0.0005419625],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000745736,"about_ca_system_score_gemma":0.001207424,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006840166,"about_ca_topic_score_gemma":0.0003021152,"domain_scores_codex":[0.9388121,0.005541166,0.001455888,0.001816079,0.05110046,0.001274328],"domain_scores_gemma":[0.7242428,0.1265315,0.001107344,0.012871,0.1335459,0.001701529],"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":[0.00002087506,0.00009121836,0.007169446,0.002521436,0.00001324325,0.000009521886,0.00002616507,0.0000294739,0.000006837244,0.0001836036,0.691698,0.2982302],"study_design_scores_gemma":[0.0002459729,0.0002283798,0.0314625,0.01544363,0.00002146729,0.000009838818,0.00007448829,0.06817354,0.00001471592,0.0007085177,0.883397,0.0002199725],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.008900476,0.455631,0.01418203,0.4871189,0.001543781,0.01093427,0.007058043,0.0001573273,0.0144741],"genre_scores_gemma":[0.2751945,0.6040291,0.01877737,0.009009569,0.0006369592,0.0006736663,0.007404114,0.0002089855,0.08406579],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.4781094,"threshold_uncertainty_score":0.9988548,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","bibliometrics"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","scholarly_communication"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4408038925","doi":"10.1186/s41073-025-00158-y","title":"Policies on artificial intelligence chatbots among academic publishers: a cross-sectional audit","year":2025,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Ottawa Hospital; Impact","funders":"","keywords":"Chatbot; Audit; Computer science; Transformative learning; World Wide Web; Library science; Public relations; Political science; Psychology; Business; Accounting","retraction":null,"screen_n_in":null,"score":{"opus":0.6237231184858077,"gpt":0.6160312006266995,"spread":0.007691917859108277,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.008352813,0.0001684647,0.0003710777,0.0004487284,0.0004701788,0.0003019768,0.0002616634,0.0003534938,0.000660137],"category_scores_gemma":[0.01577879,0.000134996,0.0001220683,0.00120543,0.0006705475,0.0003913959,0.0001101551,0.003730647,0.0002794098],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002946195,"about_ca_system_score_gemma":0.0009628538,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002790486,"about_ca_topic_score_gemma":0.0002185245,"domain_scores_codex":[0.9965356,0.0003653231,0.0007683762,0.0004910725,0.001280224,0.0005594142],"domain_scores_gemma":[0.9956701,0.0009248956,0.0000818857,0.0003963979,0.002611365,0.0003153998],"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":[0.0002851589,0.000333677,0.2333959,0.006994495,0.00007403106,0.00001307895,0.0007839406,0.000002187484,0.00006997331,0.09507502,0.1112669,0.5517056],"study_design_scores_gemma":[0.00005877093,0.000936456,0.3473027,0.02571178,0.0000741181,0.00007046577,0.002363883,0.0004868837,0.005004071,0.1324263,0.4851007,0.0004638037],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5445651,0.05322115,0.0004908576,0.3713416,0.001495821,0.002853712,0.00002949944,0.0001425287,0.02585961],"genre_scores_gemma":[0.9324075,0.04827111,0.00007507778,0.005156912,0.0009213196,0.000255044,0.00007359401,0.00001615762,0.01282325],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5512418,"threshold_uncertainty_score":0.9985678,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","scholarly_communication"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3132467447","doi":"10.1186/s41073-020-00107-x","title":"Re-evaluation of solutions to the problem of unprofessionalism in peer review","year":2021,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic Publishing and Open Access","field":"Decision Sciences","cited_by":17,"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; Fisheries and Oceans Canada; Environment and Climate Change Canada; University of Victoria","funders":"","keywords":"Offensive; Conversation; Peer review; Social media; Psychology; Sociology; Social psychology; Computer science; Law; Political science; Operations research; World Wide Web","retraction":null,"screen_n_in":null,"score":{"opus":0.7094531083437046,"gpt":0.6287024361511451,"spread":0.08075067219255949,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2819699,0.00009575415,0.0005718995,0.0001826249,0.0001854319,0.0001552393,0.001547731,0.0001183457,0.001433996],"category_scores_gemma":[0.201267,0.00005101323,0.000116526,0.003699407,0.0001826376,0.000514662,0.0009163097,0.001775188,0.0000642083],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006998257,"about_ca_system_score_gemma":0.001371292,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006018666,"about_ca_topic_score_gemma":0.001005011,"domain_scores_codex":[0.9691109,0.01035616,0.001741365,0.0006284803,0.01770201,0.0004610716],"domain_scores_gemma":[0.955028,0.004851614,0.0003956603,0.001194209,0.03830307,0.0002274694],"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":[0.000003933058,0.00004810102,0.0005970548,0.001593876,0.000006950005,0.000001008172,0.0003656769,0.00000239444,0.000009592864,0.00675879,0.8097025,0.1809101],"study_design_scores_gemma":[0.0001145576,0.0000192341,0.003019061,0.02888561,0.00002894348,0.000005791974,0.0008566795,0.0001142422,0.00004753977,0.03025464,0.9365847,0.00006899334],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.001186619,0.3764853,0.00005340109,0.607312,0.00006538704,0.001058891,0.00005780896,0.000002890493,0.01377769],"genre_scores_gemma":[0.1653191,0.752661,0.002119474,0.01391011,0.0002438522,0.0007509947,0.0001785737,0.00002534027,0.06479159],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.5934018,"threshold_uncertainty_score":0.9994788,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity","scholarly_communication"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4213005437","doi":"10.1186/s41073-022-00121-1","title":"Characteristics of ‘mega’ peer-reviewers","year":2022,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; St. Michael's Hospital; University of Ottawa; Ottawa Public Health; University of Toronto; McGill University; Institute for Work & Health; Ottawa Hospital","funders":"","keywords":"Peer review; Receipt; Mega-; Control (management); Peer group; Psychology; Medical education; Medicine; Computer science; Social psychology; World Wide Web; Political science","retraction":null,"screen_n_in":null,"score":{"opus":0.850221086539673,"gpt":0.6727948306867655,"spread":0.1774262558529075,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","bibliometrics"],"category_scores_codex":[0.2737059,0.0001510799,0.0008125117,0.01487053,0.0006951483,0.000750122,0.003101848,0.00007152838,0.008623855],"category_scores_gemma":[0.2952558,0.0001029139,0.0002532827,0.09895887,0.0004685553,0.0002879187,0.002872334,0.002721567,0.0001916322],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001495078,"about_ca_system_score_gemma":0.0005934954,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002895991,"about_ca_topic_score_gemma":0.000007003721,"domain_scores_codex":[0.9455048,0.003957142,0.00147334,0.0008800773,0.04729683,0.0008878273],"domain_scores_gemma":[0.9345699,0.0145399,0.0006186424,0.002243379,0.0469275,0.001100709],"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":[0.00001566444,0.0001534083,0.009204247,0.0006625865,0.00001267432,0.00001779304,0.00007860716,1.313724e-7,0.00003621878,0.005264833,0.5241783,0.4603755],"study_design_scores_gemma":[0.000115795,0.0002698557,0.01133053,0.0002559369,0.000007696708,0.00002553641,0.000202586,0.0001376688,0.00002366545,0.004009641,0.9835073,0.0001138034],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.07430355,0.5997828,0.0003899149,0.2703451,0.001583376,0.003260995,0.001869465,0.00004638562,0.04841843],"genre_scores_gemma":[0.5085676,0.4039107,0.0005851237,0.001452885,0.0001750858,0.000249198,0.0001042961,0.00002879157,0.0849263],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.4602617,"threshold_uncertainty_score":0.9995792,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","bibliometrics"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","bibliometrics"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W2484353070","doi":"10.1186/s41073-016-0012-9","title":"Proceedings of the 4th World Conference on Research Integrity","year":2016,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Government of Canada; Dalhousie University; Treasury Board of Canada Secretariat; University of Calgary; McGill University; University of Alberta; Université de Montréal; Brandon University; Queen's University; Fonds de recherche du Québec; Fonds de Recherche du Québec - Santé","funders":"National Research Foundation of Korea; University of New South Wales; National Research Foundation","keywords":"Research integrity; Political science; Engineering ethics; Computer science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.8209765703090168,"gpt":0.6145609508358271,"spread":0.2064156194731898,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.01752,0.0001181206,0.0003627552,0.000245634,0.0002972816,0.00004490025,0.0003518973,0.0001377997,0.0006784876],"category_scores_gemma":[0.01691943,0.00005445781,0.00009524931,0.00111405,0.0009299681,0.0001221979,0.0001640074,0.003212725,0.0001770559],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002003054,"about_ca_system_score_gemma":0.0007982599,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001830325,"about_ca_topic_score_gemma":0.0004332511,"domain_scores_codex":[0.9960629,0.0004483806,0.0004856851,0.0003675807,0.002120031,0.0005154674],"domain_scores_gemma":[0.988508,0.001470753,0.00009011937,0.0004379693,0.00917258,0.0003206031],"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":[0.0002551998,0.0003750639,0.03001672,0.008421923,0.00002871497,0.000002686321,0.0009031129,2.193879e-9,0.002414503,0.1922182,0.327184,0.4381799],"study_design_scores_gemma":[0.0001436311,0.001383,0.02360809,0.09573432,0.00004292327,0.00004713318,0.001808975,0.00001957614,0.03772318,0.08459098,0.7546656,0.0002325841],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2059508,0.01110763,0.000006504814,0.7280154,0.0002308133,0.002291376,0.00002190031,0.00002646729,0.05234905],"genre_scores_gemma":[0.9173712,0.05203346,0.00006558238,0.0007476716,0.000250483,0.0001192123,0.000002545598,0.0000120437,0.02939777],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7272677,"threshold_uncertainty_score":0.9990869,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["research_integrity"],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["research_integrity"],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2626981796","doi":"10.1186/s41073-017-0037-8","title":"Reviewer training to assess knowledge translation in funding applications is long overdue","year":2017,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Michael Smith Health Research BC","funders":"Michael Smith Health Research BC","keywords":"Knowledge translation; Excellence; Relevance (law); Accreditation; Political science; Public relations; Health care; Medical education; Modalities; Business; Engineering ethics; Knowledge management; Medicine; Computer science; Sociology; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.9780282322318703,"gpt":0.8145573755709148,"spread":0.1634708566609555,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04339055,0.0001284868,0.0004502865,0.0002541782,0.002358649,0.0001091012,0.0006051311,0.0001305701,0.000900335],"category_scores_gemma":[0.0374746,0.0001097535,0.000050411,0.0006750843,0.000142104,0.000482388,0.0002175547,0.00213282,0.0008200983],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002680535,"about_ca_system_score_gemma":0.001273749,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008469955,"about_ca_topic_score_gemma":0.003266067,"domain_scores_codex":[0.9942305,0.002516332,0.0009091821,0.0004517504,0.0009852786,0.0009069495],"domain_scores_gemma":[0.9917337,0.003856732,0.0002250189,0.0007860528,0.002844299,0.0005541955],"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":[0.000008074674,0.00004107846,0.0208257,0.01790752,0.000007963665,0.000002322494,0.02645496,5.226758e-8,0.00004339022,0.01103776,0.1565444,0.7671267],"study_design_scores_gemma":[0.0002086863,0.00002404374,0.03377288,0.01040514,0.000005429134,0.00000104109,0.0008299635,0.00002518947,0.000004067213,0.0005532522,0.954066,0.0001043091],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.01098381,0.06029652,0.009648537,0.7901394,0.0008749405,0.02033677,0.0003315138,0.00009749741,0.107291],"genre_scores_gemma":[0.4301021,0.4108197,0.009017775,0.08422105,0.003290204,0.01918552,0.0001137446,0.0001346633,0.04311527],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.7975215,"threshold_uncertainty_score":0.9999579,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4414042524","doi":"10.1186/s41073-025-00178-8","title":"Attitudes and perceptions of biomedical journal editors in chief towards the use of artificial intelligence chatbots in the scholarly publishing process: a cross-sectional survey","year":2025,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Impact; Ottawa Hospital; University of Ottawa","funders":"","keywords":"Publishing; Perception; Chatbot; Human intelligence; Applications of artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.6469506747522318,"gpt":0.5911150780951746,"spread":0.05583559665705717,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.02927154,0.00008368504,0.0002801647,0.0003905101,0.0002399986,0.0006258233,0.000279545,0.0001662963,0.00007484509],"category_scores_gemma":[0.05384826,0.00004642928,0.00005230947,0.001559176,0.00081633,0.0009734841,0.0000729067,0.003029228,0.00000145652],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008394357,"about_ca_system_score_gemma":0.001296723,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007626436,"about_ca_topic_score_gemma":0.002434705,"domain_scores_codex":[0.9960789,0.001313329,0.0008760002,0.0002065054,0.001274019,0.0002512508],"domain_scores_gemma":[0.993565,0.002239192,0.0001141374,0.00022403,0.003746616,0.0001109813],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001069268,0.0003635967,0.849413,0.002919183,0.00001630126,0.000002930904,0.002647341,0.000001192277,0.00002408384,0.001308693,0.002687544,0.1405092],"study_design_scores_gemma":[0.00002375633,0.0001683113,0.9837444,0.004773661,0.000007912343,0.00004104876,0.002264587,0.0001522843,0.00004106792,0.004462424,0.004278178,0.0000423239],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9221755,0.0128866,0.0002336741,0.06347555,0.0004052693,0.0007363022,0.00002568503,0.000003188768,0.00005818481],"genre_scores_gemma":[0.9692884,0.02967983,0.0000890833,0.0003927158,0.000427971,0.00004399423,0.0000235447,0.000003592588,0.00005087384],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1404669,"threshold_uncertainty_score":0.9995692,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4410336353","doi":"10.1186/s41073-025-00164-0","title":"Research on policy mechanisms to address funding bias and conflicts of interest in biomedical research: a scoping review","year":2025,"lang":"en","type":"review","venue":"Research Integrity and Peer Review","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":6,"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":"U.S. National Library of Medicine; National Institute of General Medical Sciences","keywords":"Conflict of interest; Political science; Management science; Public relations; Engineering ethics; Economics; Engineering; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.9766357963041572,"gpt":0.7866428425180303,"spread":0.1899929537861269,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.1473637,0.0007197057,0.004270055,0.004364772,0.0008481689,0.0001321074,0.0017339,0.0024166,0.001087575],"category_scores_gemma":[0.06506041,0.0005686487,0.0003630357,0.008788653,0.00213564,0.0001467559,0.00251634,0.03759196,0.0002437498],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008070603,"about_ca_system_score_gemma":0.005038619,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008451713,"about_ca_topic_score_gemma":0.0003655236,"domain_scores_codex":[0.9473125,0.04271347,0.002667989,0.001704215,0.002726099,0.002875715],"domain_scores_gemma":[0.9602922,0.03122833,0.0003437246,0.001139495,0.004008885,0.002987371],"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":[0.00006684933,0.0002447952,0.00000197437,0.5019943,0.00008216542,0.0001160362,0.0001049963,1.945823e-8,0.000003355199,0.02868346,0.02361383,0.4450882],"study_design_scores_gemma":[0.0002061673,0.000374457,0.0000010167,0.4847133,0.00009922965,0.00003842868,0.00006379286,0.000002320381,0.00001528091,0.0008840082,0.5133937,0.0002082778],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001601169,0.9304728,0.000002479139,0.05336383,0.0002610325,0.01061553,0.0005510664,0.00003148795,0.004685767],"genre_scores_gemma":[0.00009488783,0.9903588,0.00004131506,0.004533336,0.0004028478,0.001987303,0.0001633208,0.00005806336,0.002360107],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4897799,"threshold_uncertainty_score":0.9998255,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"incentives","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"incentives","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W3135000317","doi":"10.1186/s41073-021-00108-4","title":"Steps toward preregistration of research on research integrity","year":2021,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Women's College Research Institute","keywords":"Research integrity; USable; Psychology; Data integrity; Set (abstract data type); Documentation; Data collection; Qualitative research; Computer science; Data science; Public relations; Sociology; Political science; World Wide Web; Database; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.6142879617107355,"gpt":0.5984877638154373,"spread":0.01580019789529818,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.1875211,0.0002137314,0.0006724013,0.0005679396,0.001898036,0.0003562797,0.001232162,0.001417156,0.001820145],"category_scores_gemma":[0.09845751,0.000182184,0.0001928642,0.0035082,0.003723683,0.000439047,0.0005327136,0.02854408,0.0003601904],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005153292,"about_ca_system_score_gemma":0.004282423,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01894887,"about_ca_topic_score_gemma":0.004573002,"domain_scores_codex":[0.9596282,0.02576439,0.0008615333,0.001002087,0.01110722,0.001636511],"domain_scores_gemma":[0.9682428,0.00992643,0.0001169684,0.0009018735,0.02017983,0.0006320587],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000933581,0.0003126375,0.000548961,0.003646466,0.00004113092,0.00006160583,0.007197763,1.250669e-7,0.0001489737,0.629034,0.3253077,0.03360724],"study_design_scores_gemma":[0.0001853498,0.0003207902,0.0005183181,0.009201858,0.00001254167,0.00001439964,0.01611089,0.00001685685,0.001324823,0.06082572,0.9112777,0.0001907301],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.02285616,0.0979009,0.0001007052,0.4291829,0.0004750595,0.002710815,0.0001796428,0.00008953255,0.4465043],"genre_scores_gemma":[0.5347802,0.4146788,0.0006302941,0.0006894488,0.001073769,0.0002033821,0.0001208087,0.00003360252,0.04778963],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.58597,"threshold_uncertainty_score":0.9998792,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity","open_science"],"domain":"reproducibility","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3102946754","doi":"10.1186/s41073-020-00102-2","title":"Doing better: eleven ways to improve the integration of sex and gender in health research proposals","year":2020,"lang":"en","type":"letter","venue":"Research Integrity and Peer Review","topic":"Sex and Gender in Healthcare","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Public Health Ontario; Women's College Hospital","funders":"Canadian Institutes of Health Research; Diabetes Action Research and Education Foundation","keywords":"Gender identity; Relevance (law); Psychology; Medical education; Medicine; Political science; Social psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.5665995736985263,"gpt":0.5392911544766058,"spread":0.02730841922192051,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.03102079,0.0003007553,0.001226765,0.0004951235,0.0003014108,0.00009551726,0.0004064453,0.0004941071,0.00008162264],"category_scores_gemma":[0.006944293,0.0001837462,0.0001163147,0.001178168,0.0004107731,0.00007333532,0.000423964,0.01567206,0.00002812082],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003042324,"about_ca_system_score_gemma":0.001585996,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003731957,"about_ca_topic_score_gemma":0.0002829063,"domain_scores_codex":[0.9890624,0.004209132,0.001128003,0.001000347,0.003513085,0.001086994],"domain_scores_gemma":[0.9944538,0.001956046,0.0001275869,0.0008370818,0.00221856,0.0004069064],"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":[0.00004060342,0.00002468725,0.0001628296,0.05884688,0.00003777658,0.00008529025,0.003271348,6.563495e-9,0.00006242913,0.0002427518,0.6601764,0.277049],"study_design_scores_gemma":[0.0002393863,0.0009145175,0.0002284678,0.01583119,0.00001880273,0.00007242207,0.0007963245,0.00002392722,0.00009249752,0.004464317,0.9771733,0.0001448922],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0003031652,0.1070422,0.00008023688,0.8865591,0.00007280881,0.005054783,0.0001057363,0.00001569424,0.0007662222],"genre_scores_gemma":[0.003411049,0.2195916,0.002418173,0.7689329,0.002330897,0.0007444006,0.0004470239,0.00008954419,0.002034345],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.3169969,"threshold_uncertainty_score":0.997768,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4403162971","doi":"10.1186/s41073-024-00152-w","title":"An evaluation of the preprints produced at the beginning of the 2022 mpox public health emergency","year":2024,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic Publishing and Open Access","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Guelph; Public Health Agency of Canada","funders":"","keywords":"Observational study; Quality (philosophy); Preprint; Scale (ratio); Critical appraisal; Public health; Medicine; Psychology; Computer science; World Wide Web; Alternative medicine; Geography; Nursing; Internal medicine; Pathology","retraction":null,"screen_n_in":null,"score":{"opus":0.7197556863314704,"gpt":0.6343461286640703,"spread":0.0854095576674001,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.276915,0.00009196311,0.0002815051,0.0001075473,0.0006118762,0.000394325,0.003549468,0.00008118233,0.002212278],"category_scores_gemma":[0.09536604,0.00003376755,0.0001488873,0.002954163,0.0003921333,0.0007960654,0.001370279,0.002053831,0.00004280829],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001280992,"about_ca_system_score_gemma":0.001682641,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007908361,"about_ca_topic_score_gemma":0.0003110161,"domain_scores_codex":[0.9734395,0.01288223,0.001072699,0.0006751082,0.01153744,0.0003930898],"domain_scores_gemma":[0.9911002,0.001418436,0.0003625616,0.001926132,0.005047713,0.0001449362],"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":[0.000004422249,0.00004341534,0.01830688,0.0007724628,0.00002376789,1.006544e-7,0.002420458,0.000004086389,0.0001342247,0.004659719,0.4699679,0.5036626],"study_design_scores_gemma":[0.00008370062,0.00003932952,0.04860717,0.003090966,0.00002951149,0.00001297629,0.00120703,0.003026658,0.0004610021,0.02822045,0.9151257,0.00009548046],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2519448,0.1786062,0.00002128594,0.5611451,0.0007082709,0.00203765,0.00005132662,0.00001117261,0.005474202],"genre_scores_gemma":[0.9669334,0.0245701,0.000007523251,0.0003770936,0.00009763451,0.00008601362,0.000005410987,0.000007308281,0.007915563],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7149886,"threshold_uncertainty_score":0.9986998,"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","scholarly_communication","open_science"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4411363529","doi":"10.1186/s41073-025-00166-y","title":"Gaps in the Ottawa Statement on the Ethical Design and Conduct of Cluster Randomized Trials: a citation analysis reveals a need for updated ethics guidelines","year":2025,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Bruyère; Public Health Ontario; St. Joseph’s Healthcare Hamilton; Université de Montréal; Centre Hospitalier de l’Université de Montréal; McMaster University; Western University; University of Toronto; University Health Network; Ottawa Hospital; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Statement (logic); Citation; Cluster (spacecraft); Randomized controlled trial; Research ethics; Engineering ethics; Political science; Psychology; Library science; Medicine; Law; Engineering; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.9072430311787694,"gpt":0.7270061267709451,"spread":0.1802369044078242,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6945913,0.000159033,0.001934384,0.0004761047,0.0002159825,0.0001014945,0.0003649794,0.0006702728,0.0000808367],"category_scores_gemma":[0.8636664,0.00006873191,0.0004221501,0.002035551,0.001119949,0.00003157284,0.0001372049,0.01027589,0.000002011741],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006766484,"about_ca_system_score_gemma":0.001266427,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005086185,"about_ca_topic_score_gemma":0.000428024,"domain_scores_codex":[0.897341,0.09093803,0.004635852,0.0007989822,0.005639344,0.0006467879],"domain_scores_gemma":[0.07802435,0.9020792,0.0005509135,0.00120961,0.01792403,0.0002119379],"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":[0.05901787,0.0004481026,0.0001881379,0.02692214,0.003009379,0.000009985219,0.002595547,0.00001273987,0.0002865691,0.4531465,0.4364926,0.01787043],"study_design_scores_gemma":[0.08170848,0.0015937,0.0003849425,0.04638517,0.004448672,0.000006205564,0.00323575,0.01135047,0.0003727045,0.6810714,0.1691453,0.000297193],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.0021761,0.02842633,0.01051537,0.9490704,0.00003572689,0.009301721,0.0000382568,0.000006374712,0.0004297443],"genre_scores_gemma":[0.1447584,0.7350038,0.01533523,0.0958292,0.000172679,0.003614863,0.0002231463,0.00002975727,0.005032953],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.8532411,"threshold_uncertainty_score":0.9920074,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4389155531","doi":"10.1186/s41073-023-00139-z","title":"The quizzical failure of a nudge on academic integrity education: a randomized controlled trial","year":2023,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital","funders":"Magyar Tudományos Akadémia; European Commission","keywords":"Academic integrity; Randomized controlled trial; Nudge theory; Research integrity; Psychology; Political science; Medicine; Social psychology; Public relations; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.186925579545847,"gpt":0.5119654922768404,"spread":0.3250399127309934,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1378661,0.0002100581,0.001518551,0.0002130446,0.00135338,0.0001373835,0.000987616,0.001100602,0.0003793431],"category_scores_gemma":[0.1619852,0.0001148299,0.0005672288,0.001339295,0.002447705,0.0001751127,0.0001647464,0.01453034,0.0002593405],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001264417,"about_ca_system_score_gemma":0.002838335,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003009554,"about_ca_topic_score_gemma":0.0006043278,"domain_scores_codex":[0.9796174,0.01451619,0.001162818,0.0004188227,0.003515522,0.0007693088],"domain_scores_gemma":[0.972492,0.02411231,0.0002825272,0.0003734322,0.002405079,0.0003346083],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.1170005,0.00007282056,0.0000107597,0.0003129878,0.00007585494,0.000001442318,0.001993831,2.099194e-8,0.000003070193,0.473337,0.3900286,0.01716311],"study_design_scores_gemma":[0.1370864,0.0001207271,0.000007375916,0.002369501,0.0000726131,0.000002399956,0.003065432,0.00002920488,0.000006703823,0.04967918,0.8074293,0.0001311896],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.0127219,0.07131412,0.00002235784,0.8393352,0.001188657,0.01212501,0.00003673583,0.0001166484,0.06313936],"genre_scores_gemma":[0.07860731,0.85456,0.00009727312,0.00364801,0.001955495,0.003012512,0.0000524458,0.00002596514,0.05804101],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.8356872,"threshold_uncertainty_score":0.9999467,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4396214656","doi":"10.1186/s41073-024-00145-9","title":"Extent, transparency and impact of industry funding for pelvic mesh research: a review of the literature","year":2024,"lang":"en","type":"review","venue":"Research Integrity and Peer Review","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"Queen's University","keywords":"Transparency (behavior); Declaration; Urinary incontinence; Medicine; Conflict of interest; Payment; Accounting; Family medicine; Business; Political science; Surgery; Finance; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.9203878257127537,"gpt":0.748278658587292,"spread":0.1721091671254616,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.04236021,0.0006156278,0.003047858,0.0004379078,0.0005203458,0.00007707109,0.001143959,0.002793811,0.0008172584],"category_scores_gemma":[0.008576307,0.0003406541,0.001335712,0.003008196,0.001448355,0.0001518138,0.0005692553,0.03475903,0.00001822659],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002788163,"about_ca_system_score_gemma":0.002535825,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007531668,"about_ca_topic_score_gemma":0.00001091279,"domain_scores_codex":[0.9839969,0.01061426,0.001805169,0.0008747433,0.001460404,0.0012485],"domain_scores_gemma":[0.990273,0.004615862,0.0004079899,0.0007391567,0.003076539,0.0008874374],"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":[0.00002805852,0.00009190098,0.00001530685,0.568518,0.0003627613,0.00001291388,0.0001250908,6.854969e-9,0.00000290932,0.002055645,0.09429628,0.3344911],"study_design_scores_gemma":[0.0001217508,0.0001818565,0.000005958663,0.4300235,0.0012944,0.00008534304,0.00001975106,0.000004108632,0.00000423162,0.00069954,0.5674029,0.0001566432],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00002211022,0.976049,4.540314e-7,0.01292093,0.0003858208,0.006304102,0.003231498,0.00001814335,0.001067927],"genre_scores_gemma":[0.0001245743,0.9945471,0.00001415742,0.000570514,0.000392021,0.000779482,0.0002032763,0.00005699878,0.003311826],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4731066,"threshold_uncertainty_score":0.9999046,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4386880731","doi":"10.1186/s41073-023-00138-0","title":"Bridges of perspectives: representation of people with lived experience of spinal cord injury in editorial boards and peer review","year":2023,"lang":"en","type":"article","venue":"Research Integrity and Peer Review","topic":"Spinal Cord Injury Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia Hospital; NeuroDevNet; University of British Columbia","funders":"Government of Canada","keywords":"Diversity (politics); Lived experience; Spinal cord injury; Representation (politics); Best practice; Publishing; Psychology; Quality of life (healthcare); Medicine; Public relations; Political science; Nursing; Spinal cord; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.2538299194266069,"gpt":0.5465424787309605,"spread":0.2927125593043536,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009440184,0.000146988,0.001012911,0.0003573325,0.00004861723,0.00001471403,0.0002130011,0.00009865934,0.0001422099],"category_scores_gemma":[0.02088176,0.0001073097,0.0000926355,0.002135672,0.0006987138,0.0001665791,0.0002245167,0.001180314,0.00000547501],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007739518,"about_ca_system_score_gemma":0.0004935469,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003180855,"about_ca_topic_score_gemma":0.00008866534,"domain_scores_codex":[0.9941384,0.000687025,0.0006504435,0.0004241838,0.003763754,0.0003361652],"domain_scores_gemma":[0.9915802,0.0003737432,0.0001677449,0.0004617774,0.007241275,0.0001752104],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.03517473,0.001137381,0.1333279,0.2918286,0.0002769673,0.0001146026,0.007644009,1.44401e-7,0.0108666,0.001915506,0.3964931,0.1212205],"study_design_scores_gemma":[0.003316403,0.04717989,0.4875409,0.2986793,0.0003206944,0.0001737208,0.03073905,0.00008681729,0.009003446,0.0003630849,0.1220044,0.0005922951],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"review","genre_scores_codex":[0.8170268,0.1556384,0.000008900503,0.02367708,0.0002833883,0.002587646,0.00006834324,0.00001830679,0.00069118],"genre_scores_gemma":[0.4507478,0.5473468,0.0001726706,0.00004107427,0.0005580622,0.0001607844,0.00003548032,0.00001637306,0.0009210408],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3917084,"threshold_uncertainty_score":0.9873658,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","scholarly_communication"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4285393211","doi":"10.1186/s41073-022-00124-y","title":"Correction: Characteristics of ‘mega’ peer-reviewers","year":2022,"lang":"en","type":"erratum","venue":"Research Integrity and Peer Review","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; St. Michael's Hospital; University of Ottawa; Ottawa Public Health; University of Toronto; McGill University; Institute for Work & Health; Ottawa Hospital","funders":"","keywords":"Mega-; Computer science; Peer review; Information retrieval; Library science; Political science; Physics; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.3677572237548847,"gpt":0.5524705717150645,"spread":0.1847133479601798,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07627618,0.0003340057,0.00143403,0.0005289738,0.001233196,0.0002547402,0.001946379,0.0007441907,0.01588217],"category_scores_gemma":[0.0930514,0.0003086169,0.0003893071,0.002224591,0.001849673,0.0001940527,0.00102413,0.01428942,0.0001057139],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007368953,"about_ca_system_score_gemma":0.00432285,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009795533,"about_ca_topic_score_gemma":0.001013603,"domain_scores_codex":[0.9780976,0.006801805,0.001062524,0.000762812,0.01222409,0.001051143],"domain_scores_gemma":[0.9848223,0.002291836,0.0004444537,0.001041855,0.0108897,0.0005098222],"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":[0.00001012701,0.00007755429,0.00004152237,0.008713983,0.00003673188,0.00003134344,0.0004372912,2.593598e-9,4.52268e-7,0.003205119,0.8747287,0.1127171],"study_design_scores_gemma":[0.00004295937,0.0001627521,0.00009312796,0.01369456,0.00005040116,0.00001351761,0.0006805927,0.000005179073,0.000001292755,0.0007176122,0.9842725,0.000265506],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.000003833826,0.2850667,0.00001326318,0.1406567,0.02076826,0.002949537,0.0005526582,0.0001138297,0.5498752],"genre_scores_gemma":[0.00004557864,0.4956273,0.00004988104,0.0002092178,0.000907696,0.0003148048,0.0004118058,0.00002894645,0.5024048],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.2105606,"threshold_uncertainty_score":0.9999366,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"}]}