{"meta":{"query_hash":"c83ea7bf0091","filters":{"venue":"Accountability in Research"},"cohort_total":36,"direct_labels_cover":24,"predictions_cover":36,"exported":36,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/c83ea7bf0091","api":"https://metacan.xera.ac/api/v1/cohort?venue=Accountability+in+Research"},"results":[{"id":"W1557543835","doi":"10.1080/08989621.2012.650949","title":"Ethical Issues in the Translation of Social Neuroscience: A Policy Analysis of Current Guidelines for Public Dialogue in Human Research","year":2012,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Psychology of Moral and Emotional Judgment","field":"Neuroscience","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Montreal Clinical Research Institute; McGill University","funders":"","keywords":"Mainstream; Knowledge translation; Thematic analysis; Strengths and weaknesses; Public policy; Research ethics; Engineering ethics; Political science; Public relations; Social research; Psychology; Sociology; Qualitative research; Social science; Knowledge management; Computer science; Social psychology","score_opus":0.8656019701839802,"score_gpt":0.6518773040487474,"score_spread":0.21372466613523278,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1557543835","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93652135,0.00009426754,0.000012738006,0.06233983,0.000064119005,0.00067888486,0.00004549674,0.0000035000817,0.00023978058],"genre_scores_gemma":[0.99926776,0.00006083004,0.00002316847,0.00032190175,0.0001330231,0.00017188887,0.000011270777,0.000004458265,0.000005674119],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9926682,0.0035750382,0.0007389615,0.00042233762,0.0018164648,0.00077898806],"domain_scores_gemma":[0.9959737,0.0029868,0.00006584229,0.00033382603,0.0005865505,0.000053269232],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.034514133,0.00008735557,0.00029664414,0.0021723215,0.0002068589,0.000036059322,0.00089415023,0.00018108111,0.000027200498],"category_scores_gemma":[0.012669976,0.00006513062,0.00010884241,0.0073449574,0.0015663073,0.00027787403,0.00013045037,0.0012342097,0.0000017133386],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004968376,0.0060412264,0.6903933,0.00028794754,0.000010285315,0.0000015365264,0.03076039,0.00017887662,0.1510088,0.10397431,0.0022300119,0.014616461],"study_design_scores_gemma":[0.000489409,0.00012918074,0.9770913,0.000018413737,0.000004811821,4.0925397e-7,0.0012194501,0.0009563076,0.0016535067,0.015919885,0.0024455094,0.000071815455],"about_ca_topic_score_codex":0.0036257198,"about_ca_topic_score_gemma":0.004272861,"teacher_disagreement_score":0.28669798,"about_ca_system_score_codex":0.00016120067,"about_ca_system_score_gemma":0.00020084264,"threshold_uncertainty_score":0.9956467},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W1588909286","doi":"10.1080/08989620590957175","title":"Haunted Manuscripts: Ghost Authorship in the Medical Literature","year":2005,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":75,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Queen's University; Institute for Clinical Evaluative Sciences","funders":"","keywords":"Credibility; Conflict of interest; Accountability; Psychology; Neutrality; Clinical trial; Public relations; Political science; Law; Medicine","score_opus":0.8261096242099695,"score_gpt":0.7045818707680983,"score_spread":0.12152775344187117,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1588909286","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.84510934,0.0017761888,0.0000011689051,0.13659936,0.00036693906,0.0009717062,0.00005424096,0.000041997788,0.015079038],"genre_scores_gemma":[0.98139435,0.00042867474,0.000019633491,0.016520975,0.000811865,0.00029485757,0.00004629389,0.00001675305,0.00046659124],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.98718727,0.008462169,0.000761406,0.0006318362,0.001401137,0.0015561691],"domain_scores_gemma":[0.99379843,0.004660639,0.000045476525,0.000593142,0.0003236327,0.0005786509],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.03432653,0.00024264588,0.00031808342,0.00038829472,0.0003948528,0.000119092954,0.0015311678,0.0018587216,0.006836003],"category_scores_gemma":[0.0039575384,0.00017858633,0.00008229762,0.0022008982,0.00090731325,0.00039246984,0.00027718165,0.01689769,0.00052807655],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003156754,0.0052597336,0.80033565,0.00073003024,0.000057222784,0.0014329519,0.032196958,0.00010431641,0.00076040736,0.018666971,0.05388927,0.08340971],"study_design_scores_gemma":[0.002812216,0.00016358924,0.1926813,0.00024875192,0.00001361503,0.00014159478,0.00479177,0.0059198644,0.0011105471,0.010107966,0.78161603,0.0003927413],"about_ca_topic_score_codex":0.0007046929,"about_ca_topic_score_gemma":0.005496151,"teacher_disagreement_score":0.72772676,"about_ca_system_score_codex":0.0005197475,"about_ca_system_score_gemma":0.00049232005,"threshold_uncertainty_score":0.9994371},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["research_integrity","scholarly_communication"],"domain":null,"study_design":"design_other","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W1969530644","doi":"10.1080/08989621.2011.584763","title":"Who Cares about Consent Requirements for Sourcing Human Embryonic Stem Cells? Are Errors<i>In</i>the Past Really Errors<i>Of</i>the Past?","year":2011,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Biomedical Ethics and Regulation","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Remedial education; Informed consent; Embryonic stem cell; Human services; Business; Common Rule; Law; Psychology; Medicine; Political science; Biology; Alternative medicine; Pathology","score_opus":0.2649465839690857,"score_gpt":0.4353178113501529,"score_spread":0.1703712273810672,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1969530644","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98685086,0.00023564772,0.000020870004,0.009032589,0.0002690495,0.0023778612,0.000024075634,0.000015553967,0.0011734698],"genre_scores_gemma":[0.99865717,0.00004920908,0.000062217485,0.00038492554,0.00019020158,0.00020212219,0.00001286752,0.00002766761,0.00041361363],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9953022,0.0008657092,0.000809503,0.0005475718,0.0017241669,0.0007508242],"domain_scores_gemma":[0.99687433,0.0009010853,0.00022147308,0.0011318792,0.0007392566,0.0001319661],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009488163,0.00020424894,0.00044846916,0.00027153024,0.0003590496,0.000042288975,0.0007028243,0.00038197776,0.00006462917],"category_scores_gemma":[0.0002830358,0.00012522958,0.00017860119,0.00092550815,0.0020599705,0.00010812231,0.000269523,0.0013107249,0.000006120967],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009356119,0.0014523127,0.94921297,0.0032853372,0.0000580881,0.000021319265,0.020648181,0.00001282347,0.014686531,0.0021621028,0.006030675,0.0014940386],"study_design_scores_gemma":[0.002538049,0.0005191535,0.96577954,0.0015559008,0.00004743782,0.000008148048,0.018737799,0.0002862707,0.0033045118,0.0038993466,0.0031012115,0.00022262293],"about_ca_topic_score_codex":0.0043660626,"about_ca_topic_score_gemma":0.01185865,"teacher_disagreement_score":0.01656657,"about_ca_system_score_codex":0.00030399393,"about_ca_system_score_gemma":0.00047675218,"threshold_uncertainty_score":0.7590047},"labels":[{"model":"gemma","categories":["research_integrity"],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2013071091","doi":"10.1080/08989621003708451","title":"Dr. Cézanne and the Art of Re(peat)search: Competing Interests and Obligations in Clinical Research","year":2010,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University; Dalhousie University","funders":"","keywords":"Competition (biology); Narrative; Public relations; Political science; Psychology; Engineering ethics; Engineering","score_opus":0.8296990260947004,"score_gpt":0.7185244154404423,"score_spread":0.11117461065425804,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2013071091","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9326347,0.00021143926,0.000002392029,0.05624989,0.00009612403,0.0018586641,0.000009444419,0.000013803238,0.008923507],"genre_scores_gemma":[0.9969847,0.0014388841,0.00048476557,0.0001575439,0.00018276523,0.00014100983,0.000007882373,0.000028412991,0.0005740389],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9831901,0.007337507,0.002254005,0.0012545439,0.004529612,0.0014342079],"domain_scores_gemma":[0.61869127,0.36936566,0.00018183333,0.0038874447,0.006953075,0.00092075387],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2585603,0.00016842528,0.0007584988,0.0010098624,0.00031786662,0.00015870182,0.0008749567,0.000754837,0.0002914224],"category_scores_gemma":[0.24716386,0.00012029665,0.00010616912,0.0022823648,0.017138485,0.00020592532,0.002287151,0.021752164,0.000058019883],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024541258,0.00078468106,0.9562612,0.0005584138,0.000021952068,0.000027714414,0.0027460204,8.231088e-7,0.0011256693,0.031562336,0.00043584633,0.0040211803],"study_design_scores_gemma":[0.004636155,0.00053863827,0.89319146,0.00069629005,0.000005721349,0.000015276228,0.008336089,0.0013361654,0.00033946804,0.08996605,0.0008293798,0.000109318076],"about_ca_topic_score_codex":0.0056360057,"about_ca_topic_score_gemma":0.05844948,"teacher_disagreement_score":0.36449888,"about_ca_system_score_codex":0.00015740584,"about_ca_system_score_gemma":0.0013958695,"threshold_uncertainty_score":0.9855363},"labels":[],"label_agreement":null},{"id":"W2026647402","doi":"10.1080/08989621.2013.866045","title":"Disclosure, Consent, and the Exercise of Patient Autonomy in Surgical Innovation: A Systematic Content Analysis of the Conceptual Literature","year":2014,"lang":"en","type":"review","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Université de Montréal; Montreal Clinical Research Institute","funders":"Canadian Institutes of Health Research; Society of University Surgeons","keywords":"Informed consent; Autonomy; Psychology; Terminology; Ambiguity; Conceptual framework; Harm; Medicine; Social psychology; Alternative medicine; Sociology; Political science; Computer science; Law","score_opus":0.7046772079286926,"score_gpt":0.6201766418666188,"score_spread":0.08450056606207379,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2026647402","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05812062,0.93130326,7.6060223e-7,0.0008947825,0.00007465231,0.008803663,0.00010707834,0.0000051335896,0.00069006474],"genre_scores_gemma":[0.25743106,0.74127954,0.000012466424,0.0000328637,0.000027185124,0.00091591064,0.00005173901,0.000019476423,0.00022976141],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.98343986,0.007876333,0.0042072693,0.0007395846,0.0032301138,0.0005068188],"domain_scores_gemma":[0.8219902,0.16721357,0.0014011664,0.003774129,0.005466109,0.00015480725],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0571439,0.00031038898,0.0050706123,0.0014245171,0.00007816248,0.00006728802,0.00094130525,0.0009391525,0.00006631439],"category_scores_gemma":[0.07419674,0.00014046539,0.0007478113,0.010889276,0.0067831883,0.000057856385,0.0010231961,0.006149786,0.0000027054657],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013171103,0.0012413702,0.050945032,0.83999914,0.0017114464,0.000036918897,0.008464855,0.000004374072,4.2332724e-7,0.079172075,0.000040364586,0.017066874],"study_design_scores_gemma":[0.007494422,0.0006150439,0.012070455,0.9279769,0.005480725,0.00004947628,0.009812418,0.00041300515,0.000006622653,0.0073154066,0.02818642,0.0005790888],"about_ca_topic_score_codex":0.0012399488,"about_ca_topic_score_gemma":0.0014484946,"teacher_disagreement_score":0.19931044,"about_ca_system_score_codex":0.0005488673,"about_ca_system_score_gemma":0.0033320384,"threshold_uncertainty_score":0.99614304},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2034471287","doi":"10.1080/08989621.2011.542681","title":"Bad News about Bad News: The Disclosure of Risks to Insurability in Research Consent Processes","year":2011,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Insurability; Scope (computer science); Subject (documents); Business; Business ethics; Quality of Life Research; Actuarial science; Psychology; Political science; Public relations; Insurance policy; Medicine; Public health; Insurance law; General insurance; Computer science","score_opus":0.89119809906341,"score_gpt":0.692302295726708,"score_spread":0.19889580333670198,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2034471287","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9499691,0.0006038302,0.000008668019,0.021885447,0.00011030629,0.0050451034,0.00005560615,0.000033015018,0.022288933],"genre_scores_gemma":[0.9952917,0.0021674389,0.00038022056,0.0003552443,0.00015532465,0.0009573299,0.000009544241,0.000055329012,0.00062787073],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9781872,0.0069170794,0.0024248627,0.0019014562,0.0080977855,0.0024715853],"domain_scores_gemma":[0.7415402,0.22769794,0.00031409413,0.009450416,0.019397331,0.001600027],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10922265,0.0003408677,0.0010177731,0.0012423247,0.00034485973,0.0001224753,0.0025059634,0.00093736214,0.001156952],"category_scores_gemma":[0.37628397,0.00023463431,0.00018263276,0.0076063327,0.0062293485,0.00030867456,0.0023120204,0.01220982,0.0002837563],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0037864042,0.003068975,0.97219557,0.002618575,0.000031218562,0.000046642457,0.010975638,0.000004513621,0.00052944897,0.0017678473,0.0009961615,0.0039789933],"study_design_scores_gemma":[0.0013397286,0.0012148928,0.9115607,0.0012796111,0.000007639314,0.0000058412948,0.015121365,0.000018216058,0.0029452113,0.06510897,0.0011960746,0.0002017121],"about_ca_topic_score_codex":0.17040673,"about_ca_topic_score_gemma":0.47086865,"teacher_disagreement_score":0.30046192,"about_ca_system_score_codex":0.0012608316,"about_ca_system_score_gemma":0.010146665,"threshold_uncertainty_score":0.9997561},"labels":[],"label_agreement":null},{"id":"W2035490430","doi":"10.1080/08989620902984106","title":"Safety First: Recognizing and Managing the Risks to Child Participants in Magnetic Resonance Imaging Research","year":2009,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Radiation Dose and Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Focus (optics); Key (lock); Pediatric research; Magnetic resonance imaging; Engineering ethics; Psychology; Medicine; Business; Knowledge management; Computer science; Engineering; Computer security","score_opus":0.24041135265841806,"score_gpt":0.4912848617688953,"score_spread":0.25087350911047723,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2035490430","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.89274204,0.0050281654,0.0000044312555,0.090584606,0.000037459253,0.0014012745,0.0000032856235,0.000022013484,0.01017674],"genre_scores_gemma":[0.9973071,0.0009156065,0.00011873444,0.0012817887,0.00011996505,0.00010606407,0.0000028099603,0.000017667364,0.00013025013],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9953649,0.0008178034,0.0005555276,0.0007060807,0.0011831276,0.0013725534],"domain_scores_gemma":[0.9970185,0.0016233651,0.00002552405,0.00082214887,0.0002859189,0.00022450708],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017710192,0.00014990462,0.00029307214,0.000864588,0.0005583718,0.00020062599,0.0003724125,0.00006675813,0.0001255765],"category_scores_gemma":[0.0045556887,0.000121635734,0.000033914792,0.0026167857,0.00035701666,0.00035619896,0.00026752055,0.0018548804,0.00008237392],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00043384102,0.0001419552,0.6638254,0.00004265452,0.0000016996274,0.000047082704,0.004327424,0.000029368774,0.00010259677,0.00010956871,0.00061730994,0.3303211],"study_design_scores_gemma":[0.001009177,0.00013542078,0.9738526,0.0005536767,0.0000035484802,0.00001639106,0.004050004,0.0071894308,0.00015663811,0.0022447256,0.010669226,0.000119194796],"about_ca_topic_score_codex":0.004820253,"about_ca_topic_score_gemma":0.002834259,"teacher_disagreement_score":0.3302019,"about_ca_system_score_codex":0.0005732429,"about_ca_system_score_gemma":0.00015299219,"threshold_uncertainty_score":0.80586356},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2043925940","doi":"10.1080/08989620802388762","title":"Perspectives on Engaging the Public in the Ethics of Emerging Biotechnologies: From Salmon to Biobanks to Neuroethics","year":2008,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Neuroethics, Human Enhancement, Biomedical Innovations","field":"Neuroscience","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Concordia University","funders":"Genome British Columbia; Genome Canada","keywords":"Neuroethics; Biobank; Anticipation (artificial intelligence); Public engagement; Engineering ethics; Ethical issues; Political science; Psychology; Public relations; Neuroscience; Sociology; Biology; Bioinformatics; Computer science; Engineering","score_opus":0.5869436868712309,"score_gpt":0.5262730286507429,"score_spread":0.060670658220488005,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2043925940","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7383671,0.00002073414,0.00009399606,0.25991854,0.00012430576,0.00093221874,0.000026071999,0.00004896693,0.00046810758],"genre_scores_gemma":[0.9915324,0.00023627652,0.00015278903,0.007775509,0.000057488825,0.00020740367,0.0000014063589,0.00001982216,0.00001688967],"study_design_codex":"bench_or_experimental","study_design_gemma":"qualitative","domain_scores_codex":[0.9910876,0.0036418126,0.0006698323,0.0010496537,0.0027093594,0.0008417516],"domain_scores_gemma":[0.9726056,0.0248172,0.000095577816,0.0018536334,0.00054743496,0.00008056034],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.013533767,0.00019321645,0.00023928497,0.0011444273,0.0008760157,0.00013308368,0.0027709315,0.00029960222,0.0000652123],"category_scores_gemma":[0.123480834,0.00013151161,0.00005176994,0.007598115,0.0030810146,0.00022815402,0.0009256631,0.008267077,0.000069289206],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031971012,0.0024128663,0.017232576,0.000108500324,0.00001030593,0.00011947245,0.38239524,0.00013088163,0.4618184,0.12714367,0.0018485541,0.0064598187],"study_design_scores_gemma":[0.0013126781,0.0018814105,0.1468458,0.000501348,0.000006824702,0.000026085148,0.3870359,0.0008854484,0.34598532,0.08556581,0.028935725,0.0010176154],"about_ca_topic_score_codex":0.0021513759,"about_ca_topic_score_gemma":0.0017797817,"teacher_disagreement_score":0.25316536,"about_ca_system_score_codex":0.0003571903,"about_ca_system_score_gemma":0.00039967254,"threshold_uncertainty_score":0.999632},"labels":[],"label_agreement":null},{"id":"W2046743472","doi":"10.1080/08989621003641165","title":"Disclosure of Unknown Harms in Magnetic Resonance Imaging Research","year":2010,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Radiation Dose and Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Canadian Institutes of Health Research; U.S. Food and Drug Administration","keywords":"Magnetic resonance imaging; Research ethics; Medicine; Ethical issues; Modality (human–computer interaction); Informed consent; Institutional review board; Medical physics; Psychology; Computer science; Engineering ethics; Radiology; Pathology; Psychiatry; Alternative medicine; Artificial intelligence; Engineering","score_opus":0.123535798155459,"score_gpt":0.4855899429385898,"score_spread":0.3620541447831308,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2046743472","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9797937,0.0015753614,8.3162e-7,0.0068786168,0.000099545214,0.0008341175,0.0000070012143,0.000015391091,0.0107954275],"genre_scores_gemma":[0.9988093,0.00012747107,0.0001760627,0.00006891114,0.00013013359,0.00010227472,0.000007771475,0.00002260528,0.0005554959],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99576956,0.0005774953,0.0005839352,0.00054910895,0.0016377385,0.0008821406],"domain_scores_gemma":[0.996697,0.0013505097,0.000032757573,0.0010156706,0.0007604128,0.00014361291],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.014295291,0.00012010421,0.00032293046,0.0011619516,0.00010011216,0.00006344214,0.00044670523,0.00011850787,0.0007917312],"category_scores_gemma":[0.0048550013,0.00010792396,0.000053858657,0.0026622133,0.0010316431,0.00036992,0.0002466021,0.0030444525,0.00007764607],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022200892,0.00040459374,0.9180909,0.00012433334,0.0000012423795,0.000032921103,0.0012729105,0.0000015401275,0.02359874,0.0005346597,0.0009232126,0.054792978],"study_design_scores_gemma":[0.0012774345,0.00009551044,0.9713286,0.00018154609,0.0000017734768,0.0000119242495,0.0014976277,0.0016317276,0.0051321886,0.0023634457,0.01638483,0.00009340316],"about_ca_topic_score_codex":0.006139844,"about_ca_topic_score_gemma":0.004095318,"teacher_disagreement_score":0.054699574,"about_ca_system_score_codex":0.0002588099,"about_ca_system_score_gemma":0.0006174463,"threshold_uncertainty_score":0.99925554},"labels":[],"label_agreement":null},{"id":"W2065302798","doi":"10.1080/08989621.2012.692971","title":"Disciplining Bioethics: Towards a Standard of Methodological Rigor in Bioethics Research","year":2012,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in medical practice","field":"Health Professions","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Hospital for Sick Children","funders":"","keywords":"Bioethics; Discipline; Engineering ethics; Rigour; Variety (cybernetics); Sociology; Research ethics; Social science; Political science; Epistemology; Law; Philosophy; Computer science; Engineering","score_opus":0.920944469478084,"score_gpt":0.7754212306248044,"score_spread":0.14552323885327956,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065302798","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9076426,0.000573866,0.00021400092,0.053923022,0.00078704226,0.0023451485,0.000057233778,0.000053757565,0.03440331],"genre_scores_gemma":[0.9812781,0.00065711234,0.01642033,0.00041305416,0.00047475882,0.0005607509,0.000010469653,0.000042741125,0.00014267213],"study_design_codex":"observational","study_design_gemma":"qualitative","domain_scores_codex":[0.8363091,0.14643559,0.0028875405,0.0012022331,0.008835919,0.0043296395],"domain_scores_gemma":[0.61138827,0.37955132,0.0003787483,0.0024165858,0.005485731,0.0007793336],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.63150376,0.00024891217,0.00096035504,0.0013780595,0.0014270898,0.000036739268,0.0015667976,0.0025388184,0.0019524895],"category_scores_gemma":[0.4610638,0.0002008342,0.00010130523,0.0045662685,0.012481272,0.0007822313,0.0031391215,0.05640023,0.00033398482],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024414593,0.00093675195,0.7338591,0.0018124573,0.000018609091,0.000013294739,0.082064405,0.0000041800927,0.0003846777,0.17392145,0.00086720707,0.0036763775],"study_design_scores_gemma":[0.001800414,0.0005744392,0.22490211,0.0013767501,0.0000087906565,0.0000022830657,0.44021863,0.00015415752,0.0006073512,0.31993932,0.010070979,0.00034478752],"about_ca_topic_score_codex":0.038060006,"about_ca_topic_score_gemma":0.028311959,"teacher_disagreement_score":0.508957,"about_ca_system_score_codex":0.0022324182,"about_ca_system_score_gemma":0.0042185565,"threshold_uncertainty_score":0.9998729},"labels":[],"label_agreement":null},{"id":"W2087645370","doi":"10.1080/08989621.2011.557302","title":"The Responsible Conduct of Bioethics Research","year":2011,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"National Institute of Environmental Health Sciences; National Institutes of Health","keywords":"Bioethics; Government (linguistics); Engineering ethics; Sociology; Research ethics; Political science; Law; Psychology; Philosophy","score_opus":0.9759870882107871,"score_gpt":0.7668027338591651,"score_spread":0.20918435435162197,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2087645370","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.90189767,0.0006567705,0.0000021768012,0.01992313,0.00013890595,0.0016440934,0.0000108197,0.000026527523,0.07569993],"genre_scores_gemma":[0.99295956,0.0022007972,0.00089310773,0.00009887592,0.000097784905,0.0001882598,0.0000030742724,0.000036552316,0.0035219637],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9792647,0.007183234,0.001520648,0.0010438843,0.008926488,0.0020610807],"domain_scores_gemma":[0.6895828,0.28149468,0.00016298634,0.006675487,0.021345666,0.0007383374],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.26051643,0.00015818859,0.0004748315,0.0009247004,0.0006410179,0.00007674303,0.001746451,0.000938252,0.0007824323],"category_scores_gemma":[0.2425724,0.00010596599,0.00014663828,0.0041500637,0.017596206,0.00017739365,0.0015306773,0.015824266,0.00033848203],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013170795,0.0026232407,0.46133718,0.0013756681,0.00009374511,0.00011919507,0.01027862,5.701753e-7,0.0097766975,0.49153316,0.003923989,0.005767145],"study_design_scores_gemma":[0.0014309083,0.00209733,0.24037473,0.0005488462,0.0000077046025,0.0000149654525,0.023567501,0.00006649611,0.04479177,0.6806973,0.0062335175,0.00016889615],"about_ca_topic_score_codex":0.01945857,"about_ca_topic_score_gemma":0.02036097,"teacher_disagreement_score":0.28968182,"about_ca_system_score_codex":0.00058386906,"about_ca_system_score_gemma":0.006799751,"threshold_uncertainty_score":0.9988308},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2113503937","doi":"10.1080/08989620802689821","title":"Cost of the NSERC Science Grant Peer Review System Exceeds the Cost of Giving Every Qualified Researcher a Baseline Grant","year":2009,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Innovation Policy and R&D","field":"Economics, Econometrics and Finance","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Lakehead University; University of Manitoba","funders":"","keywords":"Scrutiny; Baseline (sea); Grant writing; Promotion (chess); Grant funding; Government (linguistics); Political science; Work (physics); Business; Public relations; Library science; Engineering; Public administration; Computer science","score_opus":0.4562942085277187,"score_gpt":0.46371875813006747,"score_spread":0.0074245496023487845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113503937","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9065026,0.0038231516,0.00020239853,0.060478505,0.00033205588,0.0042556776,0.00053100055,0.00002286968,0.02385175],"genre_scores_gemma":[0.9979595,0.00043158577,0.000030964417,0.0006333627,0.000053867043,0.000093317605,0.0000054816633,0.000009707407,0.0007822037],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9957574,0.0005749381,0.0014637999,0.00047534722,0.0010469144,0.000681572],"domain_scores_gemma":[0.99483556,0.0011007662,0.00035438157,0.0014342008,0.0022105551,0.00006453163],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0879907,0.00013258509,0.0004943995,0.0005429188,0.00039973902,0.00008214664,0.0016202347,0.00010103137,0.00019556453],"category_scores_gemma":[0.013606127,0.00008729158,0.00012753512,0.0052017854,0.0014168487,0.00034342468,0.00030745796,0.0008394627,0.000050170536],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004117277,0.001235095,0.14664002,0.0064477716,0.000042984862,0.0000037513134,0.0068214117,0.00016078078,0.0020581349,0.8025495,0.02515132,0.008477544],"study_design_scores_gemma":[0.0013356424,0.0001947891,0.8936141,0.0027793874,0.000009007673,0.000015075038,0.0047313566,0.0055731265,0.0032705248,0.021490991,0.066520974,0.00046498442],"about_ca_topic_score_codex":0.008055931,"about_ca_topic_score_gemma":0.0006128708,"teacher_disagreement_score":0.7810585,"about_ca_system_score_codex":0.0005268497,"about_ca_system_score_gemma":0.00047300966,"threshold_uncertainty_score":0.9985495},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"observational","genre":"empirical","about_ca_system":true,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":true,"about_ca_topic":true,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2115882960","doi":"10.1080/08989620902855033","title":"Supervisor–Student Relations: Examining the Spectrum of Conflicts of Interest in Bioscience Laboratories","year":2009,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"Canadian Institutes of Health Research","keywords":"Supervisor; Conflict of interest; Dual (grammatical number); Broad spectrum; Psychology; Spectrum (functional analysis); Graduate students; Public relations; Social psychology; Political science; Accounting; Business; Law; Pedagogy","score_opus":0.8768308426194927,"score_gpt":0.6745009165671052,"score_spread":0.20232992605238753,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115882960","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98619455,0.0005930738,9.766634e-7,0.0099476315,0.00017539282,0.00061313924,0.000036769998,0.000010244577,0.0024282266],"genre_scores_gemma":[0.9990738,0.00022537558,0.000013241789,0.00054667844,0.000050223192,0.00003226334,0.000003452573,0.0000046223913,0.0000503141],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9961694,0.0018841396,0.00070287625,0.0003134594,0.0003398136,0.0005903266],"domain_scores_gemma":[0.9961827,0.0028907787,0.0000865796,0.00036794768,0.00033942598,0.00013256076],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.010616815,0.00011790685,0.000273041,0.0002613068,0.00016110991,0.000018489145,0.00072048244,0.00035751914,0.0005816169],"category_scores_gemma":[0.001640934,0.00009307867,0.000029458723,0.0019381961,0.0012250624,0.00022732155,0.00017557871,0.0028856837,0.000016228565],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032039118,0.0008136261,0.9600964,0.00008321129,0.000008802139,0.000009161991,0.007901578,0.00010581466,0.013230833,0.015197259,0.00013767765,0.002095218],"study_design_scores_gemma":[0.0007698106,0.00032229823,0.9483223,0.000087962035,0.0000063724306,0.0000022970808,0.008138695,0.00051020633,0.031884033,0.0062884283,0.0035626648,0.00010495991],"about_ca_topic_score_codex":0.0008387577,"about_ca_topic_score_gemma":0.0021159851,"teacher_disagreement_score":0.0186532,"about_ca_system_score_codex":0.00020283958,"about_ca_system_score_gemma":0.00033207968,"threshold_uncertainty_score":0.9994147},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"incentives","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"incentives","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2155569949","doi":"10.1080/08989621.2012.638595","title":"Unfinished Business: Ongoing Ethical Exceptionalism in the Oversight of Human Pluripotent Stem Cell Research in Canada","year":2012,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Biomedical Ethics and Regulation","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Exceptionalism; Research ethics; Agency (philosophy); Human research; Political science; Honor; Funding Agency; Public administration; Engineering ethics; Law; Public relations; Sociology; Politics; Social science","score_opus":0.3079548212433394,"score_gpt":0.47843168366478767,"score_spread":0.17047686242144827,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2155569949","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9604979,0.0002790559,0.0000015115601,0.03594626,0.00008421949,0.0007445067,0.0000065779595,0.0000030636945,0.0024368784],"genre_scores_gemma":[0.99930024,0.00007143336,0.000018957942,0.00031250282,0.00014607506,0.000060393086,0.000026429852,0.000010693534,0.000053243366],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9923502,0.002131932,0.0006376204,0.00032503216,0.0037323055,0.0008229078],"domain_scores_gemma":[0.99522626,0.0031803376,0.00004849929,0.0005813264,0.0008114809,0.00015208204],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.03132682,0.00010606572,0.00029624268,0.00057110295,0.00013693885,0.000023110053,0.00037613514,0.00047348806,0.00023215743],"category_scores_gemma":[0.00041535718,0.000073320436,0.000037010497,0.0027704104,0.0013064998,0.000117333395,0.00022739137,0.004404328,0.0000071613094],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004804173,0.0018314358,0.95730007,0.0016984623,0.000011079648,0.00006119526,0.013427524,0.000037763675,0.009658931,0.010154067,0.0016868827,0.0036522013],"study_design_scores_gemma":[0.0010036335,0.00006403853,0.9857295,0.00028732498,0.00000224729,0.0000038496987,0.0083848415,0.00017347655,0.00026575787,0.0022457165,0.0017688214,0.000070756774],"about_ca_topic_score_codex":0.88888556,"about_ca_topic_score_gemma":0.83005524,"teacher_disagreement_score":0.05883032,"about_ca_system_score_codex":0.0019757312,"about_ca_system_score_gemma":0.0031328069,"threshold_uncertainty_score":0.99789256},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":true,"about_ca_topic":true,"confidence":"low"},{"model":"gpt","categories":["metaresearch","sts"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":true,"about_ca_topic":true,"confidence":"high"}],"label_agreement":"split"},{"id":"W2164540493","doi":"10.1080/08989620701670179","title":"Offering Results to Research Subjects: U.S. Institutional Review Board Policy","year":2007,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Izaak Walton Killam Health Centre; Dalhousie University","funders":"IWK Health Centre; Dalhousie University","keywords":"Confidence interval; Institutional review board; Odds ratio; Odds; Medicine; Actuarial science; Psychology; Accounting; Medical education; Family medicine; Business; Internal medicine; Surgery; Logistic regression","score_opus":0.8055258204019985,"score_gpt":0.7221545411890221,"score_spread":0.08337127921297638,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164540493","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7416387,0.002866647,0.000043015767,0.12101445,0.00018859489,0.004804688,0.00005666456,0.00009167169,0.12929556],"genre_scores_gemma":[0.98181194,0.009377095,0.0013846408,0.0034949966,0.0009741264,0.00028561545,0.00004306465,0.00005164851,0.0025768492],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97849244,0.0019513017,0.002259415,0.0018657418,0.012064141,0.0033669537],"domain_scores_gemma":[0.8145508,0.14785372,0.00020084532,0.008911976,0.024448425,0.004034209],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25698984,0.00026223133,0.0007546538,0.0028004674,0.0005902298,0.00013171595,0.0014497649,0.0007514317,0.00041850068],"category_scores_gemma":[0.40883997,0.00023640071,0.00018052266,0.010246932,0.0025086433,0.0002893372,0.0020043156,0.012574249,0.0015807337],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.044259828,0.009652996,0.44317788,0.047567584,0.0002490344,0.003086124,0.0077383053,0.00005893972,0.015395562,0.18818323,0.08401228,0.15661824],"study_design_scores_gemma":[0.0042161276,0.0022058513,0.69847083,0.019168364,0.00001268419,0.00007913991,0.001975424,0.000043361037,0.0037630561,0.04526287,0.2242151,0.00058716576],"about_ca_topic_score_codex":0.024374992,"about_ca_topic_score_gemma":0.025815789,"teacher_disagreement_score":0.25529298,"about_ca_system_score_codex":0.0046722083,"about_ca_system_score_gemma":0.007715416,"threshold_uncertainty_score":0.99919665},"labels":[],"label_agreement":null},{"id":"W2294826128","doi":"10.1080/08989620802388788","title":"Biomedical Research and the Commercialization Agenda: A Review of main Considerations for Neuroscience","year":2008,"lang":"en","type":"review","venue":"Accountability in Research","topic":"Intellectual Property and Patents","field":"Business, Management and Accounting","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Fondation pour la Recherche Médicale; Stem Cell Network","keywords":"Commercialization; Context (archaeology); Engineering ethics; Normative; Bioethics; Neuroscience; Political science; Sociology; Psychology; Biology; Engineering","score_opus":0.8127094389806951,"score_gpt":0.5201771416261698,"score_spread":0.29253229735452535,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2294826128","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00003629589,0.9894469,0.000027351933,0.0027292063,0.00018402912,0.0057374286,0.00004451828,0.000012556577,0.0017817202],"genre_scores_gemma":[0.0006205922,0.99681383,0.000023123095,0.0011142192,0.00029361856,0.000921422,0.00008648949,0.000027136213,0.00009954638],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.99518645,0.0011328383,0.0010683513,0.0006009175,0.001386654,0.0006248089],"domain_scores_gemma":[0.9886612,0.008681685,0.00021820988,0.0006800308,0.0017373996,0.000021467515],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.021601787,0.00022153613,0.0010908684,0.00086751755,0.00091006147,0.00025000633,0.0007361634,0.00022822112,0.0001805997],"category_scores_gemma":[0.032924157,0.0001331157,0.00018212284,0.0031639887,0.0039004078,0.00039988873,0.0007597651,0.0011938312,0.000049193113],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028652922,0.00090389483,0.0001184706,0.45135444,0.00004300808,0.000018037308,0.0005313945,5.0302333e-7,0.0000019563838,0.035249542,0.22740187,0.28409034],"study_design_scores_gemma":[0.0003686779,0.000034102915,0.000015526712,0.020246843,0.000029452001,0.000014802985,0.000090529815,0.0002619084,1.8512488e-7,0.0067106355,0.9720991,0.00012818197],"about_ca_topic_score_codex":0.0025368105,"about_ca_topic_score_gemma":0.0003376335,"teacher_disagreement_score":0.7446973,"about_ca_system_score_codex":0.00013089714,"about_ca_system_score_gemma":0.00064766506,"threshold_uncertainty_score":0.9988104},"labels":[{"model":"gemma","categories":["sts"],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2583047471","doi":"10.1080/08989621.2017.1287567","title":"Best Practice to Order Authors in Multi/Interdisciplinary Health Sciences Research Publications","year":2017,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal","funders":"Social Sciences and Humanities Research Council; Social Sciences and Humanities Research Council of Canada; Université de Montréal","keywords":"Order (exchange); Engineering ethics; Declaration; Best practice; Accountability; Attribution; Confusion; Computer science; Political science; Psychology; Public relations; Management science; Sociology; Business; Social psychology; Law; Engineering","score_opus":0.9220851009964572,"score_gpt":0.8041078746149898,"score_spread":0.11797722638146746,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2583047471","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.49487793,0.00016828784,0.000032426728,0.46371067,0.00014435167,0.0027955768,0.000014978618,0.000031217718,0.038224548],"genre_scores_gemma":[0.9812374,0.00050185126,0.010245589,0.00068074267,0.00016002847,0.00080450135,0.000008671418,0.000036586865,0.0063246395],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9828082,0.004035822,0.0013436239,0.001906417,0.0073317336,0.0025742126],"domain_scores_gemma":[0.87530917,0.09518246,0.0003301474,0.0088287275,0.018183576,0.0021659336],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.24063195,0.000195284,0.00054918,0.0025186692,0.003158268,0.0009636302,0.0032817672,0.00056866097,0.00038890913],"category_scores_gemma":[0.47745505,0.00017388062,0.00007686738,0.005028972,0.0059076007,0.001270809,0.0067274207,0.012147867,0.0010139198],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016528885,0.010305473,0.9273467,0.00094474683,0.000027366583,0.0001012623,0.012268398,0.00004069092,0.00033328918,0.02359899,0.007547678,0.015832542],"study_design_scores_gemma":[0.0020524848,0.0032484746,0.8936264,0.0018301769,0.0000036959002,0.000025376597,0.0444391,0.0014880868,0.00010416491,0.039833933,0.013050763,0.0002973254],"about_ca_topic_score_codex":0.045821983,"about_ca_topic_score_gemma":0.13085762,"teacher_disagreement_score":0.48635945,"about_ca_system_score_codex":0.0024100074,"about_ca_system_score_gemma":0.009573539,"threshold_uncertainty_score":0.9997639},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["research_integrity","scholarly_communication"],"domain":null,"study_design":"design_other","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2740628686","doi":"10.1080/08989621.2017.1362557","title":"The Inappropriate Use of Risk-Benefit Analysis in the Risk Assessment of Experimental Trauma-Focused Research","year":2017,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Risk assessment; Risk analysis (engineering); Psychology; Medicine; Computer science; Computer security","score_opus":0.7439250363914514,"score_gpt":0.6797050059099999,"score_spread":0.06422003048145142,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2740628686","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9907122,0.0002489119,0.0000075133594,0.0036462233,0.00005110907,0.0020515672,0.000077511315,0.0000067313676,0.003198193],"genre_scores_gemma":[0.99668735,0.0021216727,0.00058052916,0.000015996651,0.000065383654,0.0003510559,0.0000085417705,0.000024980249,0.00014445926],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.978365,0.008318594,0.0017669885,0.0010059215,0.009177534,0.0013659797],"domain_scores_gemma":[0.8692165,0.1196119,0.0004821429,0.0065088123,0.0039873226,0.00019330127],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.19739775,0.00020743973,0.0008454493,0.0011005118,0.001428691,0.000466053,0.0033583518,0.00053793885,0.00017549931],"category_scores_gemma":[0.14838572,0.000120402816,0.00038220276,0.0030695414,0.00669578,0.00039036514,0.0017016026,0.011068957,0.00001214232],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014120756,0.0020213728,0.98236364,0.00012715298,0.00022630817,0.0000151342465,0.003493721,0.000057490837,0.0006207018,0.0049239337,0.000049587183,0.0046888823],"study_design_scores_gemma":[0.0016039016,0.0006422504,0.9582189,0.00014070392,0.000057143072,6.4741175e-7,0.010013112,0.0032107334,0.0024620665,0.023434196,0.00012277142,0.00009362303],"about_ca_topic_score_codex":0.11110694,"about_ca_topic_score_gemma":0.09545394,"teacher_disagreement_score":0.11129331,"about_ca_system_score_codex":0.00081383355,"about_ca_system_score_gemma":0.0011929029,"threshold_uncertainty_score":0.9998713},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2792860328","doi":"10.1080/08989621.2018.1442218","title":"Person-oriented research ethics: integrating relational and everyday ethics in research","year":2018,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":73,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Université de Montréal; Montreal Clinical Research Institute","funders":"Fonds de Recherche du Québec - Santé; Kids Brain Health Network; Institut de Recherche Clinique De Montréal","keywords":"Research ethics; Engineering ethics; Acknowledgement; Personhood; Sociology; Information ethics; Epistemology; Computer science","score_opus":0.9237284042116157,"score_gpt":0.7351510082954018,"score_spread":0.18857739591621392,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2792860328","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8826406,0.00050091685,0.00003162708,0.08068707,0.00019887825,0.0024567933,0.000020760514,0.000052733893,0.033410616],"genre_scores_gemma":[0.99121445,0.0011259047,0.0025764054,0.00052045664,0.00076466793,0.0004252143,0.000033089716,0.00008366533,0.0032561193],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9363393,0.029866306,0.0021092207,0.0031702404,0.024133706,0.0043812315],"domain_scores_gemma":[0.28311595,0.6534308,0.00017927052,0.0050423937,0.056527838,0.0017037141],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.4943303,0.00033747253,0.0007715576,0.0040167375,0.0020965005,0.00033549254,0.0012474728,0.0037194805,0.0010848717],"category_scores_gemma":[0.63835067,0.0003068309,0.00012002105,0.009627937,0.01778913,0.0006095369,0.0023715468,0.103502125,0.00061288994],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0031780882,0.0013350359,0.7938584,0.0015556874,0.00003987396,0.00015991896,0.06475119,0.0000018866988,0.0020477448,0.12847398,0.0021015888,0.0024965985],"study_design_scores_gemma":[0.004038907,0.0033357795,0.63959783,0.0048516556,0.0000073379138,0.00004821811,0.113604955,0.0038199346,0.000985225,0.21718624,0.012010195,0.0005136977],"about_ca_topic_score_codex":0.030960837,"about_ca_topic_score_gemma":0.08336213,"teacher_disagreement_score":0.65322334,"about_ca_system_score_codex":0.0034961244,"about_ca_system_score_gemma":0.012379374,"threshold_uncertainty_score":0.99993837},"labels":[],"label_agreement":null},{"id":"W2986767685","doi":"10.1080/08989621.2019.1684906","title":"Breaking barriers to ethical research: An analysis of the effectiveness of nonhuman animal research approval in Canada","year":2019,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Animal testing and alternatives","field":"Veterinary","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Mandate; Research ethics; Protocol (science); Animal testing; Public trust; Public relations; Quality (philosophy); Political science; Certification; Engineering ethics; Business; Medical education; Medicine; Law; Alternative medicine; Engineering; Biology","score_opus":0.4771556329383189,"score_gpt":0.5741450469816433,"score_spread":0.09698941404332434,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2986767685","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9960156,0.00001883823,9.943658e-7,0.0002787272,0.000044513352,0.0013410095,0.00005607493,0.0000078179,0.002236429],"genre_scores_gemma":[0.9997493,0.0000017361293,0.00004210466,0.000017310185,0.000027660053,0.000114234375,0.0000050058525,0.000023894387,0.000018758717],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9732027,0.019975541,0.0006835343,0.0009349437,0.004000929,0.0012023372],"domain_scores_gemma":[0.9769349,0.019331425,0.00007015665,0.0014688513,0.001923798,0.0002708861],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.08196381,0.00016320005,0.000630057,0.0017293879,0.00025974552,0.000058239453,0.0018562368,0.00021781045,0.00029851717],"category_scores_gemma":[0.017518751,0.00013081501,0.000108671884,0.008742091,0.0010766624,0.00018916302,0.0014114704,0.004303665,0.000010461375],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0051435186,0.00019182137,0.9518516,0.00036682282,0.00006335517,0.000022581598,0.0031823982,0.00046756206,0.037185658,0.0012922183,0.000013381623,0.0002190718],"study_design_scores_gemma":[0.00034770404,0.0013963146,0.97342277,0.0003197719,0.000006951344,0.0000018617653,0.016219473,0.0051522455,0.00225232,0.0007333879,0.000025376694,0.00012182026],"about_ca_topic_score_codex":0.9678085,"about_ca_topic_score_gemma":0.9073222,"teacher_disagreement_score":0.064445056,"about_ca_system_score_codex":0.002443665,"about_ca_system_score_gemma":0.0037203713,"threshold_uncertainty_score":0.99799347},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":true,"about_ca_topic":true,"confidence":"high"},{"model":"gpt","categories":["research_integrity","metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":true,"about_ca_topic":true,"confidence":"low"}],"label_agreement":"split"},{"id":"W3002767658","doi":"10.1080/08989621.2020.1721288","title":"The norms of authorship credit: Challenging the definition of authorship in The European Code of Conduct for Research Integrity","year":2020,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Canadian Medical Association; Dublin City University; European Commission","keywords":"Normative; Code of conduct; Research integrity; Medical journal; Code (set theory); Order (exchange); Scientific misconduct; Engineering ethics; Computer science; Political science; Psychology; Law; Public relations; Library science; Medicine; Alternative medicine; Business","score_opus":0.7205593581249751,"score_gpt":0.5510722679019152,"score_spread":0.16948709022305986,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3002767658","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.89185613,0.00034295156,0.000053627547,0.07780925,0.0001125907,0.0016486606,0.00009335683,0.000010441174,0.028072994],"genre_scores_gemma":[0.9991139,0.00034435655,0.000075176686,0.00007712214,0.00026076433,0.00008542467,0.0000075596467,0.00001102061,0.00002466194],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","domain_scores_codex":[0.96549726,0.028339505,0.0011537487,0.00049925625,0.0033516812,0.0011585776],"domain_scores_gemma":[0.96147954,0.03603966,0.00017813819,0.00057185197,0.001635848,0.00009493826],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22952725,0.00011323255,0.00029183118,0.00017975157,0.0009121173,0.00008118248,0.0026465254,0.00048632256,0.00004727293],"category_scores_gemma":[0.06607984,0.00006212987,0.00011831534,0.0020166638,0.0040113623,0.00023356048,0.00031126238,0.007337388,0.000008502868],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007275684,0.00026000995,0.03476147,0.00042938272,0.000021450669,0.000002177523,0.46823764,0.000013274575,0.00080684334,0.48708892,0.0031096144,0.0045416653],"study_design_scores_gemma":[0.0004999893,0.0003356923,0.06546278,0.00024590216,0.000008938074,7.826677e-7,0.48351732,0.0005841596,0.0013406328,0.4377931,0.010063325,0.0001473673],"about_ca_topic_score_codex":0.024719901,"about_ca_topic_score_gemma":0.023981351,"teacher_disagreement_score":0.16344742,"about_ca_system_score_codex":0.00018530079,"about_ca_system_score_gemma":0.00068030774,"threshold_uncertainty_score":0.9986991},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"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"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3122064493","doi":"10.1080/08989620801946917","title":"The Therapeutic Misconception: A Threat to Valid Parental Consent for Pediatric Neuroimaging Research","year":2008,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"National Research Council Institute for Biodiagnostics; Izaak Walton Killam Health Centre; Nova Scotia Health Authority; Dalhousie University","funders":"Canadian Institutes of Health Research; University of California, Santa Cruz","keywords":"Neuroimaging; Context (archaeology); Psychology; Informed consent; Research ethics; Neuroethics; Medicine; Psychiatry; Alternative medicine; Pathology","score_opus":0.9007893159834213,"score_gpt":0.7061770508876674,"score_spread":0.1946122650957539,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3122064493","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9400295,0.0010289208,0.00003750017,0.050616153,0.00023284425,0.0046424684,0.000018981402,0.000046901867,0.003346744],"genre_scores_gemma":[0.9909009,0.0032539566,0.00028659086,0.0006522165,0.00074319233,0.0013961026,0.000010123912,0.000053586187,0.0027033794],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9880972,0.0022611332,0.0009388209,0.0011911722,0.005447047,0.0020646183],"domain_scores_gemma":[0.85163265,0.140057,0.000056643687,0.0023258755,0.0052748243,0.0006529754],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.057630386,0.00021568197,0.0004316895,0.0006366485,0.0018241802,0.00020222519,0.0011962701,0.00034884398,0.0001858568],"category_scores_gemma":[0.06495445,0.00015435806,0.000203364,0.0026659956,0.0031694034,0.00014731674,0.00089330197,0.006273189,0.0003691415],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034344268,0.0010362351,0.973023,0.0005563976,0.00003224966,0.00009863661,0.003717144,0.000011085377,0.00088665186,0.0043835626,0.0088992035,0.00392139],"study_design_scores_gemma":[0.0039659473,0.0027507816,0.8854146,0.00025540395,0.000023199058,0.00013846632,0.012725882,0.00059732137,0.0008811107,0.07512065,0.017734665,0.00039196206],"about_ca_topic_score_codex":0.0018952874,"about_ca_topic_score_gemma":0.0029830197,"teacher_disagreement_score":0.13779587,"about_ca_system_score_codex":0.0010610875,"about_ca_system_score_gemma":0.0036896605,"threshold_uncertainty_score":0.9995434},"labels":[],"label_agreement":null},{"id":"W3175444759","doi":"10.1080/08989621.2021.1947810","title":"What difference might retractions make? An estimate of the potential epistemic cost of retractions on meta-analyses","year":2021,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Meta-analysis; Sample size determination; Statistics; Psychology; Publication bias; Epistemology; Econometrics; Confidence interval; Philosophy; Medicine; Mathematics","score_opus":0.47990707220513623,"score_gpt":0.5656053467790647,"score_spread":0.08569827457392848,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3175444759","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9727046,0.0003739653,0.000052056348,0.020913986,0.00068866427,0.00047866144,0.000100514,0.000013336971,0.004674234],"genre_scores_gemma":[0.99787116,0.00063404406,0.00010629616,0.00007215408,0.0001171989,0.000041438907,0.000012171021,0.0000064905084,0.0011390304],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9928186,0.0039499537,0.00056577584,0.00039571946,0.0018487659,0.00042121002],"domain_scores_gemma":[0.9948234,0.0027639575,0.00017361811,0.00080272864,0.0013416811,0.00009460321],"candidate_categories":["research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009877126,0.00011490348,0.00040032994,0.00016906203,0.00065366336,0.00013943981,0.0007623197,0.0005894023,0.0013676108],"category_scores_gemma":[0.005506287,0.00008181314,0.0002707014,0.0015635951,0.0013169206,0.0006118213,0.0001305622,0.0031384407,0.000014911791],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020006527,0.014529781,0.275624,0.0011832573,0.0043874444,0.000111271336,0.1400471,0.004142243,0.22248296,0.27181393,0.008045426,0.05563194],"study_design_scores_gemma":[0.00058211177,0.0001996008,0.6015941,0.00036147932,0.0010420519,0.00002159283,0.23003125,0.0015976117,0.05071883,0.103680305,0.009637693,0.00053338986],"about_ca_topic_score_codex":0.030984357,"about_ca_topic_score_gemma":0.06109786,"teacher_disagreement_score":0.32597008,"about_ca_system_score_codex":0.000311684,"about_ca_system_score_gemma":0.0008689714,"threshold_uncertainty_score":0.9995453},"labels":[],"label_agreement":null},{"id":"W4293511218","doi":"10.1080/08989621.2022.2119136","title":"Appropriate inclusion of adult research participants with intellectual disability: an in-depth review of guidelines and policy statements","year":2022,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre","funders":"","keywords":"Intellectual disability; Terminology; Inclusion (mineral); Documentation; Informed consent; Psychology; Research ethics; Medical model of disability; Population; Medical education; Applied psychology; Medicine; Psychiatry; Social psychology; Alternative medicine","score_opus":0.8001993503299486,"score_gpt":0.7145235454461236,"score_spread":0.08567580488382498,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4293511218","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9797761,0.001692349,0.000002657239,0.014742257,0.000020351217,0.002880061,0.00004980063,0.000010919966,0.0008255156],"genre_scores_gemma":[0.98233795,0.016092537,0.0004608944,0.00037770777,0.00005039509,0.00053387135,0.000030737265,0.000034665478,0.00008126918],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9794958,0.0065769125,0.0024614097,0.0011718185,0.008889792,0.0014042569],"domain_scores_gemma":[0.93221265,0.047017753,0.00036089725,0.0041689146,0.015411261,0.00082851545],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.11843317,0.00019039336,0.00091587316,0.00094381167,0.00036063185,0.000020915497,0.0009627918,0.00019757159,0.0005509679],"category_scores_gemma":[0.19734819,0.00014971472,0.00006493749,0.0051391604,0.0038002685,0.00026661484,0.0069286493,0.0054560346,0.0000045428314],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008357234,0.006066829,0.9130301,0.028536709,0.00005276861,0.00003420332,0.012986919,0.000028716828,0.0020357848,0.0023109242,0.00022041403,0.026339421],"study_design_scores_gemma":[0.015020969,0.029349446,0.5630691,0.04516428,0.000075681535,0.000043728258,0.08576573,0.0056049004,0.007339715,0.2451641,0.002411463,0.000990877],"about_ca_topic_score_codex":0.03269121,"about_ca_topic_score_gemma":0.02032034,"teacher_disagreement_score":0.34996095,"about_ca_system_score_codex":0.0013131363,"about_ca_system_score_gemma":0.0036023892,"threshold_uncertainty_score":0.99891084},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4308656581","doi":"10.1080/08989621.2022.2145470","title":"“Researchers’ perceptions and awareness of predatory publishing: A survey”","year":2022,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Academic Publishing and Open Access","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Science North","funders":"","keywords":"Publishing; Reputation; Commission; Transparency (behavior); Public relations; Research ethics; Library science; Political science; Sociology; Psychology; Law","score_opus":0.6390324670345174,"score_gpt":0.5817224741848028,"score_spread":0.05730999284971461,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4308656581","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99096155,0.00024291265,0.000024954013,0.0036263694,0.00015486187,0.00048168364,0.00040983644,0.000016852353,0.004080995],"genre_scores_gemma":[0.9985186,0.00002661663,0.000042187137,0.000095542164,0.000039714814,0.0001805831,0.00003624514,0.000012899959,0.0010476493],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9747624,0.012419309,0.0012248395,0.001270179,0.009388166,0.00093514787],"domain_scores_gemma":[0.96610594,0.02795331,0.00024126883,0.0021313087,0.003192789,0.00037540807],"candidate_categories":["metaresearch","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19966447,0.00012494891,0.00039582912,0.0015415681,0.00082544616,0.0025872965,0.0058546453,0.00018286577,0.0038363454],"category_scores_gemma":[0.111970395,0.000108258195,0.00006234799,0.0066009215,0.0011167696,0.005220801,0.0061822054,0.003363967,0.000022783455],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013919466,0.00019049278,0.9590752,0.00001391068,0.000004794695,0.0000021715553,0.0040327185,0.000082563645,0.00007269611,0.00032149753,0.028353063,0.007711703],"study_design_scores_gemma":[0.00032212815,0.00006107522,0.93681973,0.000006675528,7.507858e-7,0.000004088454,0.035565693,0.0010790084,0.000014835587,0.016834095,0.009188738,0.00010317985],"about_ca_topic_score_codex":0.0448349,"about_ca_topic_score_gemma":0.007985401,"teacher_disagreement_score":0.08769407,"about_ca_system_score_codex":0.0004877675,"about_ca_system_score_gemma":0.0014568011,"threshold_uncertainty_score":0.9995242},"labels":[],"label_agreement":null},{"id":"W4320710261","doi":"10.1080/08989621.2023.2179919","title":"Can ChatGPT be trusted to provide reliable estimates?","year":2023,"lang":"en","type":"letter","venue":"Accountability in Research","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Thompson Rivers University","funders":"","keywords":"Direct Anonymous Attestation; Computer security; Internet privacy; Computer science; Scientific misconduct; Trusted Computing; Medicine","score_opus":0.25348492229708014,"score_gpt":0.44915107833577544,"score_spread":0.1956661560386953,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4320710261","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011622091,0.00003833435,0.00077556807,0.98088056,0.00093584607,0.0029494304,0.000058883274,0.0005706608,0.0021686046],"genre_scores_gemma":[0.040229656,0.000111558315,0.014381901,0.9107402,0.0039546546,0.006423162,0.0003753536,0.0004902538,0.023293281],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98970926,0.0007376699,0.0010391623,0.0023780651,0.0032403101,0.0028955436],"domain_scores_gemma":[0.990083,0.003800906,0.0001268163,0.0042900327,0.001413774,0.0002855085],"candidate_categories":["metaepi_narrow","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009342914,0.00052732934,0.00075349887,0.0018584328,0.00044109623,0.0013843076,0.0064173047,0.0010551644,0.00018457054],"category_scores_gemma":[0.006289001,0.00053592236,0.00014589034,0.006809411,0.00040913562,0.0009048234,0.0031179374,0.00566786,0.0027527756],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000027384378,0.00009874188,0.001560197,0.0004169177,0.000016399745,0.00070996705,0.00288145,0.00030402114,0.000112462636,0.00074519176,0.9902774,0.0028498722],"study_design_scores_gemma":[0.00014520543,0.00035344914,0.0008600486,0.00040855244,0.000005800862,0.000013959849,0.0005753678,0.015451848,0.004539295,0.037045702,0.9396437,0.00095707114],"about_ca_topic_score_codex":0.055247303,"about_ca_topic_score_gemma":0.014672861,"teacher_disagreement_score":0.0701404,"about_ca_system_score_codex":0.0021552693,"about_ca_system_score_gemma":0.0015869522,"threshold_uncertainty_score":0.99970925},"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":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4383186065","doi":"10.1080/08989621.2023.2233419","title":"Deploying an ethics needs assessment to inform a navigational tool for research compliance pathways at a provincial Canadian health authority","year":2023,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Simon Fraser University; Provincial Health Services Authority; Centre Hospitalier Universitaire Sainte-Justine; University of British Columbia","funders":"","keywords":"CLARITY; Variety (cybernetics); Health care; Compliance (psychology); Knowledge management; Business; Service (business); Public relations; Process management; Computer science; Political science; Psychology","score_opus":0.9098084295788003,"score_gpt":0.7336634680487633,"score_spread":0.176144961530037,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4383186065","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8837421,0.00002286721,0.00007564894,0.106744446,0.00019013474,0.007156204,0.0005362011,0.00012002368,0.0014123842],"genre_scores_gemma":[0.9897767,0.00008557623,0.0035934877,0.0020714905,0.00043898317,0.0025282796,0.0004910101,0.00007229698,0.0009421875],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98230684,0.0024397087,0.0013757874,0.001265142,0.009119248,0.0034932739],"domain_scores_gemma":[0.9311243,0.05589784,0.000087334876,0.0020585547,0.00841249,0.0024194678],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21809953,0.00024880655,0.0006238113,0.0022735249,0.0029547175,0.0003194146,0.0012499768,0.0010603971,0.00014724022],"category_scores_gemma":[0.109956555,0.000249705,0.00013394355,0.007590806,0.0012137099,0.00039717392,0.0013535841,0.014241235,0.0004900284],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.0034074502,0.0014195313,0.7972743,0.0073184683,0.00006587076,0.00011151757,0.03596833,0.00016894953,0.0013488552,0.1139861,0.013427709,0.025502924],"study_design_scores_gemma":[0.0019378928,0.003188736,0.87054205,0.001208939,0.0000039527595,0.000010735322,0.010068694,0.0080542555,0.00011201718,0.08699285,0.017502805,0.00037707496],"about_ca_topic_score_codex":0.3090714,"about_ca_topic_score_gemma":0.83789736,"teacher_disagreement_score":0.528826,"about_ca_system_score_codex":0.014274697,"about_ca_system_score_gemma":0.049097776,"threshold_uncertainty_score":0.9999955},"labels":[],"label_agreement":null},{"id":"W4400620696","doi":"10.1080/08989621.2024.2377723","title":"The punishment intensity for research misconduct and its related factors: An exploratory study on hospitals in Mainland China","year":2024,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Mainland China; Misconduct; Exploratory research; China; Scientific misconduct; Punishment (psychology); Psychology; Criminology; Applied psychology; Social psychology; Political science; Law; Sociology; Medicine; Social science; Alternative medicine","score_opus":0.2987618849254758,"score_gpt":0.5154437601731929,"score_spread":0.21668187524771715,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400620696","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.984134,0.0007023267,4.262744e-7,0.009965893,0.00030626246,0.0030662657,0.000024884996,0.000036294274,0.0017636182],"genre_scores_gemma":[0.99879146,0.0003693768,0.000002985292,0.000022126103,0.000119328404,0.00030059993,0.0000070394235,0.000014728759,0.00037237324],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.99078655,0.0053157522,0.00043395066,0.0007378984,0.0016763491,0.0010494826],"domain_scores_gemma":[0.9903879,0.008395216,0.000023552122,0.00041675474,0.0006026813,0.0001738734],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09036314,0.00014682287,0.00024388851,0.00043882037,0.0013702399,0.0005955793,0.0007673922,0.00056425435,0.000049238675],"category_scores_gemma":[0.0106707625,0.000102672784,0.00004084499,0.0013706548,0.0009988857,0.0007341721,0.0002717329,0.0056827236,0.000028696564],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005758767,0.00083586754,0.33092114,0.000107668304,0.000042072563,0.000040757557,0.6350852,0.000003716611,0.00017031333,0.025296727,0.0036114932,0.0033091283],"study_design_scores_gemma":[0.000435563,0.00087097473,0.27742106,0.00017236017,0.000003641403,5.748825e-7,0.6468742,0.00035000953,0.00019087842,0.065742746,0.0077384715,0.00019951323],"about_ca_topic_score_codex":0.0155771235,"about_ca_topic_score_gemma":0.035015758,"teacher_disagreement_score":0.07969238,"about_ca_system_score_codex":0.00093830645,"about_ca_system_score_gemma":0.00066457846,"threshold_uncertainty_score":0.99992985},"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":["research_integrity"],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4402602478","doi":"10.1080/08989621.2024.2398104","title":"Status of animal experimentation in nutrition and dietetic research: Policies of 100 leading journals and new approach methodologies","year":2024,"lang":"en","type":"review","venue":"Accountability in Research","topic":"Animal testing and alternatives","field":"Veterinary","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Engineering ethics; Research integrity; Management science; Political science; Psychology; Engineering","score_opus":0.957616822100931,"score_gpt":0.7360190920615394,"score_spread":0.22159773003939165,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402602478","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.22126901,0.77721936,0.000004102692,0.000031987653,0.000019227658,0.0008683448,0.000038020986,0.000011602919,0.0005383449],"genre_scores_gemma":[0.095050745,0.90075874,0.003870632,8.3428904e-7,0.00007323844,0.00015122887,0.000011064743,0.00002898574,0.000054554268],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9923046,0.004475323,0.0010014782,0.00063683925,0.0008917474,0.0006900052],"domain_scores_gemma":[0.9917518,0.0073656384,0.00018043366,0.00031005812,0.00028731217,0.000104761544],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014082122,0.00023558077,0.0012344529,0.0019061392,0.000078218145,0.00014306171,0.00031672622,0.00023588425,0.000029019247],"category_scores_gemma":[0.005933596,0.00019228419,0.00009225657,0.0015317022,0.001055952,0.00027465602,0.0006827588,0.0016634811,0.000004571959],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001883376,0.0012485266,0.012627241,0.26030222,0.00023219544,0.000060268525,0.03602656,9.321652e-7,0.0067060236,0.0018865106,0.00055103854,0.6784751],"study_design_scores_gemma":[0.004753009,0.022229433,0.06761631,0.38842365,0.0006431666,0.0006655887,0.34278175,0.00067093887,0.0013975714,0.08636139,0.081490375,0.0029667742],"about_ca_topic_score_codex":0.009484913,"about_ca_topic_score_gemma":0.00008082065,"teacher_disagreement_score":0.6755083,"about_ca_system_score_codex":0.00041276403,"about_ca_system_score_gemma":0.0004039494,"threshold_uncertainty_score":0.997111},"labels":[{"model":"gpt","categories":["metaresearch","scholarly_communication"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","scholarly_communication"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W4404044359","doi":"10.1080/08989621.2024.2420812","title":"Allowing AI co–authors is a disregard for humanization","year":2024,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Ethics and Social Impacts of AI","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Engineering ethics; Psychology; Engineering","score_opus":0.4544497266217594,"score_gpt":0.6545098781449135,"score_spread":0.20006015152315404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404044359","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8423279,0.00032999032,0.00021255438,0.116151586,0.0006338167,0.001503305,0.000052665164,0.00014440737,0.038643803],"genre_scores_gemma":[0.9967706,0.00013618016,0.00006350981,0.0007088697,0.00040550902,0.00008575057,0.000012172462,0.000017545572,0.0017998294],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.9970919,0.00053257105,0.00023762934,0.00037291655,0.0011147122,0.00065027206],"domain_scores_gemma":[0.99661297,0.0022047353,0.000013595933,0.00021637989,0.000825228,0.0001271167],"candidate_categories":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.018030632,0.00008045858,0.00014023585,0.00021953338,0.0009926517,0.0012293308,0.00038093564,0.0002887353,0.00037325965],"category_scores_gemma":[0.0061199665,0.000080038786,0.0000873797,0.0009563943,0.0006342429,0.0009314907,0.00006906225,0.0008262699,0.00007134994],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000078984514,0.0002471877,0.047095437,0.0004536104,0.00004584519,0.000009277869,0.41065767,0.000008604331,0.00031410943,0.45517987,0.07016202,0.015747381],"study_design_scores_gemma":[0.00025862228,0.0001318827,0.008824402,0.00019149136,0.000008631616,1.5708105e-7,0.047292992,0.0010328564,0.00018538104,0.45813972,0.48365998,0.00027389685],"about_ca_topic_score_codex":0.019586815,"about_ca_topic_score_gemma":0.023664026,"teacher_disagreement_score":0.41349795,"about_ca_system_score_codex":0.0006300006,"about_ca_system_score_gemma":0.00094955426,"threshold_uncertainty_score":0.9998075},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"evaluation","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","scholarly_communication","research_integrity"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4406034614","doi":"10.1080/08989621.2024.2449041","title":"Using mixed methods research to study research integrity: Current status, issues, and guidelines","year":2025,"lang":"en","type":"review","venue":"Accountability in Research","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Research integrity; Current (fluid); Academic integrity; Engineering ethics; Scientific misconduct; Research ethics; Psychology; Medicine; Engineering; Alternative medicine; Pathology","score_opus":0.9748763335874832,"score_gpt":0.8580912241453093,"score_spread":0.1167851094421739,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406034614","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004535987,0.9673771,0.000049028444,0.00419296,0.0007451403,0.021666184,0.00014691324,0.00008503023,0.0012016423],"genre_scores_gemma":[0.00017703886,0.9682961,0.024693133,0.000057741836,0.0009682867,0.0033927865,0.00007523852,0.00021263931,0.00212704],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.79093754,0.14492597,0.009847914,0.009836591,0.03379824,0.010653735],"domain_scores_gemma":[0.4754586,0.44181016,0.00038451067,0.012875435,0.06577154,0.0036997355],"candidate_categories":["metaresearch","metaepi_narrow","bibliometrics","sts","scholarly_communication","open_science","research_integrity"],"consensus_categories":["metaresearch","bibliometrics","sts","research_integrity"],"category_scores_codex":[0.5591915,0.0010887673,0.00583178,0.01246134,0.001533718,0.0010854983,0.003991198,0.003156809,0.0004718747],"category_scores_gemma":[0.62543166,0.0008833486,0.0005887072,0.023805138,0.004584848,0.00031808185,0.01768442,0.07668133,0.0003126059],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.00049582345,0.003115605,0.004739342,0.0671479,0.00021758114,0.00013064814,0.0012054053,4.0685677e-7,0.000004907704,0.0010911679,0.005797877,0.91605335],"study_design_scores_gemma":[0.0011213553,0.0017450729,0.0010614743,0.0829601,0.00018539568,0.000021081492,0.010350967,0.00007323145,0.000009676704,0.024194472,0.8776367,0.0006405057],"about_ca_topic_score_codex":0.03698132,"about_ca_topic_score_gemma":0.011099815,"teacher_disagreement_score":0.91541284,"about_ca_system_score_codex":0.0075080716,"about_ca_system_score_gemma":0.030262437,"threshold_uncertainty_score":0.9999515},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4406183280","doi":"10.1080/08989621.2025.2450451","title":"Publisher and journal reciprocity for peer review: Not so much","year":2025,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"","keywords":"Peer review; Publication; Incentive; Publishing; Reciprocity (cultural anthropology); Public relations; Revenue; Psychology; Political science; Business; Economics; Social psychology; Accounting; Law","score_opus":0.9255360272906197,"score_gpt":0.7014977101768047,"score_spread":0.22403831711381494,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406183280","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.54533553,0.028644674,0.0035213951,0.3351773,0.0015582094,0.009134911,0.00004998569,0.000009674653,0.07656832],"genre_scores_gemma":[0.7592888,0.003261808,0.0072193397,0.00926051,0.00032141936,0.0011446916,0.000016797196,0.000028121216,0.21945854],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9478865,0.021174002,0.010599201,0.0017815189,0.01766499,0.00089375366],"domain_scores_gemma":[0.83149,0.082300626,0.0034954993,0.012757758,0.06926929,0.0006868272],"candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.65358,0.00019879629,0.0023716302,0.00086502545,0.00051550264,0.004670162,0.002603834,0.00015267663,0.00543487],"category_scores_gemma":[0.5238494,0.00010100873,0.00078749005,0.003610549,0.00033169927,0.00088511535,0.00050142413,0.0009547751,0.00034818525],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029275709,0.000064700595,0.068405576,0.00050557446,0.000034748628,8.1475974e-7,0.00020735801,5.096453e-7,0.00001126361,0.00093569205,0.8870756,0.0427289],"study_design_scores_gemma":[0.0002621485,0.000023986673,0.05763252,0.00027196715,0.000032331438,0.0000053123485,0.0010944877,0.00033658222,0.0000070814267,0.08608914,0.85414916,0.00009524411],"about_ca_topic_score_codex":0.00015723908,"about_ca_topic_score_gemma":0.0005673433,"teacher_disagreement_score":0.3259168,"about_ca_system_score_codex":0.00021229615,"about_ca_system_score_gemma":0.0005811258,"threshold_uncertainty_score":0.9963631},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4408464518","doi":"10.1080/08989621.2025.2470860","title":"Mapping nine decades of research integrity studies (1935–2024): A scientometric analysis","year":2025,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Research integrity; Engineering ethics; Engineering","score_opus":0.45889575399146626,"score_gpt":0.6163961553527757,"score_spread":0.15750040136130944,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408464518","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9309634,0.0031846208,0.00009269847,0.010060912,0.00032566427,0.0007892075,0.000021612881,0.00003569687,0.054526176],"genre_scores_gemma":[0.99155396,0.0018434364,0.0003592763,0.00006802715,0.00010994828,0.00009850514,0.0000044907783,0.0000054416996,0.005956931],"study_design_codex":"observational","study_design_gemma":"qualitative","domain_scores_codex":[0.9862051,0.0063991817,0.0008806075,0.0008367576,0.0042427294,0.0014356134],"domain_scores_gemma":[0.9722717,0.020591484,0.00009720324,0.0008019654,0.006070072,0.00016758699],"candidate_categories":["metaresearch","bibliometrics","sts","research_integrity"],"consensus_categories":["metaresearch","bibliometrics","sts"],"category_scores_codex":[0.14753808,0.00015371991,0.00066008145,0.015338601,0.0013358498,0.00018553842,0.001844001,0.0008792377,0.00067962724],"category_scores_gemma":[0.08271674,0.0001392617,0.00019403668,0.094564624,0.0059930077,0.00047634658,0.0009819049,0.007628629,0.00004548741],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010213554,0.00032096947,0.90093744,0.00028304817,0.00033935232,0.0000046821056,0.050807208,0.000017360964,0.00012958131,0.03063578,0.0109924935,0.0054299333],"study_design_scores_gemma":[0.000434239,0.000079517886,0.31847978,0.00048529985,0.000054262822,2.9210497e-7,0.5517645,0.0005354264,0.0006379452,0.083461,0.043808717,0.00025904976],"about_ca_topic_score_codex":0.11227362,"about_ca_topic_score_gemma":0.058114495,"teacher_disagreement_score":0.58245766,"about_ca_system_score_codex":0.0019670653,"about_ca_system_score_gemma":0.0018358424,"threshold_uncertainty_score":0.9999643},"labels":[],"label_agreement":null},{"id":"W4413041880","doi":"10.1080/08989621.2025.2542197","title":"AI disclosure, moral shame, and the punishment of honesty","year":2025,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Honesty; Shame; Transparency (behavior); Psychology; Punishment (psychology); Constructive; Perspective (graphical); Narrative; Social psychology; Process (computing); Law; Political science; Computer science","score_opus":0.10953065969817855,"score_gpt":0.4822537312601409,"score_spread":0.3727230715619624,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413041880","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8547173,0.00042317083,0.00001791873,0.085683495,0.0001258064,0.00068318314,0.000010777917,0.000010816596,0.058327522],"genre_scores_gemma":[0.9980212,0.0003927811,0.000024775216,0.00052209676,0.000048862294,0.000047599522,0.0000015137946,0.0000021176643,0.0009390851],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9964663,0.0018488104,0.00027657548,0.00022181566,0.0008087014,0.0003778391],"domain_scores_gemma":[0.9962835,0.003028119,0.000029597868,0.00027621206,0.00033962645,0.00004294869],"candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.024393415,0.0000632701,0.00018860218,0.0001225648,0.00041188288,0.00007876083,0.0005385655,0.00033763202,0.00013179777],"category_scores_gemma":[0.0035643624,0.00004240441,0.000040177347,0.0008698033,0.003631497,0.00020259667,0.00025295527,0.0019455338,0.0000065153813],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002355228,0.00008648157,0.44663644,0.000049170856,0.000010335985,6.7999406e-7,0.017381182,0.0000018326642,0.000010601006,0.52684516,0.0051758303,0.0035667978],"study_design_scores_gemma":[0.0011545091,0.000028909104,0.30528906,0.00011134792,0.000009014176,3.4099932e-7,0.04927852,0.00025375123,0.00015929305,0.58401746,0.05958318,0.00011459509],"about_ca_topic_score_codex":0.095867254,"about_ca_topic_score_gemma":0.022395257,"teacher_disagreement_score":0.14330386,"about_ca_system_score_codex":0.0001622598,"about_ca_system_score_gemma":0.0004824844,"threshold_uncertainty_score":0.99908006},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4414516464","doi":"10.1080/08989621.2025.2560886","title":"A classroom exercise for improving mentor/mentee relationships","year":2025,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Mentoring and Academic Development","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Columbia College","funders":"","keywords":"Physical activity; Component (thermodynamics); Research ethics","score_opus":0.24161579935727323,"score_gpt":0.4966661895393897,"score_spread":0.25505039018211645,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414516464","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94869155,0.00078814995,0.00056853023,0.0030780449,0.002528299,0.0023962434,0.000017989558,0.00008212767,0.04184906],"genre_scores_gemma":[0.97423726,0.000018817933,0.00075211696,0.00006926127,0.00007942687,0.0022071858,0.000016871905,0.000014255011,0.02260481],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9971579,0.0004290384,0.00051063916,0.00060883444,0.00045311265,0.0008405071],"domain_scores_gemma":[0.99693114,0.0020994009,0.00004466009,0.00060703786,0.00023335767,0.00008442209],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012079605,0.0001291428,0.00020854456,0.00047388152,0.00031428377,0.00007118831,0.00047463446,0.00020666452,0.00037106295],"category_scores_gemma":[0.0016825387,0.0001282667,0.000053131585,0.00081951247,0.00012176795,0.00012539593,0.00020918138,0.0010347301,0.00015211833],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010052791,0.00090936985,0.77465504,0.0005598544,0.00006432498,0.000005269539,0.010351988,0.000007143447,0.0006942179,0.032206845,0.07711222,0.10242842],"study_design_scores_gemma":[0.0048120464,0.00015169899,0.5893614,0.00058355316,0.000034136636,0.000001673145,0.030796895,0.00045507884,0.001111546,0.053249367,0.3188963,0.00054628024],"about_ca_topic_score_codex":0.0011638461,"about_ca_topic_score_gemma":0.00018644943,"teacher_disagreement_score":0.2417841,"about_ca_system_score_codex":0.00080402114,"about_ca_system_score_gemma":0.00024376645,"threshold_uncertainty_score":0.52305645},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4417358357","doi":"10.1080/08989621.2025.2600404","title":"Commercial funding of randomized controlled trials of weight-loss interventions using dietary supplements: A rapid review","year":2025,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Pharmacology and Obesity Treatment","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Randomized controlled trial; Psychological intervention; Weight loss; MEDLINE; Clinical trial; Dietary supplement","score_opus":0.4723859013700117,"score_gpt":0.6036886997669948,"score_spread":0.1313027983969831,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417358357","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94195396,0.043007933,0.00007323216,0.0046273367,0.00022356205,0.008993104,0.00008419179,0.000009937408,0.0010267592],"genre_scores_gemma":[0.9838897,0.014846035,0.00025221522,0.00023700402,0.000036177047,0.0005946477,0.00005440063,0.0000060417633,0.00008377802],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98732126,0.0082892645,0.0031270252,0.00029219818,0.0005949405,0.00037530414],"domain_scores_gemma":[0.983124,0.014903876,0.00051822077,0.00052538386,0.0008520914,0.00007641369],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06699459,0.0001537015,0.0049938746,0.0006741619,0.00009544673,0.0000065986437,0.00021164253,0.000112880516,0.002586523],"category_scores_gemma":[0.018412255,0.00010408385,0.0015306269,0.00076400844,0.0006522695,0.00009588421,0.00019789315,0.00047235782,0.0000032726136],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.62222743,0.010393994,0.30791825,0.030645149,0.0075984243,0.000036755708,0.00086703186,0.0000051305137,0.0068173897,0.0009882096,0.0055379057,0.0069643664],"study_design_scores_gemma":[0.91403633,0.0005413575,0.037341967,0.028027501,0.0066534565,0.0000064634505,0.00068299135,0.0004138038,0.008582266,0.002871095,0.0006554518,0.00018731462],"about_ca_topic_score_codex":0.0012579589,"about_ca_topic_score_gemma":0.0001620224,"teacher_disagreement_score":0.29180893,"about_ca_system_score_codex":0.00034185633,"about_ca_system_score_gemma":0.000480482,"threshold_uncertainty_score":0.9983252},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow"],"domain":"incentives","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"}]}