{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":36,"total_is_capped":false,"direct_labels_cover":24,"predictions_cover":36,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"c83ea7bf0091","filters":{"venue":"Accountability in Research"}},"results":[{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8261096242099695,"gpt":0.7045818707680983,"spread":0.1215277534418712,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.03432653,0.0002426459,0.0003180834,0.0003882947,0.0003948528,0.000119093,0.001531168,0.001858722,0.006836003],"category_scores_gemma":[0.003957538,0.0001785863,0.00008229762,0.002200898,0.0009073133,0.0003924698,0.0002771817,0.01689769,0.0005280765],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005197475,"about_ca_system_score_gemma":0.00049232,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007046929,"about_ca_topic_score_gemma":0.005496151,"domain_scores_codex":[0.9871873,0.008462169,0.000761406,0.0006318362,0.001401137,0.001556169],"domain_scores_gemma":[0.9937984,0.004660639,0.00004547652,0.000593142,0.0003236327,0.0005786509],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003156754,0.005259734,0.8003356,0.0007300302,0.00005722278,0.001432952,0.03219696,0.0001043164,0.0007604074,0.01866697,0.05388927,0.08340971],"study_design_scores_gemma":[0.002812216,0.0001635892,0.1926813,0.0002487519,0.00001361503,0.0001415948,0.00479177,0.005919864,0.001110547,0.01010797,0.781616,0.0003927413],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8451093,0.001776189,0.000001168905,0.1365994,0.0003669391,0.0009717062,0.00005424096,0.00004199779,0.01507904],"genre_scores_gemma":[0.9813944,0.0004286747,0.00001963349,0.01652098,0.000811865,0.0002948576,0.00004629389,0.00001675305,0.0004665912],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7277268,"threshold_uncertainty_score":0.9994371,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9237284042116157,"gpt":0.7351510082954018,"spread":0.1885773959162139,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.4943303,0.0003374725,0.0007715576,0.004016737,0.0020965,0.0003354925,0.001247473,0.00371948,0.001084872],"category_scores_gemma":[0.6383507,0.0003068309,0.0001200211,0.009627937,0.01778913,0.0006095369,0.002371547,0.1035021,0.0006128899],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003496124,"about_ca_system_score_gemma":0.01237937,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03096084,"about_ca_topic_score_gemma":0.08336213,"domain_scores_codex":[0.9363393,0.02986631,0.002109221,0.00317024,0.02413371,0.004381231],"domain_scores_gemma":[0.283116,0.6534308,0.0001792705,0.005042394,0.05652784,0.001703714],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003178088,0.001335036,0.7938584,0.001555687,0.00003987396,0.000159919,0.06475119,0.000001886699,0.002047745,0.128474,0.002101589,0.002496599],"study_design_scores_gemma":[0.004038907,0.00333578,0.6395978,0.004851656,0.000007337914,0.00004821811,0.113605,0.003819935,0.000985225,0.2171862,0.0120102,0.0005136977],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8826406,0.0005009169,0.00003162708,0.08068707,0.0001988782,0.002456793,0.00002076051,0.00005273389,0.03341062],"genre_scores_gemma":[0.9912145,0.001125905,0.002576405,0.0005204566,0.0007646679,0.0004252143,0.00003308972,0.00008366533,0.003256119],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6532233,"threshold_uncertainty_score":0.9999384,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.4562942085277187,"gpt":0.4637187581300675,"spread":0.007424549602348784,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0879907,0.0001325851,0.0004943995,0.0005429188,0.000399739,0.00008214664,0.001620235,0.0001010314,0.0001955645],"category_scores_gemma":[0.01360613,0.00008729158,0.0001275351,0.005201785,0.001416849,0.0003434247,0.000307458,0.0008394627,0.00005017054],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005268497,"about_ca_system_score_gemma":0.0004730097,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008055931,"about_ca_topic_score_gemma":0.0006128708,"domain_scores_codex":[0.9957574,0.0005749381,0.0014638,0.0004753472,0.001046914,0.000681572],"domain_scores_gemma":[0.9948356,0.001100766,0.0003543816,0.001434201,0.002210555,0.00006453163],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0004117277,0.001235095,0.14664,0.006447772,0.00004298486,0.000003751313,0.006821412,0.0001607808,0.002058135,0.8025495,0.02515132,0.008477544],"study_design_scores_gemma":[0.001335642,0.0001947891,0.8936141,0.002779387,0.000009007673,0.00001507504,0.004731357,0.005573126,0.003270525,0.02149099,0.06652097,0.0004649844],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9065026,0.003823152,0.0002023985,0.0604785,0.0003320559,0.004255678,0.0005310006,0.00002286968,0.02385175],"genre_scores_gemma":[0.9979595,0.0004315858,0.00003096442,0.0006333627,0.00005386704,0.00009331761,0.000005481663,0.000009707407,0.0007822037],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7810585,"threshold_uncertainty_score":0.9985495,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9220851009964572,"gpt":0.8041078746149898,"spread":0.1179772263814675,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.240632,0.000195284,0.00054918,0.002518669,0.003158268,0.0009636302,0.003281767,0.000568661,0.0003889091],"category_scores_gemma":[0.477455,0.0001738806,0.00007686738,0.005028972,0.005907601,0.001270809,0.006727421,0.01214787,0.00101392],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002410007,"about_ca_system_score_gemma":0.009573539,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.04582198,"about_ca_topic_score_gemma":0.1308576,"domain_scores_codex":[0.9828082,0.004035822,0.001343624,0.001906417,0.007331734,0.002574213],"domain_scores_gemma":[0.8753092,0.09518246,0.0003301474,0.008828728,0.01818358,0.002165934],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001652888,0.01030547,0.9273467,0.0009447468,0.00002736658,0.0001012623,0.0122684,0.00004069092,0.0003332892,0.02359899,0.007547678,0.01583254],"study_design_scores_gemma":[0.002052485,0.003248475,0.8936264,0.001830177,0.0000036959,0.0000253766,0.0444391,0.001488087,0.0001041649,0.03983393,0.01305076,0.0002973254],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4948779,0.0001682878,0.00003242673,0.4637107,0.0001443517,0.002795577,0.00001497862,0.00003121772,0.03822455],"genre_scores_gemma":[0.9812374,0.0005018513,0.01024559,0.0006807427,0.0001600285,0.0008045014,0.000008671418,0.00003658687,0.00632464],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4863594,"threshold_uncertainty_score":0.9997639,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.4799070722051362,"gpt":0.5656053467790647,"spread":0.08569827457392848,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009877126,0.0001149035,0.0004003299,0.000169062,0.0006536634,0.0001394398,0.0007623197,0.0005894023,0.001367611],"category_scores_gemma":[0.005506287,0.00008181314,0.0002707014,0.001563595,0.001316921,0.0006118213,0.0001305622,0.003138441,0.00001491179],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000311684,"about_ca_system_score_gemma":0.0008689714,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03098436,"about_ca_topic_score_gemma":0.06109786,"domain_scores_codex":[0.9928186,0.003949954,0.0005657758,0.0003957195,0.001848766,0.00042121],"domain_scores_gemma":[0.9948234,0.002763957,0.0001736181,0.0008027286,0.001341681,0.00009460321],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002000653,0.01452978,0.275624,0.001183257,0.004387444,0.0001112713,0.1400471,0.004142243,0.222483,0.2718139,0.008045426,0.05563194],"study_design_scores_gemma":[0.0005821118,0.0001996008,0.6015941,0.0003614793,0.001042052,0.00002159283,0.2300313,0.001597612,0.05071883,0.1036803,0.009637693,0.0005333899],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9727046,0.0003739653,0.00005205635,0.02091399,0.0006886643,0.0004786614,0.000100514,0.00001333697,0.004674234],"genre_scores_gemma":[0.9978712,0.0006340441,0.0001062962,0.00007215408,0.0001171989,0.00004143891,0.00001217102,0.000006490508,0.00113903],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3259701,"threshold_uncertainty_score":0.9995453,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.7205593581249751,"gpt":0.5510722679019152,"spread":0.1694870902230599,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2295272,0.0001132325,0.0002918312,0.0001797516,0.0009121173,0.00008118248,0.002646525,0.0004863226,0.00004727293],"category_scores_gemma":[0.06607984,0.00006212987,0.0001183153,0.002016664,0.004011362,0.0002335605,0.0003112624,0.007337388,0.000008502868],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001853008,"about_ca_system_score_gemma":0.0006803077,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.0247199,"about_ca_topic_score_gemma":0.02398135,"domain_scores_codex":[0.9654973,0.02833951,0.001153749,0.0004992562,0.003351681,0.001158578],"domain_scores_gemma":[0.9614795,0.03603966,0.0001781382,0.000571852,0.001635848,0.00009493826],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.0007275684,0.00026001,0.03476147,0.0004293827,0.00002145067,0.000002177523,0.4682376,0.00001327458,0.0008068433,0.4870889,0.003109614,0.004541665],"study_design_scores_gemma":[0.0004999893,0.0003356923,0.06546278,0.0002459022,0.000008938074,7.826677e-7,0.4835173,0.0005841596,0.001340633,0.4377931,0.01006333,0.0001473673],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8918561,0.0003429516,0.00005362755,0.07780925,0.0001125907,0.001648661,0.00009335683,0.00001044117,0.02807299],"genre_scores_gemma":[0.9991139,0.0003443566,0.00007517669,0.00007712214,0.0002607643,0.00008542467,0.000007559647,0.00001102061,0.00002466194],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1634474,"threshold_uncertainty_score":0.9986991,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.5869436868712309,"gpt":0.5262730286507429,"spread":0.06067065822048801,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.01353377,0.0001932164,0.000239285,0.001144427,0.0008760157,0.0001330837,0.002770931,0.0002996022,0.0000652123],"category_scores_gemma":[0.1234808,0.0001315116,0.00005176994,0.007598115,0.003081015,0.000228154,0.0009256631,0.008267077,0.00006928921],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003571903,"about_ca_system_score_gemma":0.0003996725,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002151376,"about_ca_topic_score_gemma":0.001779782,"domain_scores_codex":[0.9910876,0.003641813,0.0006698323,0.001049654,0.002709359,0.0008417516],"domain_scores_gemma":[0.9726056,0.0248172,0.00009557782,0.001853633,0.000547435,0.00008056034],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"qualitative","study_design_scores_codex":[0.0003197101,0.002412866,0.01723258,0.0001085003,0.00001030593,0.0001194724,0.3823952,0.0001308816,0.4618184,0.1271437,0.001848554,0.006459819],"study_design_scores_gemma":[0.001312678,0.00188141,0.1468458,0.000501348,0.000006824702,0.00002608515,0.3870359,0.0008854484,0.3459853,0.08556581,0.02893572,0.001017615],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7383671,0.00002073414,0.00009399606,0.2599185,0.0001243058,0.0009322187,0.000026072,0.00004896693,0.0004681076],"genre_scores_gemma":[0.9915324,0.0002362765,0.000152789,0.007775509,0.00005748883,0.0002074037,0.000001406359,0.00001982216,0.00001688967],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2531654,"threshold_uncertainty_score":0.999632,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.7046772079286926,"gpt":0.6201766418666188,"spread":0.08450056606207379,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0571439,0.000310389,0.005070612,0.001424517,0.00007816248,0.00006728802,0.0009413052,0.0009391525,0.00006631439],"category_scores_gemma":[0.07419674,0.0001404654,0.0007478113,0.01088928,0.006783188,0.00005785638,0.001023196,0.006149786,0.000002705466],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005488673,"about_ca_system_score_gemma":0.003332038,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001239949,"about_ca_topic_score_gemma":0.001448495,"domain_scores_codex":[0.9834399,0.007876333,0.004207269,0.0007395846,0.003230114,0.0005068188],"domain_scores_gemma":[0.8219902,0.1672136,0.001401166,0.003774129,0.005466109,0.0001548072],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00131711,0.00124137,0.05094503,0.8399991,0.001711446,0.0000369189,0.008464855,0.000004374072,4.233272e-7,0.07917207,0.00004036459,0.01706687],"study_design_scores_gemma":[0.007494422,0.0006150439,0.01207045,0.9279769,0.005480725,0.00004947628,0.009812418,0.0004130052,0.000006622653,0.007315407,0.02818642,0.0005790888],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.05812062,0.9313033,7.606022e-7,0.0008947825,0.00007465231,0.008803663,0.0001070783,0.00000513359,0.0006900647],"genre_scores_gemma":[0.2574311,0.7412795,0.00001246642,0.0000328637,0.00002718512,0.0009159106,0.00005173901,0.00001947642,0.0002297614],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1993104,"threshold_uncertainty_score":0.996143,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Thompson Rivers University","funders":"","keywords":"Direct Anonymous Attestation; Computer security; Internet privacy; Computer science; Scientific misconduct; Trusted Computing; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.2534849222970801,"gpt":0.4491510783357754,"spread":0.1956661560386953,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009342914,0.0005273293,0.0007534989,0.001858433,0.0004410962,0.001384308,0.006417305,0.001055164,0.0001845705],"category_scores_gemma":[0.006289001,0.0005359224,0.0001458903,0.006809411,0.0004091356,0.0009048234,0.003117937,0.00566786,0.002752776],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002155269,"about_ca_system_score_gemma":0.001586952,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0552473,"about_ca_topic_score_gemma":0.01467286,"domain_scores_codex":[0.9897093,0.0007376699,0.001039162,0.002378065,0.00324031,0.002895544],"domain_scores_gemma":[0.990083,0.003800906,0.0001268163,0.004290033,0.001413774,0.0002855085],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002738438,0.00009874188,0.001560197,0.0004169177,0.00001639974,0.000709967,0.00288145,0.0003040211,0.0001124626,0.0007451918,0.9902774,0.002849872],"study_design_scores_gemma":[0.0001452054,0.0003534491,0.0008600486,0.0004085524,0.000005800862,0.00001395985,0.0005753678,0.01545185,0.004539295,0.0370457,0.9396437,0.0009570711],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.01162209,0.00003833435,0.0007755681,0.9808806,0.0009358461,0.00294943,0.00005888327,0.0005706608,0.002168605],"genre_scores_gemma":[0.04022966,0.0001115583,0.0143819,0.9107402,0.003954655,0.006423162,0.0003753536,0.0004902538,0.02329328],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.0701404,"threshold_uncertainty_score":0.9997092,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8055258204019985,"gpt":0.7221545411890221,"spread":0.08337127921297638,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2569898,0.0002622313,0.0007546538,0.002800467,0.0005902298,0.0001317159,0.001449765,0.0007514317,0.0004185007],"category_scores_gemma":[0.40884,0.0002364007,0.0001805227,0.01024693,0.002508643,0.0002893372,0.002004316,0.01257425,0.001580734],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.004672208,"about_ca_system_score_gemma":0.007715416,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.02437499,"about_ca_topic_score_gemma":0.02581579,"domain_scores_codex":[0.9784924,0.001951302,0.002259415,0.001865742,0.01206414,0.003366954],"domain_scores_gemma":[0.8145508,0.1478537,0.0002008453,0.008911976,0.02444842,0.004034209],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.04425983,0.009652996,0.4431779,0.04756758,0.0002490344,0.003086124,0.007738305,0.00005893972,0.01539556,0.1881832,0.08401228,0.1566182],"study_design_scores_gemma":[0.004216128,0.002205851,0.6984708,0.01916836,0.00001268419,0.00007913991,0.001975424,0.00004336104,0.003763056,0.04526287,0.2242151,0.0005871658],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7416387,0.002866647,0.00004301577,0.1210145,0.0001885949,0.004804688,0.00005666456,0.00009167169,0.1292956],"genre_scores_gemma":[0.9818119,0.009377095,0.001384641,0.003494997,0.0009741264,0.0002856154,0.00004306465,0.00005164851,0.002576849],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.255293,"threshold_uncertainty_score":0.9991966,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.920944469478084,"gpt":0.7754212306248044,"spread":0.1455232388532796,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts","research_integrity"],"category_scores_codex":[0.6315038,0.0002489122,0.000960355,0.00137806,0.00142709,0.00003673927,0.001566798,0.002538818,0.001952489],"category_scores_gemma":[0.4610638,0.0002008342,0.0001013052,0.004566269,0.01248127,0.0007822313,0.003139121,0.05640023,0.0003339848],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002232418,"about_ca_system_score_gemma":0.004218556,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03806001,"about_ca_topic_score_gemma":0.02831196,"domain_scores_codex":[0.8363091,0.1464356,0.00288754,0.001202233,0.008835919,0.004329639],"domain_scores_gemma":[0.6113883,0.3795513,0.0003787483,0.002416586,0.005485731,0.0007793336],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.002441459,0.000936752,0.7338591,0.001812457,0.00001860909,0.00001329474,0.08206441,0.000004180093,0.0003846777,0.1739215,0.0008672071,0.003676377],"study_design_scores_gemma":[0.001800414,0.0005744392,0.2249021,0.00137675,0.000008790656,0.000002283066,0.4402186,0.0001541575,0.0006073512,0.3199393,0.01007098,0.0003447875],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9076426,0.000573866,0.0002140009,0.05392302,0.0007870423,0.002345148,0.00005723378,0.00005375757,0.03440331],"genre_scores_gemma":[0.9812781,0.0006571123,0.01642033,0.0004130542,0.0004747588,0.0005607509,0.00001046965,0.00004274113,0.0001426721],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.508957,"threshold_uncertainty_score":0.9998729,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8768308426194927,"gpt":0.6745009165671052,"spread":0.2023299260523875,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.01061681,0.0001179068,0.000273041,0.0002613068,0.0001611099,0.00001848915,0.0007204824,0.0003575191,0.0005816169],"category_scores_gemma":[0.001640934,0.00009307867,0.00002945872,0.001938196,0.001225062,0.0002273216,0.0001755787,0.002885684,0.00001622857],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002028396,"about_ca_system_score_gemma":0.0003320797,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008387577,"about_ca_topic_score_gemma":0.002115985,"domain_scores_codex":[0.9961694,0.00188414,0.0007028762,0.0003134594,0.0003398136,0.0005903266],"domain_scores_gemma":[0.9961827,0.002890779,0.0000865796,0.0003679477,0.000339426,0.0001325608],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003203912,0.0008136261,0.9600964,0.00008321129,0.000008802139,0.000009161991,0.007901578,0.0001058147,0.01323083,0.01519726,0.0001376776,0.002095218],"study_design_scores_gemma":[0.0007698106,0.0003222982,0.9483223,0.00008796203,0.000006372431,0.000002297081,0.008138695,0.0005102063,0.03188403,0.006288428,0.003562665,0.0001049599],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9861946,0.0005930738,9.766634e-7,0.009947632,0.0001753928,0.0006131392,0.00003677,0.00001024458,0.002428227],"genre_scores_gemma":[0.9990738,0.0002253756,0.00001324179,0.0005466784,0.00005022319,0.00003226334,0.000003452573,0.000004622391,0.0000503141],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0186532,"threshold_uncertainty_score":0.9994147,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9007893159834213,"gpt":0.7061770508876674,"spread":0.1946122650957539,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.05763039,0.000215682,0.0004316895,0.0006366485,0.00182418,0.0002022252,0.00119627,0.000348844,0.0001858568],"category_scores_gemma":[0.06495445,0.0001543581,0.000203364,0.002665996,0.003169403,0.0001473167,0.000893302,0.006273189,0.0003691415],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001061087,"about_ca_system_score_gemma":0.00368966,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001895287,"about_ca_topic_score_gemma":0.00298302,"domain_scores_codex":[0.9880972,0.002261133,0.0009388209,0.001191172,0.005447047,0.002064618],"domain_scores_gemma":[0.8516327,0.140057,0.00005664369,0.002325875,0.005274824,0.0006529754],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003434427,0.001036235,0.973023,0.0005563976,0.00003224966,0.00009863661,0.003717144,0.00001108538,0.0008866519,0.004383563,0.008899204,0.00392139],"study_design_scores_gemma":[0.003965947,0.002750782,0.8854146,0.000255404,0.00002319906,0.0001384663,0.01272588,0.0005973214,0.0008811107,0.07512065,0.01773467,0.0003919621],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9400295,0.001028921,0.00003750017,0.05061615,0.0002328442,0.004642468,0.0000189814,0.00004690187,0.003346744],"genre_scores_gemma":[0.9909009,0.003253957,0.0002865909,0.0006522165,0.0007431923,0.001396103,0.00001012391,0.00005358619,0.002703379],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1377959,"threshold_uncertainty_score":0.9995434,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9759870882107871,"gpt":0.7668027338591651,"spread":0.209184354351622,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2605164,0.0001581886,0.0004748315,0.0009247004,0.0006410179,0.00007674303,0.001746451,0.000938252,0.0007824323],"category_scores_gemma":[0.2425724,0.000105966,0.0001466383,0.004150064,0.01759621,0.0001773937,0.001530677,0.01582427,0.000338482],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005838691,"about_ca_system_score_gemma":0.006799751,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01945857,"about_ca_topic_score_gemma":0.02036097,"domain_scores_codex":[0.9792647,0.007183234,0.001520648,0.001043884,0.008926488,0.002061081],"domain_scores_gemma":[0.6895828,0.2814947,0.0001629863,0.006675487,0.02134567,0.0007383374],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.01317079,0.002623241,0.4613372,0.001375668,0.00009374511,0.0001191951,0.01027862,5.701753e-7,0.009776697,0.4915332,0.003923989,0.005767145],"study_design_scores_gemma":[0.001430908,0.00209733,0.2403747,0.0005488462,0.000007704602,0.00001496545,0.0235675,0.00006649611,0.04479177,0.6806973,0.006233518,0.0001688961],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9018977,0.0006567705,0.000002176801,0.01992313,0.0001389059,0.001644093,0.0000108197,0.00002652752,0.07569993],"genre_scores_gemma":[0.9929596,0.002200797,0.0008931077,0.00009887592,0.0000977849,0.0001882598,0.000003074272,0.00003655232,0.003521964],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2896818,"threshold_uncertainty_score":0.9988308,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.6390324670345174,"gpt":0.5817224741848028,"spread":0.05730999284971461,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1996645,0.0001249489,0.0003958291,0.001541568,0.0008254462,0.002587297,0.005854645,0.0001828658,0.003836345],"category_scores_gemma":[0.1119704,0.0001082582,0.00006234799,0.006600922,0.00111677,0.005220801,0.006182205,0.003363967,0.00002278346],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004877675,"about_ca_system_score_gemma":0.001456801,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0448349,"about_ca_topic_score_gemma":0.007985401,"domain_scores_codex":[0.9747624,0.01241931,0.001224839,0.001270179,0.009388166,0.0009351479],"domain_scores_gemma":[0.9661059,0.02795331,0.0002412688,0.002131309,0.003192789,0.0003754081],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001391947,0.0001904928,0.9590752,0.00001391068,0.000004794695,0.000002171555,0.004032718,0.00008256365,0.00007269611,0.0003214975,0.02835306,0.007711703],"study_design_scores_gemma":[0.0003221282,0.00006107522,0.9368197,0.000006675528,7.507858e-7,0.000004088454,0.03556569,0.001079008,0.00001483559,0.0168341,0.009188738,0.0001031799],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9909616,0.0002429126,0.00002495401,0.003626369,0.0001548619,0.0004816836,0.0004098364,0.00001685235,0.004080995],"genre_scores_gemma":[0.9985186,0.00002661663,0.00004218714,0.00009554216,0.00003971481,0.0001805831,0.00003624514,0.00001289996,0.001047649],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08769407,"threshold_uncertainty_score":0.9995242,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.89119809906341,"gpt":0.692302295726708,"spread":0.198895803336702,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1092227,0.0003408677,0.001017773,0.001242325,0.0003448597,0.0001224753,0.002505963,0.0009373621,0.001156952],"category_scores_gemma":[0.376284,0.0002346343,0.0001826328,0.007606333,0.006229348,0.0003086746,0.00231202,0.01220982,0.0002837563],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001260832,"about_ca_system_score_gemma":0.01014667,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1704067,"about_ca_topic_score_gemma":0.4708686,"domain_scores_codex":[0.9781872,0.006917079,0.002424863,0.001901456,0.008097786,0.002471585],"domain_scores_gemma":[0.7415402,0.2276979,0.0003140941,0.009450416,0.01939733,0.001600027],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003786404,0.003068975,0.9721956,0.002618575,0.00003121856,0.00004664246,0.01097564,0.000004513621,0.000529449,0.001767847,0.0009961615,0.003978993],"study_design_scores_gemma":[0.001339729,0.001214893,0.9115607,0.001279611,0.000007639314,0.000005841295,0.01512136,0.00001821606,0.002945211,0.06510897,0.001196075,0.0002017121],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9499691,0.0006038302,0.000008668019,0.02188545,0.0001103063,0.005045103,0.00005560615,0.00003301502,0.02228893],"genre_scores_gemma":[0.9952917,0.002167439,0.0003802206,0.0003552443,0.0001553247,0.0009573299,0.000009544241,0.00005532901,0.0006278707],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3004619,"threshold_uncertainty_score":0.9997561,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8127094389806951,"gpt":0.5201771416261698,"spread":0.2925322973545254,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.02160179,0.0002215361,0.001090868,0.0008675175,0.0009100615,0.0002500063,0.0007361634,0.0002282211,0.0001805997],"category_scores_gemma":[0.03292416,0.0001331157,0.0001821228,0.003163989,0.003900408,0.0003998887,0.0007597651,0.001193831,0.00004919311],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001308971,"about_ca_system_score_gemma":0.0006476651,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00253681,"about_ca_topic_score_gemma":0.0003376335,"domain_scores_codex":[0.9951864,0.001132838,0.001068351,0.0006009175,0.001386654,0.0006248089],"domain_scores_gemma":[0.9886612,0.008681685,0.0002182099,0.0006800308,0.0017374,0.00002146751],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002865292,0.0009038948,0.0001184706,0.4513544,0.00004300808,0.00001803731,0.0005313945,5.030233e-7,0.000001956384,0.03524954,0.2274019,0.2840903],"study_design_scores_gemma":[0.0003686779,0.00003410292,0.00001552671,0.02024684,0.000029452,0.00001480298,0.00009052981,0.0002619084,1.851249e-7,0.006710635,0.9720991,0.000128182],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00003629589,0.9894469,0.00002735193,0.002729206,0.0001840291,0.005737429,0.00004451828,0.00001255658,0.00178172],"genre_scores_gemma":[0.0006205922,0.9968138,0.0000231231,0.001114219,0.0002936186,0.000921422,0.00008648949,0.00002713621,0.00009954638],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7446973,"threshold_uncertainty_score":0.9988104,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8001993503299486,"gpt":0.7145235454461236,"spread":0.08567580488382498,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1184332,0.0001903934,0.0009158732,0.0009438117,0.0003606318,0.0000209155,0.0009627918,0.0001975716,0.0005509679],"category_scores_gemma":[0.1973482,0.0001497147,0.00006493749,0.00513916,0.003800269,0.0002666148,0.006928649,0.005456035,0.000004542831],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001313136,"about_ca_system_score_gemma":0.003602389,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.03269121,"about_ca_topic_score_gemma":0.02032034,"domain_scores_codex":[0.9794958,0.006576912,0.00246141,0.001171818,0.008889792,0.001404257],"domain_scores_gemma":[0.9322127,0.04701775,0.0003608972,0.004168915,0.01541126,0.0008285154],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.008357234,0.006066829,0.9130301,0.02853671,0.00005276861,0.00003420332,0.01298692,0.00002871683,0.002035785,0.002310924,0.000220414,0.02633942],"study_design_scores_gemma":[0.01502097,0.02934945,0.5630691,0.04516428,0.00007568154,0.00004372826,0.08576573,0.0056049,0.007339715,0.2451641,0.002411463,0.000990877],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9797761,0.001692349,0.000002657239,0.01474226,0.00002035122,0.002880061,0.00004980063,0.00001091997,0.0008255156],"genre_scores_gemma":[0.982338,0.01609254,0.0004608944,0.0003777078,0.00005039509,0.0005338714,0.00003073726,0.00003466548,0.00008126918],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.349961,"threshold_uncertainty_score":0.9989108,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8656019701839802,"gpt":0.6518773040487474,"spread":0.2137246661352328,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03451413,0.00008735557,0.0002966441,0.002172322,0.0002068589,0.00003605932,0.0008941502,0.0001810811,0.0000272005],"category_scores_gemma":[0.01266998,0.00006513062,0.0001088424,0.007344957,0.001566307,0.000277874,0.0001304504,0.00123421,0.000001713339],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001612007,"about_ca_system_score_gemma":0.0002008426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00362572,"about_ca_topic_score_gemma":0.004272861,"domain_scores_codex":[0.9926682,0.003575038,0.0007389615,0.0004223376,0.001816465,0.0007789881],"domain_scores_gemma":[0.9959737,0.0029868,0.00006584229,0.000333826,0.0005865505,0.00005326923],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004968376,0.006041226,0.6903933,0.0002879475,0.00001028531,0.000001536526,0.03076039,0.0001788766,0.1510088,0.1039743,0.002230012,0.01461646],"study_design_scores_gemma":[0.000489409,0.0001291807,0.9770913,0.00001841374,0.000004811821,4.09254e-7,0.00121945,0.0009563076,0.001653507,0.01591988,0.002445509,0.00007181546],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9365214,0.00009426754,0.00001273801,0.06233983,0.00006411901,0.0006788849,0.00004549674,0.000003500082,0.0002397806],"genre_scores_gemma":[0.9992678,0.00006083004,0.00002316847,0.0003219018,0.0001330231,0.0001718889,0.00001127078,0.000004458265,0.000005674119],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.286698,"threshold_uncertainty_score":0.9956467,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.1095306596981785,"gpt":0.4822537312601409,"spread":0.3727230715619624,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.02439342,0.0000632701,0.0001886022,0.0001225648,0.0004118829,0.00007876083,0.0005385655,0.000337632,0.0001317978],"category_scores_gemma":[0.003564362,0.00004240441,0.00004017735,0.0008698033,0.003631497,0.0002025967,0.0002529553,0.001945534,0.000006515381],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001622598,"about_ca_system_score_gemma":0.0004824844,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.09586725,"about_ca_topic_score_gemma":0.02239526,"domain_scores_codex":[0.9964663,0.00184881,0.0002765755,0.0002218157,0.0008087014,0.0003778391],"domain_scores_gemma":[0.9962835,0.003028119,0.00002959787,0.0002762121,0.0003396265,0.00004294869],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002355228,0.00008648157,0.4466364,0.00004917086,0.00001033599,6.799941e-7,0.01738118,0.000001832664,0.00001060101,0.5268452,0.00517583,0.003566798],"study_design_scores_gemma":[0.001154509,0.0000289091,0.3052891,0.0001113479,0.000009014176,3.409993e-7,0.04927852,0.0002537512,0.000159293,0.5840175,0.05958318,0.0001145951],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8547173,0.0004231708,0.00001791873,0.08568349,0.0001258064,0.0006831831,0.00001077792,0.0000108166,0.05832752],"genre_scores_gemma":[0.9980212,0.0003927811,0.00002477522,0.0005220968,0.00004886229,0.00004759952,0.000001513795,0.000002117664,0.0009390851],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1433039,"threshold_uncertainty_score":0.9990801,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Research integrity; Engineering ethics; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.4588957539914663,"gpt":0.6163961553527757,"spread":0.1575004013613094,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","sts","research_integrity"],"consensus_categories":["metaresearch","bibliometrics","sts"],"category_scores_codex":[0.1475381,0.0001537199,0.0006600815,0.0153386,0.00133585,0.0001855384,0.001844001,0.0008792377,0.0006796272],"category_scores_gemma":[0.08271674,0.0001392617,0.0001940367,0.09456462,0.005993008,0.0004763466,0.0009819049,0.007628629,0.00004548741],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001967065,"about_ca_system_score_gemma":0.001835842,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1122736,"about_ca_topic_score_gemma":0.0581145,"domain_scores_codex":[0.9862051,0.006399182,0.0008806075,0.0008367576,0.004242729,0.001435613],"domain_scores_gemma":[0.9722717,0.02059148,0.00009720324,0.0008019654,0.006070072,0.000167587],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001021355,0.0003209695,0.9009374,0.0002830482,0.0003393523,0.000004682106,0.05080721,0.00001736096,0.0001295813,0.03063578,0.01099249,0.005429933],"study_design_scores_gemma":[0.000434239,0.00007951789,0.3184798,0.0004852999,0.00005426282,2.92105e-7,0.5517645,0.0005354264,0.0006379452,0.083461,0.04380872,0.0002590498],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9309634,0.003184621,0.00009269847,0.01006091,0.0003256643,0.0007892075,0.00002161288,0.00003569687,0.05452618],"genre_scores_gemma":[0.991554,0.001843436,0.0003592763,0.00006802715,0.0001099483,0.00009850514,0.000004490778,0.0000054417,0.005956931],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5824577,"threshold_uncertainty_score":0.9999643,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.4771556329383189,"gpt":0.5741450469816433,"spread":0.09698941404332434,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.08196381,0.0001632,0.000630057,0.001729388,0.0002597455,0.00005823945,0.001856237,0.0002178104,0.0002985172],"category_scores_gemma":[0.01751875,0.000130815,0.0001086719,0.008742091,0.001076662,0.000189163,0.00141147,0.004303665,0.00001046137],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002443665,"about_ca_system_score_gemma":0.003720371,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9678085,"about_ca_topic_score_gemma":0.9073222,"domain_scores_codex":[0.9732027,0.01997554,0.0006835343,0.0009349437,0.004000929,0.001202337],"domain_scores_gemma":[0.9769349,0.01933143,0.00007015665,0.001468851,0.001923798,0.0002708861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005143519,0.0001918214,0.9518516,0.0003668228,0.00006335517,0.0000225816,0.003182398,0.0004675621,0.03718566,0.001292218,0.00001338162,0.0002190718],"study_design_scores_gemma":[0.000347704,0.001396315,0.9734228,0.0003197719,0.000006951344,0.000001861765,0.01621947,0.005152246,0.00225232,0.0007333879,0.00002537669,0.0001218203],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9960156,0.00001883823,9.943658e-7,0.0002787272,0.00004451335,0.00134101,0.00005607493,0.0000078179,0.002236429],"genre_scores_gemma":[0.9997493,0.000001736129,0.00004210466,0.00001731019,0.00002766005,0.0001142344,0.000005005852,0.00002389439,0.00001875872],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06444506,"threshold_uncertainty_score":0.9979935,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9255360272906197,"gpt":0.7014977101768047,"spread":0.2240383171138149,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.65358,0.0001987963,0.00237163,0.0008650254,0.0005155026,0.004670162,0.002603834,0.0001526766,0.00543487],"category_scores_gemma":[0.5238494,0.0001010087,0.00078749,0.003610549,0.0003316993,0.0008851154,0.0005014241,0.0009547751,0.0003481852],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002122961,"about_ca_system_score_gemma":0.0005811258,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001572391,"about_ca_topic_score_gemma":0.0005673433,"domain_scores_codex":[0.9478865,0.021174,0.0105992,0.001781519,0.01766499,0.0008937537],"domain_scores_gemma":[0.83149,0.08230063,0.003495499,0.01275776,0.06926929,0.0006868272],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002927571,0.00006470059,0.06840558,0.0005055745,0.00003474863,8.147597e-7,0.000207358,5.096453e-7,0.00001126361,0.000935692,0.8870756,0.0427289],"study_design_scores_gemma":[0.0002621485,0.00002398667,0.05763252,0.0002719671,0.00003233144,0.000005312349,0.001094488,0.0003365822,0.000007081427,0.08608914,0.8541492,0.00009524411],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5453355,0.02864467,0.003521395,0.3351773,0.001558209,0.009134911,0.00004998569,0.000009674653,0.07656832],"genre_scores_gemma":[0.7592888,0.003261808,0.00721934,0.00926051,0.0003214194,0.001144692,0.0000167972,0.00002812122,0.2194585],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3259168,"threshold_uncertainty_score":0.9963631,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2987618849254758,"gpt":0.5154437601731929,"spread":0.2166818752477171,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09036314,0.0001468229,0.0002438885,0.0004388204,0.00137024,0.0005955793,0.0007673922,0.0005642544,0.00004923868],"category_scores_gemma":[0.01067076,0.0001026728,0.00004084499,0.001370655,0.0009988857,0.0007341721,0.0002717329,0.005682724,0.00002869656],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009383064,"about_ca_system_score_gemma":0.0006645785,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01557712,"about_ca_topic_score_gemma":0.03501576,"domain_scores_codex":[0.9907866,0.005315752,0.0004339507,0.0007378984,0.001676349,0.001049483],"domain_scores_gemma":[0.9903879,0.008395216,0.00002355212,0.0004167547,0.0006026813,0.0001738734],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0005758767,0.0008358675,0.3309211,0.0001076683,0.00004207256,0.00004075756,0.6350852,0.000003716611,0.0001703133,0.02529673,0.003611493,0.003309128],"study_design_scores_gemma":[0.000435563,0.0008709747,0.2774211,0.0001723602,0.000003641403,5.748825e-7,0.6468742,0.0003500095,0.0001908784,0.06574275,0.007738471,0.0001995132],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.984134,0.0007023267,4.262744e-7,0.009965893,0.0003062625,0.003066266,0.000024885,0.00003629427,0.001763618],"genre_scores_gemma":[0.9987915,0.0003693768,0.000002985292,0.0000221261,0.0001193284,0.0003005999,0.000007039424,0.00001472876,0.0003723732],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07969238,"threshold_uncertainty_score":0.9999298,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["research_integrity"],"domain":null,"study_design":"observational","genre":"empirical","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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9748763335874832,"gpt":0.8580912241453093,"spread":0.1167851094421739,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","bibliometrics","sts","scholarly_communication","open_science","research_integrity"],"consensus_categories":["metaresearch","bibliometrics","sts","research_integrity"],"category_scores_codex":[0.5591915,0.001088767,0.00583178,0.01246134,0.001533718,0.001085498,0.003991198,0.003156809,0.0004718747],"category_scores_gemma":[0.6254317,0.0008833486,0.0005887072,0.02380514,0.004584848,0.0003180818,0.01768442,0.07668133,0.0003126059],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.007508072,"about_ca_system_score_gemma":0.03026244,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03698132,"about_ca_topic_score_gemma":0.01109982,"domain_scores_codex":[0.7909375,0.144926,0.009847914,0.009836591,0.03379824,0.01065374],"domain_scores_gemma":[0.4754586,0.4418102,0.0003845107,0.01287543,0.06577154,0.003699736],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0004958234,0.003115605,0.004739342,0.0671479,0.0002175811,0.0001306481,0.001205405,4.068568e-7,0.000004907704,0.001091168,0.005797877,0.9160534],"study_design_scores_gemma":[0.001121355,0.001745073,0.001061474,0.0829601,0.0001853957,0.00002108149,0.01035097,0.00007323145,0.000009676704,0.02419447,0.8776367,0.0006405057],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.004535987,0.9673771,0.00004902844,0.00419296,0.0007451403,0.02166618,0.0001469132,0.00008503023,0.001201642],"genre_scores_gemma":[0.0001770389,0.9682961,0.02469313,0.00005774184,0.0009682867,0.003392786,0.00007523852,0.0002126393,0.00212704],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9154128,"threshold_uncertainty_score":0.9999515,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"","keywords":"Risk assessment; Risk analysis (engineering); Psychology; Medicine; Computer science; Computer security","retraction":null,"screen_n_in":null,"score":{"opus":0.7439250363914514,"gpt":0.6797050059099999,"spread":0.06422003048145142,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.1973978,0.0002074397,0.0008454493,0.001100512,0.001428691,0.000466053,0.003358352,0.0005379388,0.0001754993],"category_scores_gemma":[0.1483857,0.0001204028,0.0003822028,0.003069541,0.00669578,0.0003903651,0.001701603,0.01106896,0.00001214232],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008138336,"about_ca_system_score_gemma":0.001192903,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1111069,"about_ca_topic_score_gemma":0.09545394,"domain_scores_codex":[0.978365,0.008318594,0.001766989,0.001005922,0.009177534,0.00136598],"domain_scores_gemma":[0.8692165,0.1196119,0.0004821429,0.006508812,0.003987323,0.0001933013],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001412076,0.002021373,0.9823636,0.000127153,0.0002263082,0.00001513425,0.003493721,0.00005749084,0.0006207018,0.004923934,0.00004958718,0.004688882],"study_design_scores_gemma":[0.001603902,0.0006422504,0.9582189,0.0001407039,0.00005714307,6.474118e-7,0.01001311,0.003210733,0.002462066,0.0234342,0.0001227714,0.00009362303],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9907122,0.0002489119,0.000007513359,0.003646223,0.00005110907,0.002051567,0.00007751131,0.000006731368,0.003198193],"genre_scores_gemma":[0.9966874,0.002121673,0.0005805292,0.00001599665,0.00006538365,0.0003510559,0.000008541771,0.00002498025,0.0001444593],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1112933,"threshold_uncertainty_score":0.9998713,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Engineering ethics; Psychology; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.4544497266217594,"gpt":0.6545098781449135,"spread":0.200060151523154,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.01803063,0.00008045858,0.0001402358,0.0002195334,0.0009926517,0.001229331,0.0003809356,0.0002887353,0.0003732597],"category_scores_gemma":[0.006119967,0.00008003879,0.0000873797,0.0009563943,0.0006342429,0.0009314907,0.00006906225,0.0008262699,0.00007134994],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006300006,"about_ca_system_score_gemma":0.0009495543,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.01958681,"about_ca_topic_score_gemma":0.02366403,"domain_scores_codex":[0.9970919,0.0005325711,0.0002376293,0.0003729165,0.001114712,0.0006502721],"domain_scores_gemma":[0.996613,0.002204735,0.00001359593,0.0002163799,0.000825228,0.0001271167],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007898451,0.0002471877,0.04709544,0.0004536104,0.00004584519,0.000009277869,0.4106577,0.000008604331,0.0003141094,0.4551799,0.07016202,0.01574738],"study_design_scores_gemma":[0.0002586223,0.0001318827,0.008824402,0.0001914914,0.000008631616,1.57081e-7,0.04729299,0.001032856,0.000185381,0.4581397,0.48366,0.0002738968],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8423279,0.0003299903,0.0002125544,0.1161516,0.0006338167,0.001503305,0.00005266516,0.0001444074,0.0386438],"genre_scores_gemma":[0.9967706,0.0001361802,0.00006350981,0.0007088697,0.000405509,0.00008575057,0.00001217246,0.00001754557,0.001799829],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.413498,"threshold_uncertainty_score":0.9998075,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2649465839690857,"gpt":0.4353178113501529,"spread":0.1703712273810672,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009488163,0.0002042489,0.0004484692,0.0002715302,0.0003590496,0.00004228898,0.0007028243,0.0003819778,0.00006462917],"category_scores_gemma":[0.0002830358,0.0001252296,0.0001786012,0.0009255082,0.002059971,0.0001081223,0.000269523,0.001310725,0.000006120967],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003039939,"about_ca_system_score_gemma":0.0004767522,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004366063,"about_ca_topic_score_gemma":0.01185865,"domain_scores_codex":[0.9953022,0.0008657092,0.000809503,0.0005475718,0.001724167,0.0007508242],"domain_scores_gemma":[0.9968743,0.0009010853,0.0002214731,0.001131879,0.0007392566,0.0001319661],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009356119,0.001452313,0.949213,0.003285337,0.0000580881,0.00002131927,0.02064818,0.00001282347,0.01468653,0.002162103,0.006030675,0.001494039],"study_design_scores_gemma":[0.002538049,0.0005191535,0.9657795,0.001555901,0.00004743782,0.000008148048,0.0187378,0.0002862707,0.003304512,0.003899347,0.003101212,0.0002226229],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9868509,0.0002356477,0.00002087,0.009032589,0.0002690495,0.002377861,0.00002407563,0.00001555397,0.00117347],"genre_scores_gemma":[0.9986572,0.00004920908,0.00006221748,0.0003849255,0.0001902016,0.0002021222,0.00001286752,0.00002766761,0.0004136136],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01656657,"threshold_uncertainty_score":0.7590047,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.3079548212433394,"gpt":0.4784316836647877,"spread":0.1704768624214483,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.03132682,0.0001060657,0.0002962427,0.000571103,0.0001369388,0.00002311005,0.0003761351,0.0004734881,0.0002321574],"category_scores_gemma":[0.0004153572,0.00007332044,0.0000370105,0.00277041,0.0013065,0.0001173334,0.0002273914,0.004404328,0.000007161309],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001975731,"about_ca_system_score_gemma":0.003132807,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8888856,"about_ca_topic_score_gemma":0.8300552,"domain_scores_codex":[0.9923502,0.002131932,0.0006376204,0.0003250322,0.003732305,0.0008229078],"domain_scores_gemma":[0.9952263,0.003180338,0.00004849929,0.0005813264,0.0008114809,0.000152082],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004804173,0.001831436,0.9573001,0.001698462,0.00001107965,0.00006119526,0.01342752,0.00003776368,0.009658931,0.01015407,0.001686883,0.003652201],"study_design_scores_gemma":[0.001003633,0.00006403853,0.9857295,0.000287325,0.00000224729,0.000003849699,0.008384841,0.0001734766,0.0002657579,0.002245717,0.001768821,0.00007075677],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9604979,0.0002790559,0.00000151156,0.03594626,0.00008421949,0.0007445067,0.00000657796,0.000003063695,0.002436878],"genre_scores_gemma":[0.9993002,0.00007143336,0.00001895794,0.0003125028,0.0001460751,0.00006039309,0.00002642985,0.00001069353,0.00005324337],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05883032,"threshold_uncertainty_score":0.9978926,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Engineering ethics; Research integrity; Management science; Political science; Psychology; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.957616822100931,"gpt":0.7360190920615394,"spread":0.2215977300393916,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01408212,0.0002355808,0.001234453,0.001906139,0.00007821815,0.0001430617,0.0003167262,0.0002358842,0.00002901925],"category_scores_gemma":[0.005933596,0.0001922842,0.00009225657,0.001531702,0.001055952,0.000274656,0.0006827588,0.001663481,0.000004571959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000412764,"about_ca_system_score_gemma":0.0004039494,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009484913,"about_ca_topic_score_gemma":0.00008082065,"domain_scores_codex":[0.9923046,0.004475323,0.001001478,0.0006368392,0.0008917474,0.0006900052],"domain_scores_gemma":[0.9917518,0.007365638,0.0001804337,0.0003100581,0.0002873122,0.0001047615],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.001883376,0.001248527,0.01262724,0.2603022,0.0002321954,0.00006026852,0.03602656,9.321652e-7,0.006706024,0.001886511,0.0005510385,0.6784751],"study_design_scores_gemma":[0.004753009,0.02222943,0.06761631,0.3884237,0.0006431666,0.0006655887,0.3427818,0.0006709389,0.001397571,0.08636139,0.08149038,0.002966774],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.221269,0.7772194,0.000004102692,0.00003198765,0.00001922766,0.0008683448,0.00003802099,0.00001160292,0.0005383449],"genre_scores_gemma":[0.09505074,0.9007587,0.003870632,8.34289e-7,0.00007323844,0.0001512289,0.00001106474,0.00002898574,0.00005455427],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6755083,"threshold_uncertainty_score":0.997111,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Columbia College","funders":"","keywords":"Physical activity; Component (thermodynamics); Research ethics","retraction":null,"screen_n_in":null,"score":{"opus":0.2416157993572732,"gpt":0.4966661895393897,"spread":0.2550503901821164,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0120796,0.0001291428,0.0002085446,0.0004738815,0.0003142838,0.00007118831,0.0004746345,0.0002066645,0.000371063],"category_scores_gemma":[0.001682539,0.0001282667,0.00005313159,0.0008195125,0.000121768,0.0001253959,0.0002091814,0.00103473,0.0001521183],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008040211,"about_ca_system_score_gemma":0.0002437665,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001163846,"about_ca_topic_score_gemma":0.0001864494,"domain_scores_codex":[0.9971579,0.0004290384,0.0005106392,0.0006088344,0.0004531126,0.0008405071],"domain_scores_gemma":[0.9969311,0.002099401,0.00004466009,0.0006070379,0.0002333577,0.00008442209],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001005279,0.0009093698,0.774655,0.0005598544,0.00006432498,0.000005269539,0.01035199,0.000007143447,0.0006942179,0.03220684,0.07711222,0.1024284],"study_design_scores_gemma":[0.004812046,0.000151699,0.5893614,0.0005835532,0.00003413664,0.000001673145,0.03079689,0.0004550788,0.001111546,0.05324937,0.3188963,0.0005462802],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9486915,0.0007881499,0.0005685302,0.003078045,0.002528299,0.002396243,0.00001798956,0.00008212767,0.04184906],"genre_scores_gemma":[0.9742373,0.00001881793,0.000752117,0.00006926127,0.00007942687,0.002207186,0.00001687191,0.00001425501,0.02260481],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2417841,"threshold_uncertainty_score":0.5230564,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.9098084295788003,"gpt":0.7336634680487633,"spread":0.176144961530037,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2180995,0.0002488066,0.0006238113,0.002273525,0.002954717,0.0003194146,0.001249977,0.001060397,0.0001472402],"category_scores_gemma":[0.1099566,0.000249705,0.0001339436,0.007590806,0.00121371,0.0003971739,0.001353584,0.01424124,0.0004900284],"about_ca_system_candidate":true,"about_ca_system_consensus":true,"about_ca_system_score_codex":0.0142747,"about_ca_system_score_gemma":0.04909778,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.3090714,"about_ca_topic_score_gemma":0.8378974,"domain_scores_codex":[0.9823068,0.002439709,0.001375787,0.001265142,0.009119248,0.003493274],"domain_scores_gemma":[0.9311243,0.05589784,0.00008733488,0.002058555,0.00841249,0.002419468],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00340745,0.001419531,0.7972743,0.007318468,0.00006587076,0.0001115176,0.03596833,0.0001689495,0.001348855,0.1139861,0.01342771,0.02550292],"study_design_scores_gemma":[0.001937893,0.003188736,0.870542,0.001208939,0.000003952759,0.00001073532,0.01006869,0.008054256,0.0001120172,0.08699285,0.01750281,0.000377075],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8837421,0.00002286721,0.00007564894,0.1067444,0.0001901347,0.007156204,0.0005362011,0.0001200237,0.001412384],"genre_scores_gemma":[0.9897767,0.00008557623,0.003593488,0.002071491,0.0004389832,0.00252828,0.0004910101,0.00007229698,0.0009421875],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.528826,"threshold_uncertainty_score":0.9999955,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2404113526584181,"gpt":0.4912848617688953,"spread":0.2508735091104772,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01771019,0.0001499046,0.0002930721,0.000864588,0.0005583718,0.000200626,0.0003724125,0.00006675813,0.0001255765],"category_scores_gemma":[0.004555689,0.0001216357,0.00003391479,0.002616786,0.0003570167,0.000356199,0.0002675205,0.00185488,0.00008237392],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005732429,"about_ca_system_score_gemma":0.0001529922,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004820253,"about_ca_topic_score_gemma":0.002834259,"domain_scores_codex":[0.9953649,0.0008178034,0.0005555276,0.0007060807,0.001183128,0.001372553],"domain_scores_gemma":[0.9970185,0.001623365,0.00002552405,0.0008221489,0.0002859189,0.0002245071],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000433841,0.0001419552,0.6638254,0.00004265452,0.000001699627,0.0000470827,0.004327424,0.00002936877,0.0001025968,0.0001095687,0.0006173099,0.3303211],"study_design_scores_gemma":[0.001009177,0.0001354208,0.9738526,0.0005536767,0.00000354848,0.00001639106,0.004050004,0.007189431,0.0001566381,0.002244726,0.01066923,0.0001191948],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.892742,0.005028165,0.000004431256,0.09058461,0.00003745925,0.001401274,0.000003285624,0.00002201348,0.01017674],"genre_scores_gemma":[0.9973071,0.0009156065,0.0001187344,0.001281789,0.0001199651,0.0001060641,0.00000280996,0.00001766736,0.0001302501],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3302019,"threshold_uncertainty_score":0.8058636,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.123535798155459,"gpt":0.4855899429385898,"spread":0.3620541447831308,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.01429529,0.0001201042,0.0003229305,0.001161952,0.0001001122,0.00006344214,0.0004467052,0.0001185079,0.0007917312],"category_scores_gemma":[0.004855001,0.000107924,0.00005385866,0.002662213,0.001031643,0.00036992,0.0002466021,0.003044453,0.00007764607],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002588099,"about_ca_system_score_gemma":0.0006174463,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006139844,"about_ca_topic_score_gemma":0.004095318,"domain_scores_codex":[0.9957696,0.0005774953,0.0005839352,0.000549109,0.001637739,0.0008821406],"domain_scores_gemma":[0.996697,0.00135051,0.00003275757,0.001015671,0.0007604128,0.0001436129],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002220089,0.0004045937,0.9180909,0.0001243333,0.00000124238,0.0000329211,0.001272911,0.000001540128,0.02359874,0.0005346597,0.0009232126,0.05479298],"study_design_scores_gemma":[0.001277435,0.00009551044,0.9713286,0.0001815461,0.000001773477,0.00001192425,0.001497628,0.001631728,0.005132189,0.002363446,0.01638483,0.00009340316],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9797937,0.001575361,8.3162e-7,0.006878617,0.00009954521,0.0008341175,0.000007001214,0.00001539109,0.01079543],"genre_scores_gemma":[0.9988093,0.0001274711,0.0001760627,0.00006891114,0.0001301336,0.0001022747,0.000007771475,0.00002260528,0.0005554959],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05469957,"threshold_uncertainty_score":0.9992555,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Randomized controlled trial; Psychological intervention; Weight loss; MEDLINE; Clinical trial; Dietary supplement","retraction":null,"screen_n_in":null,"score":{"opus":0.4723859013700117,"gpt":0.6036886997669948,"spread":0.1313027983969831,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06699459,0.0001537015,0.004993875,0.0006741619,0.00009544673,0.000006598644,0.0002116425,0.0001128805,0.002586523],"category_scores_gemma":[0.01841225,0.0001040838,0.001530627,0.0007640084,0.0006522695,0.00009588421,0.0001978932,0.0004723578,0.000003272614],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003418563,"about_ca_system_score_gemma":0.000480482,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001257959,"about_ca_topic_score_gemma":0.0001620224,"domain_scores_codex":[0.9873213,0.008289265,0.003127025,0.0002921982,0.0005949405,0.0003753041],"domain_scores_gemma":[0.983124,0.01490388,0.0005182208,0.0005253839,0.0008520914,0.00007641369],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.6222274,0.01039399,0.3079183,0.03064515,0.007598424,0.00003675571,0.0008670319,0.000005130514,0.00681739,0.0009882096,0.005537906,0.006964366],"study_design_scores_gemma":[0.9140363,0.0005413575,0.03734197,0.0280275,0.006653456,0.000006463451,0.0006829913,0.0004138038,0.008582266,0.002871095,0.0006554518,0.0001873146],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.941954,0.04300793,0.00007323216,0.004627337,0.000223562,0.008993104,0.00008419179,0.000009937408,0.001026759],"genre_scores_gemma":[0.9838897,0.01484604,0.0002522152,0.000237004,0.00003617705,0.0005946477,0.00005440063,0.000006041763,0.00008377802],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2918089,"threshold_uncertainty_score":0.9983252,"prediction_status":"machine_predicted_unvalidated"},"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"},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; Dalhousie University","funders":"","keywords":"Competition (biology); Narrative; Public relations; Political science; Psychology; Engineering ethics; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.8296990260947004,"gpt":0.7185244154404423,"spread":0.111174610654258,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2585603,0.0001684253,0.0007584988,0.001009862,0.0003178666,0.0001587018,0.0008749567,0.000754837,0.0002914224],"category_scores_gemma":[0.2471639,0.0001202967,0.0001061691,0.002282365,0.01713848,0.0002059253,0.002287151,0.02175216,0.00005801988],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001574058,"about_ca_system_score_gemma":0.00139587,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005636006,"about_ca_topic_score_gemma":0.05844948,"domain_scores_codex":[0.9831901,0.007337507,0.002254005,0.001254544,0.004529612,0.001434208],"domain_scores_gemma":[0.6186913,0.3693657,0.0001818333,0.003887445,0.006953075,0.0009207539],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002454126,0.0007846811,0.9562612,0.0005584138,0.00002195207,0.00002771441,0.00274602,8.231088e-7,0.001125669,0.03156234,0.0004358463,0.00402118],"study_design_scores_gemma":[0.004636155,0.0005386383,0.8931915,0.0006962901,0.000005721349,0.00001527623,0.008336089,0.001336165,0.000339468,0.08996605,0.0008293798,0.0001093181],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9326347,0.0002114393,0.000002392029,0.05624989,0.00009612403,0.001858664,0.000009444419,0.00001380324,0.008923507],"genre_scores_gemma":[0.9969847,0.001438884,0.0004847656,0.0001575439,0.0001827652,0.0001410098,0.000007882373,0.00002841299,0.0005740389],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3644989,"threshold_uncertainty_score":0.9855363,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}