{"meta":{"query_hash":"57e0b6922e06","filters":{"venue":"PhysioNet"},"cohort_total":17,"direct_labels_cover":0,"predictions_cover":17,"exported":17,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/57e0b6922e06","api":"https://metacan.xera.ac/api/v1/cohort?venue=PhysioNet"},"results":[{"id":"W6884851608","doi":"10.13026/1wf7-pm14","title":"Asclepius-R : Clinical Large Language Model Built On MIMIC-III Discharge Summaries","year":2023,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Usability; Resource (disambiguation); Language model; Benchmark (surveying); Data modeling","score_opus":0.06624195068223274,"score_gpt":0.3947683611672638,"score_spread":0.32852641048503106,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6884851608","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0032683953,0.00020001033,0.000008273073,0.00015373917,0.0014362017,0.0008399743,0.99312574,0.0008280398,0.00013961946],"genre_scores_gemma":[0.0010746762,0.00059644104,0.000059413305,0.00085716415,0.0020203318,0.00020139475,0.9923133,0.0005057608,0.0023714784],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9932975,0.00053834077,0.0013284641,0.0018305286,0.001541597,0.0014635606],"domain_scores_gemma":[0.99433905,0.0007504549,0.0007825882,0.0035068819,0.000127675,0.0004933508],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014076629,0.0011673435,0.0017241422,0.00047439936,0.00038860156,0.00019581387,0.0018468315,0.00089696824,0.00041520718],"category_scores_gemma":[0.00081406534,0.001045313,0.00081796694,0.00074006483,0.00026500528,0.00027631232,0.0015686675,0.0023189813,0.21456555],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005952444,0.0014575437,0.000010773869,0.00020366038,0.00027847395,0.000093519935,0.000104229,0.00026291967,0.00017638333,0.00013551945,0.9965583,0.00012340253],"study_design_scores_gemma":[0.0020179367,0.00021769125,0.0003667133,0.00040790628,0.0003270525,0.0000016680698,0.0000978875,0.005517733,0.00004514962,0.00055589946,0.9891565,0.0012878644],"about_ca_topic_score_codex":0.0014301192,"about_ca_topic_score_gemma":0.0024816324,"teacher_disagreement_score":0.21415034,"about_ca_system_score_codex":0.00016895114,"about_ca_system_score_gemma":0.00029576593,"threshold_uncertainty_score":0.9999827},"labels":[],"label_agreement":null},{"id":"W6884882244","doi":"10.13026/p8zc-cs44","title":"Annotation dataset of social determinants of health from MIMIC-III Clinical Care Database","year":2023,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Annotation; Social determinants of health; Health care; Agreement; Social care; MEDLINE; Social media; Dual (grammatical number)","score_opus":0.13220818322804898,"score_gpt":0.468883117840802,"score_spread":0.336674934612753,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6884882244","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014924532,0.0002631592,0.0000021159537,0.000014773797,0.0008401618,0.00063644064,0.9832636,0.00005457333,6.205189e-7],"genre_scores_gemma":[0.0012895192,0.00035519066,0.00017481718,0.0001008639,0.0006862434,0.000034883702,0.99721664,0.00013824596,0.000003591808],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9934413,0.0011295922,0.0027175043,0.0010693824,0.0011323178,0.0005099313],"domain_scores_gemma":[0.99301475,0.00054887,0.003991673,0.0019946024,0.00027864552,0.00017145883],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001126565,0.00053699716,0.0020599249,0.0003241635,0.00017505439,0.000019789515,0.0011561488,0.00038241714,0.00011378269],"category_scores_gemma":[0.0003648003,0.00055557024,0.0002878078,0.0006125121,0.00034717564,0.0002589304,0.00093143625,0.0006468906,0.0049297437],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003643628,0.00050457416,0.00008158012,0.00127534,0.00012856722,0.00001601366,0.00024088638,0.000003662514,0.00029117367,5.7122145e-7,0.99560606,0.0014871921],"study_design_scores_gemma":[0.0027845202,0.00032995798,0.016591372,0.0010271354,0.00047824826,8.697326e-7,0.00074638025,0.000109471395,0.00025430103,0.000041429263,0.9769242,0.00071213255],"about_ca_topic_score_codex":0.02969295,"about_ca_topic_score_gemma":0.006281891,"teacher_disagreement_score":0.023411058,"about_ca_system_score_codex":0.00012855863,"about_ca_system_score_gemma":0.0007699165,"threshold_uncertainty_score":0.9996896},"labels":[],"label_agreement":null},{"id":"W6884909244","doi":"10.13026/vc0e-th79","title":"WiDS (Women in Data Science) Datathon 2020: ICU Mortality Prediction","year":2020,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vector Institute; University of Toronto","funders":"","keywords":"Big data; Mortality rate; Intensive care unit; MEDLINE; Data source; Health care","score_opus":0.07746496477094239,"score_gpt":0.3559154355650454,"score_spread":0.278450470794103,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6884909244","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0028919447,0.000057216093,0.0000023947161,0.000054441236,0.0009098897,0.00071234483,0.99512,0.00018785216,0.000063905514],"genre_scores_gemma":[0.0013822343,0.00033116958,0.00006562767,0.00014084126,0.0011205407,0.000100336074,0.99675214,0.00008842773,0.00001866752],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9928774,0.00026249987,0.0007802586,0.0026587674,0.002257264,0.001163788],"domain_scores_gemma":[0.9924618,0.000058614358,0.00049788546,0.00629239,0.00010407173,0.0005852337],"candidate_categories":["metaepi_narrow","open_science","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0026766642,0.00061569526,0.00082545425,0.00043497028,0.00024230918,0.00025872255,0.005509618,0.00022217359,0.0005669446],"category_scores_gemma":[0.00092204125,0.000619651,0.00005054011,0.0034107994,0.00069153274,0.0026709498,0.005720089,0.0010820995,0.01605313],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007950831,0.00037250435,0.00008306358,0.00021482284,0.000048926107,0.000063338994,0.00010033255,0.000041782358,0.0030397833,0.0000036857173,0.9958556,0.00009661252],"study_design_scores_gemma":[0.0005651872,0.000080097954,0.01613795,0.00014399648,0.00010516154,0.000005215405,0.000078772435,0.0021163377,0.00003888227,0.00022060031,0.97995627,0.0005515448],"about_ca_topic_score_codex":0.0016604804,"about_ca_topic_score_gemma":0.0005456035,"teacher_disagreement_score":0.016054885,"about_ca_system_score_codex":0.00094741606,"about_ca_system_score_gemma":0.0011726379,"threshold_uncertainty_score":0.9998711},"labels":[],"label_agreement":null},{"id":"W6922363832","doi":"10.13026/zsgv-8w31","title":"Annotation dataset of social determinants of health from MIMIC-III Clinical Care Database","year":2024,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Historical Geopolitical and Social Dynamics","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Annotation; Social determinants of health; Health care; Agreement; Social care; MEDLINE; Social media; Dual (grammatical number)","score_opus":0.06688549351667016,"score_gpt":0.44420817155020814,"score_spread":0.377322678033538,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6922363832","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0018577046,0.0004901824,0.0000060993234,0.00008927498,0.0014199145,0.00025032152,0.9957805,0.000018089802,0.0000879133],"genre_scores_gemma":[0.0025451933,0.00065597106,0.000079499216,0.00018681759,0.0011581925,0.000008929314,0.99528337,0.000014633366,0.00006738932],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99648976,0.00073934445,0.0011800115,0.00046331715,0.0007567875,0.00037077704],"domain_scores_gemma":[0.9981911,0.00039129704,0.00066501537,0.00038783255,0.00017176423,0.00019294387],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00068240287,0.0002099523,0.0009496602,0.00007664116,0.00036082117,0.000021875749,0.0005647339,0.0003801351,0.00020603495],"category_scores_gemma":[0.00029835504,0.00020370868,0.00023747755,0.00032309364,0.0006908665,0.00010730734,0.00023603311,0.00045025066,0.00026262176],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000024626277,0.00019293322,0.000020674684,0.00044016208,0.000029908086,0.0000044919525,0.0025175144,1.6190909e-7,0.0000013402645,0.00051565445,0.9933165,0.002936072],"study_design_scores_gemma":[0.00019441721,0.000073978495,0.00020832023,0.00015899207,0.00011782186,3.2103785e-8,0.0019421958,0.000016046344,0.0000011481646,0.00097096607,0.9961253,0.00019079135],"about_ca_topic_score_codex":0.07089873,"about_ca_topic_score_gemma":0.014471113,"teacher_disagreement_score":0.056427613,"about_ca_system_score_codex":0.00015191274,"about_ca_system_score_gemma":0.00097083964,"threshold_uncertainty_score":0.93528825},"labels":[],"label_agreement":null},{"id":"W6922631198","doi":"10.13026/ckbf-pt70","title":"Asclepius-R : Clinical Large Language Model Built On MIMIC-III Discharge Summaries","year":2024,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Usability; Resource (disambiguation); Language model; Benchmark (surveying); Data modeling","score_opus":0.04527118146809302,"score_gpt":0.3859986484987003,"score_spread":0.3407274670306073,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6922631198","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002424475,0.0010386184,0.0000066857965,0.00019570549,0.0017593506,0.00078001333,0.9928059,0.0006088432,0.00038043017],"genre_scores_gemma":[0.0030073703,0.0005938714,0.00007466115,0.0010475212,0.002451061,0.0001693835,0.99046165,0.0004408184,0.0017536887],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99361646,0.00042862506,0.001286731,0.0019695212,0.0014114577,0.0012872044],"domain_scores_gemma":[0.9952141,0.00044713498,0.0005196224,0.0032447276,0.00010368625,0.00047072992],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0011577633,0.0012315256,0.0016254686,0.00046183722,0.00027347478,0.00031782757,0.0016327092,0.00088784826,0.0011889911],"category_scores_gemma":[0.00040131726,0.0010436261,0.00091145636,0.0005866785,0.00026648934,0.00027670883,0.0015749744,0.0029625339,0.24438642],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00053101114,0.0014968362,0.00000527157,0.00042647912,0.00035817953,0.00013127743,0.00014868892,0.00012377961,0.00020571136,0.00027559284,0.9961381,0.00015907342],"study_design_scores_gemma":[0.0011959439,0.00020686112,0.000051118623,0.00064592145,0.0005566545,0.0000034904274,0.000077468,0.0062485808,0.0000611645,0.0010109253,0.988752,0.0011898809],"about_ca_topic_score_codex":0.00041832638,"about_ca_topic_score_gemma":0.0010855888,"teacher_disagreement_score":0.24319743,"about_ca_system_score_codex":0.00018968327,"about_ca_system_score_gemma":0.00034438996,"threshold_uncertainty_score":0.99972403},"labels":[],"label_agreement":null},{"id":"W6941079881","doi":"10.13026/37yb-1t42","title":"Bridge2AI-Voice: An ethically-sourced, diverse voice dataset linked to health information","year":2025,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Mycorrhizal Fungi and Plant Interactions","field":"Agricultural and Biological Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Baycrest Hospital; Lunenfeld-Tanenbaum Research Institute; Trillium Health Centre; Université de Montréal; University of Toronto; SickKids Foundation; Dalhousie University; Hospital for Sick Children; Bridgepoint Active Healthcare; Simon Fraser University","funders":"","keywords":"Spectrogram; Voice analysis; Mood; Data collection; Big data","score_opus":0.022130786647277582,"score_gpt":0.2896273029604329,"score_spread":0.2674965163131553,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6941079881","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00085544545,0.000043305437,0.000005374011,0.003075318,0.0007165903,0.000496515,0.99466515,0.00006737626,0.00007492945],"genre_scores_gemma":[0.0000944219,0.0008481463,0.000041383002,0.014741153,0.00067041104,0.000028524737,0.9834371,6.6596795e-7,0.00013817888],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99813545,0.00021185228,0.00045540262,0.00040411667,0.00040615295,0.00038702326],"domain_scores_gemma":[0.9981018,0.00086956046,0.00026045964,0.00028836104,0.00011058584,0.00036923637],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00027693383,0.0003116413,0.00037733398,0.000062195926,0.00044744083,0.00021591678,0.0007592951,0.00030240443,0.0024713217],"category_scores_gemma":[0.00028873293,0.0001420184,0.00009770065,0.00047385067,0.000008158923,0.00050529715,0.00035620006,0.0007456161,0.005083065],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000081220234,0.00010626095,0.000007461717,0.00007130672,0.000017517545,0.0000019465328,0.000036625937,0.00003074496,0.00032865297,0.0000134070815,0.9938941,0.005410736],"study_design_scores_gemma":[0.00004447412,0.00028831643,0.0008319744,0.00018950264,0.000028039498,0.0000052884798,0.00008324904,0.000050733302,0.000004588103,0.000027622787,0.99814326,0.00030294128],"about_ca_topic_score_codex":0.024367316,"about_ca_topic_score_gemma":0.003535636,"teacher_disagreement_score":0.02083168,"about_ca_system_score_codex":0.00005969614,"about_ca_system_score_gemma":0.00005883684,"threshold_uncertainty_score":0.99844056},"labels":[],"label_agreement":null},{"id":"W6941313737","doi":"10.13026/cqt2-0b27","title":"MIMIC-IV on FHIR","year":2024,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Mycorrhizal Fungi and Plant Interactions","field":"Agricultural and Biological Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children","funders":"","keywords":"Interoperability; Terminology; Health care; ENCODE; Information system; Restructuring; Process (computing); Health informatics; Data warehouse","score_opus":0.013698811187891392,"score_gpt":0.22804762082628155,"score_spread":0.21434880963839015,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6941313737","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011241081,0.00022754334,1.0812873e-8,0.00057132466,0.0014195043,0.00011371452,0.99496067,0.000060081587,0.0015230386],"genre_scores_gemma":[0.00025736805,0.0012680629,0.0000014517782,0.0007125644,0.0015677206,0.000019504505,0.995073,8.3692805e-7,0.0010994509],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.999124,0.000033024004,0.00013749767,0.0003298264,0.00018829676,0.00018735747],"domain_scores_gemma":[0.9990838,0.0006819377,0.00005077207,0.00009123684,0.000017545435,0.000074661286],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.000039310464,0.00020174377,0.00017976132,0.000016989365,0.000103177685,0.000084747066,0.00027191345,0.00015082999,0.022061214],"category_scores_gemma":[0.00003260666,0.00006417474,0.00014370876,0.00017074392,0.0000034365887,0.000035232504,0.00007653461,0.00039750987,0.14435734],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020823967,0.00006805908,0.0000016038206,0.000021077034,0.000015925105,0.000015193321,0.0000017659275,0.0000014462686,0.0038198547,0.0000367597,0.99464357,0.0013539174],"study_design_scores_gemma":[0.0000068661006,0.00013034933,0.000033073124,0.00018752909,0.000028341763,0.000008227517,0.000006806093,0.0000131251645,0.00007614797,0.0006028084,0.9987192,0.00018750288],"about_ca_topic_score_codex":0.001120532,"about_ca_topic_score_gemma":0.0002996682,"teacher_disagreement_score":0.122296125,"about_ca_system_score_codex":0.0000156984,"about_ca_system_score_gemma":0.000005159894,"threshold_uncertainty_score":0.9788328},"labels":[],"label_agreement":null},{"id":"W6959883823","doi":"10.13026/249v-w155","title":"Bridge2AI-Voice: An ethically-sourced, diverse voice dataset linked to health information","year":2025,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Spectrogram; Voice analysis; Mood; Data collection; Big data","score_opus":0.0316463037866117,"score_gpt":0.35347112923944546,"score_spread":0.32182482545283375,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6959883823","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00013614791,0.000044401742,0.00007879345,0.0014510244,0.00081808615,0.0018290262,0.9952521,0.00035597908,0.000034442186],"genre_scores_gemma":[0.00003908453,0.00015611369,0.00043259194,0.02343855,0.00073891686,0.00012395339,0.974923,0.00006317487,0.00008462227],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9936355,0.00085278077,0.0013875671,0.0011271769,0.001844962,0.001152019],"domain_scores_gemma":[0.99304885,0.00029466036,0.0011335916,0.0039694314,0.0004231239,0.0011303445],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014561415,0.0010136629,0.0012212287,0.0011054071,0.00064925005,0.00045967594,0.0024604024,0.0007976527,0.00032943618],"category_scores_gemma":[0.00091208925,0.0010648613,0.00018217292,0.001662487,0.0001347413,0.0018405186,0.0020455215,0.0024900858,0.04794357],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00036132312,0.00037026405,0.0000011047057,0.0009714678,0.00010729595,0.000008003922,0.00032900958,0.000664305,0.00006098415,0.000012717316,0.99598044,0.001133095],"study_design_scores_gemma":[0.0010645246,0.0003395229,0.00087134854,0.00054153317,0.00016495532,0.000005568218,0.00013092751,0.00027970367,0.000005501016,0.00002463542,0.9955787,0.0009930485],"about_ca_topic_score_codex":0.013796632,"about_ca_topic_score_gemma":0.002436396,"teacher_disagreement_score":0.04761413,"about_ca_system_score_codex":0.0007122363,"about_ca_system_score_gemma":0.0011812324,"threshold_uncertainty_score":0.99981123},"labels":[],"label_agreement":null},{"id":"W6959989302","doi":"10.13026/3xt6-rf05","title":"Bridge2AI-Voice: An ethically-sourced, diverse voice dataset linked to health information","year":2025,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Plant pathogens and resistance mechanisms","field":"Agricultural and Biological Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; University of Toronto; SickKids Foundation; Dalhousie University; Hospital for Sick Children; Simon Fraser University","funders":"","keywords":"Spectrogram; Voice analysis; Mood; Data collection; Human voice","score_opus":0.02480228992026199,"score_gpt":0.2753309313074211,"score_spread":0.2505286413871591,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6959989302","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0020867137,0.000042930624,0.00001298791,0.0018736678,0.00042094605,0.0005066664,0.9949797,0.00006123933,0.0000151975855],"genre_scores_gemma":[0.00017451806,0.0003615922,0.000093438146,0.009959249,0.000469435,0.000025599102,0.988809,7.376213e-7,0.000106470136],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9979077,0.00019773043,0.00044097123,0.00047512032,0.0005193838,0.00045906825],"domain_scores_gemma":[0.99882275,0.00012813673,0.0002917216,0.00030002042,0.00009187113,0.00036551908],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00039714912,0.00033914897,0.0004486517,0.00005038348,0.00047224687,0.00017960957,0.00085738726,0.00035588906,0.00018478748],"category_scores_gemma":[0.00008109903,0.00014298013,0.000095659096,0.0004585249,0.000028898408,0.00029996235,0.00041205494,0.00052721176,0.00053937716],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000061619714,0.00009681828,6.117154e-7,0.00009412338,0.000013214968,0.0000049730247,0.000021621467,0.00001983622,0.0043674987,0.000011443276,0.9893216,0.0059866463],"study_design_scores_gemma":[0.00011359358,0.0002973301,0.0016305881,0.00015950117,0.000024740419,0.0000027800893,0.00009815842,0.000028391338,0.000021167598,0.000030994626,0.9972495,0.00034329598],"about_ca_topic_score_codex":0.0024811425,"about_ca_topic_score_gemma":0.0059890575,"teacher_disagreement_score":0.0080855815,"about_ca_system_score_codex":0.00006204317,"about_ca_system_score_gemma":0.00006087005,"threshold_uncertainty_score":0.69327825},"labels":[],"label_agreement":null},{"id":"W6960159768","doi":"10.13026/s5rz-1j65","title":"Asclepius-R : Clinical Large Language Model Built On MIMIC-III Discharge Summaries","year":2024,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Phytochemistry and Biological Activities","field":"Agricultural and Biological Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Usability; Resource (disambiguation); Language model; Benchmark (surveying); Data modeling","score_opus":0.04396920669721984,"score_gpt":0.31361604946006827,"score_spread":0.26964684276284845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6960159768","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10364128,0.00032752295,1.7087649e-7,0.0005409903,0.00034924658,0.0002009574,0.89444625,0.00012249225,0.00037105993],"genre_scores_gemma":[0.1257484,0.0005988883,0.0000046805453,0.000951977,0.0020114777,0.000052166186,0.8696866,0.0000022319905,0.0009435596],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99775213,0.0001181844,0.00042804104,0.0008356194,0.00033819332,0.00052781904],"domain_scores_gemma":[0.9989577,0.00043751017,0.00016712844,0.00023109152,0.000030182251,0.00017642124],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00028335484,0.00046504944,0.0006229211,0.000011933705,0.00023100473,0.00013614986,0.0006939378,0.0005492322,0.0009856852],"category_scores_gemma":[0.0001037176,0.00015430339,0.0004450419,0.00018402339,0.00015048431,0.00008938808,0.00051570457,0.0009452241,0.0018302425],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015010608,0.0005846491,0.000009993785,0.000062663894,0.00004639727,0.00001652353,0.000010105513,0.0000021524347,0.009598599,0.000028073997,0.98662037,0.0028703944],"study_design_scores_gemma":[0.00013330513,0.00037483042,0.00031441767,0.00015254017,0.000057303656,0.0000017048473,0.00008457907,0.00014523783,0.0013218774,0.0015340743,0.99538046,0.00049966335],"about_ca_topic_score_codex":0.00011971082,"about_ca_topic_score_gemma":0.00029338663,"teacher_disagreement_score":0.024759658,"about_ca_system_score_codex":0.000023581375,"about_ca_system_score_gemma":0.000018697885,"threshold_uncertainty_score":0.9999275},"labels":[],"label_agreement":null},{"id":"W7080721321","doi":"10.13026/5b8h-jx61","title":"mcPHASES: A Dataset of Physiological, Hormonal, and Self-reported Events and Symptoms for Menstrual Health Tracking with Wearables","year":2025,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Menstrual cycle; Distress; Wearable computer; Wearable technology; Activity tracker; Reproductive health; Tracking (education)","score_opus":0.024297886433895258,"score_gpt":0.28948594551326184,"score_spread":0.2651880590793666,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7080721321","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0026813787,0.0006098141,0.0010114029,0.0003446145,0.0000567503,0.00052770815,0.99471414,0.000048245995,0.0000059361837],"genre_scores_gemma":[0.0043568956,0.0007187649,0.0038145296,0.00029422084,0.00005044973,0.00006967592,0.9906699,0.0000026585078,0.000022896564],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9981849,0.00006852803,0.00042805047,0.00076404616,0.00022793419,0.00032654914],"domain_scores_gemma":[0.9983258,0.00018038525,0.0005704753,0.0007069254,0.00010489699,0.000111472495],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002714957,0.00031187548,0.0006061285,0.00007099812,0.00020805132,0.000053988726,0.00052100624,0.00013475816,0.0000051495635],"category_scores_gemma":[0.000072747986,0.00023261362,0.000041990428,0.00020661227,0.000097184886,0.00016400042,0.0004830603,0.0001900527,2.4944876e-7],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003263368,0.00026175947,0.000007952465,0.0011967829,0.00020239134,0.0000058237615,0.000031516556,0.000025725085,0.0002636938,0.00005501986,0.9966706,0.0012460982],"study_design_scores_gemma":[0.0014748635,0.0014208407,0.00082077,0.0010103906,0.0001611671,0.000066427696,0.00004202589,0.0025611892,0.00027249265,0.0020222303,0.98956406,0.00058354094],"about_ca_topic_score_codex":0.00011218094,"about_ca_topic_score_gemma":0.000007628479,"teacher_disagreement_score":0.0071065417,"about_ca_system_score_codex":0.000015519032,"about_ca_system_score_gemma":0.00019490563,"threshold_uncertainty_score":0.9485708},"labels":[],"label_agreement":null},{"id":"W7080721330","doi":"10.13026/zx6a-2c81","title":"mcPHASES: A Dataset of Physiological, Hormonal, and Self-reported Events and Symptoms for Menstrual Health Tracking with Wearables","year":2025,"lang":"en","type":"dataset","venue":"PhysioNet","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Menstrual cycle; Distress; Wearable computer; Wearable technology; Activity tracker; Reproductive health; Tracking (education)","score_opus":0.024297886433895258,"score_gpt":0.28948594551326184,"score_spread":0.2651880590793666,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7080721330","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0026813787,0.0006098141,0.0010114029,0.0003446145,0.0000567503,0.00052770815,0.99471414,0.000048245995,0.0000059361837],"genre_scores_gemma":[0.0043568956,0.0007187649,0.0038145296,0.00029422084,0.00005044973,0.00006967592,0.9906699,0.0000026585078,0.000022896564],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9981849,0.00006852803,0.00042805047,0.00076404616,0.00022793419,0.00032654914],"domain_scores_gemma":[0.9983258,0.00018038525,0.0005704753,0.0007069254,0.00010489699,0.000111472495],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002714957,0.00031187548,0.0006061285,0.00007099812,0.00020805132,0.000053988726,0.00052100624,0.00013475816,0.0000051495635],"category_scores_gemma":[0.000072747986,0.00023261362,0.000041990428,0.00020661227,0.000097184886,0.00016400042,0.0004830603,0.0001900527,2.4944876e-7],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003263368,0.00026175947,0.000007952465,0.0011967829,0.00020239134,0.0000058237615,0.000031516556,0.000025725085,0.0002636938,0.00005501986,0.9966706,0.0012460982],"study_design_scores_gemma":[0.0014748635,0.0014208407,0.00082077,0.0010103906,0.0001611671,0.000066427696,0.00004202589,0.0025611892,0.00027249265,0.0020222303,0.98956406,0.00058354094],"about_ca_topic_score_codex":0.00011218094,"about_ca_topic_score_gemma":0.000007628479,"teacher_disagreement_score":0.0071065417,"about_ca_system_score_codex":0.000015519032,"about_ca_system_score_gemma":0.00019490563,"threshold_uncertainty_score":0.9485708},"labels":[],"label_agreement":null},{"id":"W7115697022","doi":"10.13026/k81f-qr68","title":"Bridge2AI-Voice: An ethically-sourced, diverse voice dataset linked to health information","year":2025,"lang":"","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Lunenfeld-Tanenbaum Research Institute; Université de Montréal; University of Toronto; SickKids Foundation; Dalhousie University; Hospital for Sick Children; Simon Fraser University","funders":"","keywords":"Spectrogram; Voice analysis; Mood; Data collection; Big data","score_opus":0.03248821624541169,"score_gpt":0.35334038130166123,"score_spread":0.3208521650562495,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7115697022","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00068873476,0.00017812062,0.0004548254,0.004030309,0.0024317496,0.005940681,0.98572886,0.00049488846,0.000051830106],"genre_scores_gemma":[0.0005333649,0.0010392161,0.00106159,0.044186976,0.0017386546,0.00029455335,0.9507969,0.00019339444,0.00015532388],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9803613,0.003478069,0.004660138,0.0032772692,0.004796919,0.0034262969],"domain_scores_gemma":[0.9797586,0.0009385043,0.00394703,0.009805177,0.0014768365,0.0040738457],"candidate_categories":["metaepi_narrow","sts","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.0046349065,0.0030878996,0.0034529553,0.0029506325,0.0026201424,0.0017851477,0.0064909076,0.002195888,0.0012370485],"category_scores_gemma":[0.002726389,0.0035489576,0.000565037,0.0054936125,0.00057082425,0.006466971,0.0070847725,0.0070761777,0.15086699],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0022049102,0.0019102023,0.0000045686375,0.0036802315,0.00045943796,0.000025288244,0.0020314003,0.004091185,0.00026782815,0.00004858818,0.98040044,0.004875899],"study_design_scores_gemma":[0.0035953654,0.0016459252,0.0022377195,0.0023966972,0.00070237,0.000020182533,0.0008102097,0.0031681373,0.000017510247,0.000045971065,0.98230773,0.003052158],"about_ca_topic_score_codex":0.017310722,"about_ca_topic_score_gemma":0.0028615694,"teacher_disagreement_score":0.14962994,"about_ca_system_score_codex":0.002360609,"about_ca_system_score_gemma":0.0046259537,"threshold_uncertainty_score":0.9996759},"labels":[],"label_agreement":null},{"id":"W7115910046","doi":"10.13026/mf9s-5r03","title":"Bridge2AI-Voice Pediatric Dataset","year":2025,"lang":"","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Baycrest Hospital; Sinai Health System; Lunenfeld-Tanenbaum Research Institute; Mount Sinai Hospital; SickKids Foundation; University of Toronto; Dalhousie University; Hospital for Sick Children; Bridgepoint Active Healthcare; Simon Fraser University","funders":"","keywords":"Spectrogram; Measure (data warehouse); Data collection; Mood; Voice analysis","score_opus":0.017295758935052686,"score_gpt":0.3052028607778872,"score_spread":0.2879071018428345,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7115910046","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00015608351,0.0050079967,0.000011824148,0.00017501054,0.004779714,0.002178297,0.9872102,0.00027249,0.00020838722],"genre_scores_gemma":[0.00015580478,0.010212256,0.000112556656,0.0014303783,0.008226065,0.00029553846,0.9779612,0.0002948125,0.0013114068],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98486686,0.0014108953,0.00270816,0.004846495,0.0032158333,0.0029517463],"domain_scores_gemma":[0.98351926,0.0012587064,0.002420549,0.0110731,0.00055137865,0.0011770121],"candidate_categories":["metaepi_narrow","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.0012242331,0.0032705083,0.0030864903,0.0024029412,0.0012281412,0.0007993489,0.0068257707,0.0014082897,0.0056118667],"category_scores_gemma":[0.0011866485,0.0037165429,0.00080686726,0.0069144373,0.00044969167,0.0011986556,0.006999074,0.0039671897,0.20767331],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004869993,0.0031289808,0.000024558716,0.0039401995,0.000542726,0.0002318659,0.000019956238,0.00019828875,0.00021528779,0.000018462553,0.99091005,0.00028259747],"study_design_scores_gemma":[0.002305227,0.00023871516,0.0008262587,0.0004788315,0.0051064435,0.000025538206,0.000020401756,0.00031561375,0.00006837962,0.000099926954,0.9872753,0.0032393767],"about_ca_topic_score_codex":0.0019030764,"about_ca_topic_score_gemma":0.00016521047,"teacher_disagreement_score":0.20206144,"about_ca_system_score_codex":0.0007474229,"about_ca_system_score_gemma":0.0025239515,"threshold_uncertainty_score":0.99988806},"labels":[],"label_agreement":null},{"id":"W7115911116","doi":"10.13026/y7mp-eh56","title":"Bridge2AI-Voice Pediatric Dataset","year":2025,"lang":"","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sinai Health System; Lunenfeld-Tanenbaum Research Institute; Mount Sinai Hospital; SickKids Foundation; University of Toronto; Dalhousie University; Hospital for Sick Children; Simon Fraser University","funders":"","keywords":"Spectrogram; Measure (data warehouse); Data collection; Mood; Voice analysis","score_opus":0.017295758935052686,"score_gpt":0.3052028607778872,"score_spread":0.2879071018428345,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7115911116","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00015608351,0.0050079967,0.000011824148,0.00017501054,0.004779714,0.002178297,0.9872102,0.00027249,0.00020838722],"genre_scores_gemma":[0.00015580478,0.010212256,0.000112556656,0.0014303783,0.008226065,0.00029553846,0.9779612,0.0002948125,0.0013114068],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98486686,0.0014108953,0.00270816,0.004846495,0.0032158333,0.0029517463],"domain_scores_gemma":[0.98351926,0.0012587064,0.002420549,0.0110731,0.00055137865,0.0011770121],"candidate_categories":["metaepi_narrow","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.0012242331,0.0032705083,0.0030864903,0.0024029412,0.0012281412,0.0007993489,0.0068257707,0.0014082897,0.0056118667],"category_scores_gemma":[0.0011866485,0.0037165429,0.00080686726,0.0069144373,0.00044969167,0.0011986556,0.006999074,0.0039671897,0.20767331],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004869993,0.0031289808,0.000024558716,0.0039401995,0.000542726,0.0002318659,0.000019956238,0.00019828875,0.00021528779,0.000018462553,0.99091005,0.00028259747],"study_design_scores_gemma":[0.002305227,0.00023871516,0.0008262587,0.0004788315,0.0051064435,0.000025538206,0.000020401756,0.00031561375,0.00006837962,0.000099926954,0.9872753,0.0032393767],"about_ca_topic_score_codex":0.0019030764,"about_ca_topic_score_gemma":0.00016521047,"teacher_disagreement_score":0.20206144,"about_ca_system_score_codex":0.0007474229,"about_ca_system_score_gemma":0.0025239515,"threshold_uncertainty_score":0.99988806},"labels":[],"label_agreement":null},{"id":"W7123274677","doi":"10.13026/pk42-4v91","title":"Lunguage: A Benchmark for Structured and Sequential Chest X-ray Interpretation","year":2025,"lang":"","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Benchmark (surveying); Pairwise comparison; SNOMED CT; Semantics (computer science); Interpretation (philosophy); Gold standard (test); Resource (disambiguation)","score_opus":0.00878072575010086,"score_gpt":0.2949954740640469,"score_spread":0.28621474831394605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7123274677","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0022207347,0.0017733045,0.0011262251,0.0000801385,0.0028785118,0.004567079,0.9871877,0.00010184157,0.00006448022],"genre_scores_gemma":[0.08339707,0.00028173346,0.0017753071,0.0002433343,0.0012306533,0.000679754,0.91191965,0.00015293327,0.0003195427],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9934158,0.00053516106,0.0014676172,0.0024949708,0.00093043747,0.0011560173],"domain_scores_gemma":[0.9950029,0.00064589386,0.0014164548,0.0019571076,0.0005658505,0.00041177735],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.0005751174,0.0016970092,0.0017761408,0.00089023996,0.00071456237,0.00070409727,0.0011490161,0.0009597086,0.0012911991],"category_scores_gemma":[0.0008585011,0.0018235565,0.00055092,0.0010049217,0.00052679924,0.0007148052,0.0010993487,0.0011733067,0.00016283529],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0029825938,0.00032951796,0.000002500306,0.005131118,0.0010268259,0.0000125936,0.00091619825,0.000408486,0.036973413,0.00007491051,0.94713277,0.0050090966],"study_design_scores_gemma":[0.006744603,0.0008276976,0.0010949953,0.0032188178,0.0049947845,0.000018425884,0.00048229296,0.0351202,0.0022286945,0.002659751,0.9396658,0.002943913],"about_ca_topic_score_codex":0.0002663034,"about_ca_topic_score_gemma":0.0002618098,"teacher_disagreement_score":0.08117633,"about_ca_system_score_codex":0.0004179363,"about_ca_system_score_gemma":0.0008074968,"threshold_uncertainty_score":0.99962175},"labels":[],"label_agreement":null},{"id":"W7123307329","doi":"10.13026/t075-g517","title":"Lunguage: A Benchmark for Structured and Sequential Chest X-ray Interpretation","year":2025,"lang":"","type":"dataset","venue":"PhysioNet","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Benchmark (surveying); Pairwise comparison; SNOMED CT; Semantics (computer science); Interpretation (philosophy); Gold standard (test); Resource (disambiguation)","score_opus":0.00878072575010086,"score_gpt":0.2949954740640469,"score_spread":0.28621474831394605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7123307329","genre_codex":"dataset","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":"dataset","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0022207347,0.0017733045,0.0011262251,0.0000801385,0.0028785118,0.004567079,0.9871877,0.00010184157,0.00006448022],"genre_scores_gemma":[0.08339707,0.00028173346,0.0017753071,0.0002433343,0.0012306533,0.000679754,0.91191965,0.00015293327,0.0003195427],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9934158,0.00053516106,0.0014676172,0.0024949708,0.00093043747,0.0011560173],"domain_scores_gemma":[0.9950029,0.00064589386,0.0014164548,0.0019571076,0.0005658505,0.00041177735],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.0005751174,0.0016970092,0.0017761408,0.00089023996,0.00071456237,0.00070409727,0.0011490161,0.0009597086,0.0012911991],"category_scores_gemma":[0.0008585011,0.0018235565,0.00055092,0.0010049217,0.00052679924,0.0007148052,0.0010993487,0.0011733067,0.00016283529],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0029825938,0.00032951796,0.000002500306,0.005131118,0.0010268259,0.0000125936,0.00091619825,0.000408486,0.036973413,0.00007491051,0.94713277,0.0050090966],"study_design_scores_gemma":[0.006744603,0.0008276976,0.0010949953,0.0032188178,0.0049947845,0.000018425884,0.00048229296,0.0351202,0.0022286945,0.002659751,0.9396658,0.002943913],"about_ca_topic_score_codex":0.0002663034,"about_ca_topic_score_gemma":0.0002618098,"teacher_disagreement_score":0.08117633,"about_ca_system_score_codex":0.0004179363,"about_ca_system_score_gemma":0.0008074968,"threshold_uncertainty_score":0.99962175},"labels":[],"label_agreement":null}]}