{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":7,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":7,"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","author_layer_release":"2026-06-26"},"query_hash":"3e471ffbb115","filters":{"venue":"Psychological Test Adaptation and Development"}},"results":[{"id":"W4396520795","doi":"10.1027/2698-1866/a000061","title":"Structural Validity Evidence for the Oxford Utilitarianism Scale Across 15 Languages","year":2024,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Psychology of Moral and Emotional Judgment","field":"Neuroscience","cited_by":10,"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":"Scale (ratio); Utilitarianism; Geography; Philosophy; Epistemology; Cartography","authors":[{"name":"Briana Oshiro","is_ca":true},{"name":"William H.B. McAuliffe","is_ca":false},{"name":"Raymond Luong","is_ca":true},{"name":"Anabela Caetano Santos","is_ca":false},{"name":"Andrej Findor","is_ca":false},{"name":"A Kuźmińska","is_ca":false},{"name":"Anthony Lantian","is_ca":false},{"name":"Asil Ali Özdoğru","is_ca":false},{"name":"Balázs Aczél","is_ca":false},{"name":"Bojana M. Dinić","is_ca":false},{"name":"Christopher R. Chartier","is_ca":false},{"name":"Jasper Hidding","is_ca":false},{"name":"Job de Grefte","is_ca":false},{"name":"John Protzko","is_ca":false},{"name":"Mairead Shaw","is_ca":true},{"name":"Maximilian Primbs","is_ca":false},{"name":"Nicholas A. Coles","is_ca":false},{"name":"Patrí­cia Arriaga","is_ca":false},{"name":"Patrick S. Forscher","is_ca":false},{"name":"Savannah C Lewis","is_ca":false},{"name":"Tamás Nagy","is_ca":false},{"name":"Wieteke C. de Vries","is_ca":false},{"name":"William Jiménez‐Leal","is_ca":false},{"name":"Yansong Li","is_ca":false},{"name":"Jessica Kay Flake","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4732301068642866,"gpt":0.4263292116603027,"spread":0.04690089520398383,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02688381,0.0004918409,0.0005185577,0.002622271,0.0007720157,0.002055092,0.0006289299,0.0004107477,0.002884532],"category_scores_gemma":[0.06059569,0.0003914092,0.001021639,0.002663232,0.002964546,0.001646477,0.00311897,0.001066331,0.0003197287],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009662962,"about_ca_system_score_gemma":0.00193397,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002015355,"about_ca_topic_score_gemma":0.002994488,"domain_scores_codex":[0.9862102,0.006220972,0.001815331,0.00136576,0.003982952,0.0004047882],"domain_scores_gemma":[0.9256166,0.04004423,0.009911331,0.005876284,0.01702774,0.001523733],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005686674,0.0002056872,0.9029346,0.0003537541,0.0004978122,0.0001235708,0.02128861,0.0005725626,0.002218152,0.002797582,0.001136778,0.06730236],"study_design_scores_gemma":[0.00007056526,0.0003523611,0.9771525,0.0003740228,0.0001704141,0.0002259256,0.01105788,0.001549467,0.001971325,0.003136048,0.003878371,0.00006106705],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9897351,0.0002136254,0.002577853,0.0002289688,0.00006055866,0.0001042159,0.000323368,0.00002144044,0.00673481],"genre_scores_gemma":[0.9965714,0.00007323238,0.002554426,0.00003896801,0.000009903348,0.000121909,0.0003344049,0.00002054698,0.0002751632],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02688381,"threshold_uncertainty_score":0.1421769,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4376645007","doi":"10.1027/2698-1866/a000039","title":"Effects of a Neutral Response Option on Occupational Interest Circumplex Fit","year":2023,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Psychometric Methodologies and Testing","field":"Decision 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":"Toronto Metropolitan University; Wilfrid Laurier University","funders":"","keywords":"Generalizability theory; Equivalence (formal languages); Psychology; Econometrics; Correlation; Statistics; Sample (material); Structural equation modeling; Hexagonal crystal system; Social psychology; Mathematics; Pure mathematics","authors":[{"name":"Sabah Rasheed","is_ca":true},{"name":"Chet Robie","is_ca":true},{"name":"Peter A. Fisher","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7527729537058937,"gpt":0.5201070910133551,"spread":0.2326658626925386,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06812744,0.000825293,0.0006873286,0.0007774123,0.0009404141,0.002136352,0.001397203,0.001398036,0.01072893],"category_scores_gemma":[0.3016184,0.0004396357,0.001216574,0.0006843964,0.002461886,0.002363079,0.002470782,0.00212393,0.001004406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001247251,"about_ca_system_score_gemma":0.000861011,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006352025,"about_ca_topic_score_gemma":0.0005420638,"domain_scores_codex":[0.8849605,0.09544495,0.005600883,0.005269474,0.007893394,0.0008308515],"domain_scores_gemma":[0.3770874,0.5786052,0.01472093,0.01947189,0.007649984,0.002464542],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.04440737,0.006739254,0.7393859,0.0007905637,0.001444652,0.0005701365,0.01647078,0.006694564,0.0196698,0.009250382,0.002662528,0.1519141],"study_design_scores_gemma":[0.002393909,0.01963906,0.8723097,0.0007628432,0.001018359,0.0008916967,0.007312662,0.04948578,0.02837339,0.01036029,0.007029381,0.0004230127],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9920679,0.00005553618,0.004973794,0.000198775,0.00007260684,0.0001818951,0.00007983128,0.00004167544,0.002327848],"genre_scores_gemma":[0.9949626,0.00001547144,0.004064784,0.0000821974,0.00001238021,0.0003785045,0.0001080123,0.00002868398,0.0003475096],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06812744,"threshold_uncertainty_score":0.3602967,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4410504341","doi":"10.1027/2698-1866/a000100","title":"Cross-Cultural Adaptation of the Short Dark Tetrad (SD4) in the Brazilian Context","year":2025,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Digital Games and Media","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":"Western University","funders":"","keywords":"Adaptation (eye); Tetrad; Context (archaeology); Business; Psychology; Geography; Biology; Botany","authors":[{"name":"Carollina Souza Guilhermino","is_ca":false},{"name":"Bruno Bonfá-Araújo","is_ca":true},{"name":"Tiago Geraldo de Azevedo","is_ca":false},{"name":"Nelson Hauck Filho","is_ca":false},{"name":"Paulo Felipe Ribeiro Bandeira","is_ca":false},{"name":"Celso Francisco Tondin","is_ca":false},{"name":"Marco Antônio Silva Alvarenga","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08481528937089738,"gpt":0.3836462200946721,"spread":0.2988309307237747,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004032262,0.0003405295,0.0003313178,0.001296843,0.0006445684,0.0006909684,0.0004233565,0.0001968165,0.001580036],"category_scores_gemma":[0.01260686,0.0001960581,0.0005837749,0.0009436251,0.0006144415,0.0004661192,0.001196255,0.0004282697,0.0001832984],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006684831,"about_ca_system_score_gemma":0.001125639,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01272688,"about_ca_topic_score_gemma":0.02835163,"domain_scores_codex":[0.9984994,0.000676429,0.0001981655,0.0001653958,0.0003358308,0.0001248611],"domain_scores_gemma":[0.9962978,0.001473793,0.0005747498,0.0004506403,0.000970954,0.0002321061],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001257235,0.0002744051,0.890649,0.0001759698,0.0001065486,0.0002254484,0.0132073,0.0001997202,0.001384446,0.001175356,0.001023892,0.09145217],"study_design_scores_gemma":[0.00002228599,0.0002767917,0.9789327,0.0002143559,0.00005375507,0.0004404173,0.009272865,0.0007917781,0.0007530598,0.001005541,0.008213889,0.00002258196],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9941763,0.0002219912,0.001093863,0.0001840644,0.00004857395,0.000133103,0.0002592933,0.000009883767,0.003872888],"genre_scores_gemma":[0.9965648,0.0001960961,0.002295686,0.00005474552,0.000007425439,0.0001596132,0.0002795708,0.00001012383,0.0004318987],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01272688,"threshold_uncertainty_score":0.02530557,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4308568240","doi":"10.1027/2698-1866/a000025","title":"The German Version of the Hybrid Work Characteristics Scale","year":2022,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Job Satisfaction and Organizational Behavior","field":"Business, Management and Accounting","cited_by":3,"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":"Universität Bielefeld","keywords":"German; Human multitasking; Scale (ratio); Computer science; Reliability (semiconductor); Work (physics); Task (project management); Psychology; Cognitive psychology; Engineering; Linguistics","authors":[{"name":"Agnieszka Paruzel","is_ca":false},{"name":"Anja‐Kristin Abendroth","is_ca":false},{"name":"Axel Mayer","is_ca":false},{"name":"Jia Lin Xie","is_ca":true},{"name":"Günter W. Maier","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02834803613470245,"gpt":0.2513507258479564,"spread":0.2230026897132539,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001200283,0.000462879,0.0003428001,0.00112807,0.000354376,0.0005659207,0.0004389083,0.000336143,0.01413864],"category_scores_gemma":[0.003968749,0.0001725377,0.0004811385,0.001033747,0.0002505565,0.0004970436,0.0008453127,0.0006595065,0.002253157],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007003847,"about_ca_system_score_gemma":0.0006057148,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002570056,"about_ca_topic_score_gemma":0.004451844,"domain_scores_codex":[0.9991744,0.0002231113,0.0001506906,0.00008688729,0.0002884088,0.00007646235],"domain_scores_gemma":[0.9982548,0.0004712841,0.000425025,0.0001254114,0.0005577301,0.0001656082],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009480027,0.0009830949,0.4278584,0.001303461,0.0005609011,0.0004033255,0.004124822,0.002072708,0.003779373,0.007803786,0.1316163,0.4185458],"study_design_scores_gemma":[0.000158964,0.0004328265,0.9163669,0.0002066836,0.00009799055,0.0003673859,0.001169108,0.0008290302,0.0007514927,0.001804552,0.07774703,0.00006797544],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8916637,0.001913881,0.009460752,0.002184163,0.0004501103,0.003865879,0.04645833,0.0003302673,0.04367289],"genre_scores_gemma":[0.9146575,0.001513627,0.02230176,0.000726221,0.0001071002,0.005574125,0.02973722,0.00008745459,0.02529496],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01413864,"threshold_uncertainty_score":0.04729843,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4385879749","doi":"10.1027/2698-1866/a000050","title":"Psychometric Properties of the Brazilian Portuguese Version of the Circumplex Scales of Interpersonal Problems (CSIP)","year":2023,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Child and Adolescent Psychosocial and Emotional 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":"McMaster University","funders":"","keywords":"Interpersonal communication; Portuguese; Construct (python library); Checklist; Construct validity; Scale (ratio); Concurrent validity; Psychometrics; Interpersonal relationship; Psychology; Clinical psychology; Social psychology; Computer science; Geography; Cognitive psychology; Internal consistency; Cartography","authors":[{"name":"Vinícius Betzel Koehler","is_ca":true},{"name":"Michael J. Boudreaux","is_ca":false},{"name":"Rosana Suemi Tokumaru","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08830546072002736,"gpt":0.2925175832569814,"spread":0.204212122536954,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006417145,0.0005109693,0.0005476598,0.001857154,0.0007481079,0.0009366713,0.0004512967,0.0003526255,0.002104895],"category_scores_gemma":[0.02654781,0.0003084622,0.0007941202,0.001722205,0.0006161671,0.0005660779,0.001252454,0.0006243144,0.0002407103],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005885653,"about_ca_system_score_gemma":0.001451571,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006817392,"about_ca_topic_score_gemma":0.009679189,"domain_scores_codex":[0.9971709,0.001032654,0.0003806855,0.0002599248,0.001018521,0.0001373951],"domain_scores_gemma":[0.9921554,0.003349254,0.001525385,0.0005190343,0.001994345,0.0004566665],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002267062,0.0002997584,0.8381688,0.0005789903,0.0004173334,0.0002776129,0.006332989,0.0006418374,0.001349924,0.00170156,0.003393797,0.1466107],"study_design_scores_gemma":[0.00004560818,0.0002456376,0.9857853,0.0004052931,0.0001449016,0.000732813,0.002118393,0.001053481,0.0006609768,0.0009879945,0.007788641,0.00003085577],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.970248,0.002279372,0.004128865,0.00072043,0.0001777378,0.0006518596,0.002181977,0.00004135631,0.01957035],"genre_scores_gemma":[0.9875662,0.001230926,0.007725422,0.0001413408,0.00004043419,0.000553332,0.001499656,0.00003141555,0.001211326],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006817392,"threshold_uncertainty_score":0.03393751,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4401035029","doi":"10.1027/2698-1866/a000078","title":"Validation of the German Intimacy Scale","year":2024,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Family Support in Illness","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Saint Mary's University","funders":"","keywords":"German; Scale (ratio); Psychology; Computer science; Geography; Archaeology; Cartography","authors":[{"name":"Sonja Weilenmann","is_ca":false},{"name":"Alexander von Luckner","is_ca":false},{"name":"Charlotte Huber","is_ca":false},{"name":"Carolyn Birnie‐Porter","is_ca":true},{"name":"Monique C. Pfaltz","is_ca":false},{"name":"Billy Jansson","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07978204604697928,"gpt":0.3682736061917756,"spread":0.2884915601447963,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006678351,0.0003827879,0.0003408611,0.001449747,0.0006714026,0.0006143911,0.0005450231,0.0003974938,0.003992447],"category_scores_gemma":[0.01327597,0.0001774486,0.0004817863,0.0007679781,0.0007279745,0.0005154219,0.001636959,0.000655847,0.001061859],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000891539,"about_ca_system_score_gemma":0.001006177,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001850197,"about_ca_topic_score_gemma":0.002746108,"domain_scores_codex":[0.9971173,0.001193605,0.0003709756,0.0002208548,0.0009431323,0.0001541819],"domain_scores_gemma":[0.9953905,0.001638051,0.0005460735,0.0006469523,0.001450309,0.0003281621],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009595099,0.001475889,0.5444776,0.0007350576,0.0003089998,0.0007220398,0.01974515,0.001260784,0.00800092,0.01098182,0.02397239,0.3873599],"study_design_scores_gemma":[0.000273978,0.001415485,0.9237438,0.0003512052,0.0001201395,0.001271465,0.005826331,0.002158176,0.004629829,0.002519527,0.05759193,0.0000981712],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.9770486,0.0007263033,0.005079325,0.0006392424,0.0001725357,0.001433592,0.001881789,0.00005606031,0.01296256],"genre_scores_gemma":[0.97645,0.0006138611,0.01309002,0.000362787,0.00005248932,0.002747753,0.002806678,0.00004996339,0.003826428],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.006678351,"threshold_uncertainty_score":0.03531897,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4403896401","doi":"10.1027/2698-1866/a000087","title":"Development of a Revised Urdu Version of the 20-Item Toronto Alexithymia Scale","year":2024,"lang":"en","type":"article","venue":"Psychological Test Adaptation and Development","topic":"Psychosomatic Disorders and Their Treatments","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Trent University; Western University; University of Toronto","funders":"University of Toronto","keywords":"Alexithymia; Urdu; Psychology; Scale (ratio); Toronto Alexithymia Scale; Clinical psychology; Cartography; Geography; Linguistics; Philosophy","authors":[{"name":"Aqsa Zahid","is_ca":true},{"name":"Sharlane C. L. Lau","is_ca":true},{"name":"Laiba Rizwan","is_ca":true},{"name":"James D. A. Parker","is_ca":true},{"name":"R. Michael Bagby","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04484046673102685,"gpt":0.3112815745295487,"spread":0.2664411077985218,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002790225,0.0004258172,0.0005735693,0.001138516,0.000763413,0.0007204277,0.0006166009,0.0002329625,0.007554245],"category_scores_gemma":[0.01081058,0.0002192448,0.0005366809,0.001167635,0.0002927389,0.0006028659,0.0009755619,0.0008588302,0.001729069],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00120851,"about_ca_system_score_gemma":0.002742568,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0160708,"about_ca_topic_score_gemma":0.03596547,"domain_scores_codex":[0.9978747,0.0005055637,0.0003341653,0.0001809059,0.0009902591,0.0001143975],"domain_scores_gemma":[0.996046,0.0004787832,0.0004340036,0.0002368353,0.002517443,0.0002869007],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0004220389,0.0004623852,0.2448282,0.001114702,0.0001767414,0.0008757083,0.008351865,0.0008763737,0.005987494,0.003321185,0.08099253,0.6525907],"study_design_scores_gemma":[0.0002235507,0.0008720509,0.848029,0.0009820223,0.0001485558,0.001942614,0.003581188,0.00319268,0.002859848,0.002379564,0.1356103,0.0001785475],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.8563456,0.003306058,0.03426859,0.002741695,0.001190779,0.008616884,0.03413438,0.0008610144,0.05853498],"genre_scores_gemma":[0.823534,0.003857309,0.1214053,0.0007915834,0.0002275583,0.01113719,0.02447539,0.0001643225,0.01440729],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.0160708,"threshold_uncertainty_score":0.03195447,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}