{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":8,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":8,"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":"6b31ad525f57","filters":{"venue":"Journal on Satisfiability Boolean Modeling and Computation"}},"results":[{"id":"W1488842140","doi":"10.3233/sat190055","title":"QBF-Based Formal Verification: Experience and Perspectives","year":2008,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Formal Methods in Verification","field":"Computer Science","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","funders":"Strong","keywords":"Formal methods; Formal verification; Computer science; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.08181267425518404,"gpt":0.3148733832200596,"spread":0.2330607089648755,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0008877631,0.0001508619,0.0001667707,0.0001403325,0.0007175994,0.0001944911,0.0002054593,0.00006314079,0.000002297846],"category_scores_gemma":[0.0001735859,0.0001408248,0.00004621952,0.0001803279,0.0001406128,0.001170504,0.00004958882,0.0002750095,0.000002768543],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009449816,"about_ca_system_score_gemma":0.00008903855,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008896621,"about_ca_topic_score_gemma":4.028245e-7,"domain_scores_codex":[0.9984233,0.0002144102,0.0003973008,0.0003891244,0.0003712502,0.0002046417],"domain_scores_gemma":[0.9990609,0.0001047021,0.0001778727,0.000259178,0.0002535041,0.0001438333],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001534353,0.0002331565,0.00492361,0.00004849779,0.00001993852,0.00001308487,0.02499457,0.5683134,0.0003528818,0.05531229,0.00002847254,0.3456067],"study_design_scores_gemma":[0.0003744246,0.000214088,0.01743213,0.00002498206,0.000003628248,0.0001670556,0.0004571716,0.9750964,0.0001318192,0.005910355,0.00003150077,0.0001564301],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4515346,0.00008968512,0.5478141,0.0002666392,0.00008484986,0.0000904674,4.420567e-7,0.00004955151,0.00006967276],"genre_scores_gemma":[0.694872,0.00006282615,0.3048845,0.0001124793,0.000047827,0.00001006398,8.253608e-7,0.000005398438,0.000004129336],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.406783,"threshold_uncertainty_score":0.5742668,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2098347190","doi":"10.3233/sat190050","title":"Improved SAT-based Reachability Analysis with Observability Don’t Cares","year":2008,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Formal Methods in Verification","field":"Computer Science","cited_by":22,"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":"Observability; Reachability; Mathematics; Computer science; Theoretical computer science; Applied mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.04621811252362271,"gpt":0.2875096499397546,"spread":0.2412915374161319,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002593117,0.000277513,0.0004591298,0.0002900723,0.0006647308,0.0002283916,0.0003854988,0.0001151867,0.000003597228],"category_scores_gemma":[0.0003732226,0.0002249142,0.0001738964,0.0009489378,0.0001788826,0.0007448629,0.00006381796,0.0005244086,0.000001675787],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002956318,"about_ca_system_score_gemma":0.0002853081,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001956212,"about_ca_topic_score_gemma":0.000057161,"domain_scores_codex":[0.9967585,0.000780082,0.0007713409,0.0007363131,0.0006379648,0.0003157919],"domain_scores_gemma":[0.9975655,0.0003327188,0.0003686637,0.0007623896,0.0007277426,0.0002429718],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001775703,0.0001845316,0.04500212,0.00003751721,0.00008717346,0.000005265927,0.0007619609,0.9072565,0.0001050215,0.0007471208,0.000005371937,0.04562981],"study_design_scores_gemma":[0.0005252031,0.0005162034,0.05881302,0.0000199887,0.00007740899,0.00003488217,0.00006695159,0.9356999,0.0001410984,0.003850661,0.000009942286,0.0002447899],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4818037,0.00002498266,0.5175865,0.0002112476,0.00008054925,0.000180642,0.000002159232,0.00008666681,0.00002359096],"genre_scores_gemma":[0.6284816,0.000006513682,0.3713009,0.0001456995,0.00003272454,0.00001365359,0.000007292459,0.00000799037,0.000003563624],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1466779,"threshold_uncertainty_score":0.9171735,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2995736486","doi":"10.3233/sat190119","title":"MaxSAT Evaluation 2018: New Developments and Detailed Results","year":2019,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Formal Methods in Verification","field":"Computer Science","cited_by":20,"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":"Carnegie Mellon Portugal; National Science Foundation","keywords":"Maximum satisfiability problem; Satisfiability; Boolean satisfiability problem; Series (stratigraphy); Computer science; Radiomics; Theoretical computer science; Artificial intelligence; Algorithm; Boolean function","retraction":null,"screen_n_in":null,"score":{"opus":0.06352342221819057,"gpt":0.3266440326746796,"spread":0.263120610456489,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003729757,0.0001601767,0.0001928197,0.0001662472,0.000209437,0.0003575929,0.000190507,0.00008344475,0.000003836594],"category_scores_gemma":[0.0003335642,0.0001461521,0.00003542812,0.0001732448,0.00002080092,0.0008421093,0.00008235444,0.0002726061,0.00002442483],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001698435,"about_ca_system_score_gemma":0.0001873414,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001287532,"about_ca_topic_score_gemma":0.000002786564,"domain_scores_codex":[0.9977236,0.0003936272,0.0006040182,0.0004475312,0.0006389829,0.0001922781],"domain_scores_gemma":[0.9988642,0.0001448876,0.0002814228,0.0002693886,0.0002696762,0.0001703707],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000100316,0.00002849246,0.002396428,0.00001786865,0.00001456233,6.273966e-7,0.001064411,0.2112058,0.0001258856,0.002734965,0.00006339881,0.7822472],"study_design_scores_gemma":[0.001172027,0.0001796403,0.02420705,0.00007515655,0.00001097584,0.00003846203,0.00005013958,0.9368488,0.00006048382,0.03714147,0.00005715583,0.0001586219],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4662171,0.00005653847,0.5326582,0.0002420467,0.0002977064,0.0002531503,5.539139e-7,0.00003303947,0.0002416837],"genre_scores_gemma":[0.6526798,0.00003006202,0.3470825,0.0001172442,0.00004305145,0.000003602767,0.000003514288,0.000006119572,0.00003416058],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7820886,"threshold_uncertainty_score":0.595991,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2126735726","doi":"10.3233/sat190045","title":"A Switching Criterion for Intensification and Diversification in Local Search for SAT12","year":2008,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Constraint Satisfaction and Optimization","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of New Brunswick","funders":"Natural Sciences and Engineering Research Council of Canada; Technische Universiteit Delft","keywords":"Diversification (marketing strategy); Business; Economics; Marketing","retraction":null,"screen_n_in":null,"score":{"opus":0.06049728746025682,"gpt":0.2991865808051254,"spread":0.2386892933448686,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009051457,0.0001123495,0.0001558486,0.0002453135,0.0004363674,0.0001483588,0.00007963328,0.00006592244,5.779456e-7],"category_scores_gemma":[0.0001049369,0.0001137221,0.00004595726,0.0001302338,0.0000423385,0.0005409608,0.00002706385,0.0001695217,4.87025e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001285866,"about_ca_system_score_gemma":0.0000707736,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002431991,"about_ca_topic_score_gemma":0.00001122601,"domain_scores_codex":[0.9988335,0.00009733922,0.0003969337,0.0003329384,0.0001813132,0.0001580006],"domain_scores_gemma":[0.9990959,0.0002631643,0.0001126196,0.0001076195,0.0003330283,0.00008762734],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001157474,0.0000430071,0.003250903,0.00003918113,0.000006685515,0.000001127494,0.002518182,0.4449303,0.0002501963,0.00397627,0.00001249471,0.5448558],"study_design_scores_gemma":[0.0008914969,0.0001784843,0.01460111,0.00004662395,0.000005178335,0.00008682498,0.0003723358,0.9752865,0.00003534971,0.008365927,0.00001350123,0.0001166417],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.383905,0.00002351289,0.614745,0.0008919715,0.0000796771,0.0003208837,0.000001618853,0.00002624125,0.00000608755],"genre_scores_gemma":[0.9092379,0.00007859711,0.09044858,0.0001563331,0.00003819831,0.00002170848,0.000009044951,0.00000680005,0.000002820333],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5447392,"threshold_uncertainty_score":0.4637454,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W121484681","doi":"10.3233/sat190001","title":"Semidefinite Optimization Approaches for Satisfiability and Maximum-Satisfiability Problems","year":2005,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Complexity and Algorithms in Graphs","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"Semidefinite programming; Semidefinite embedding; Satisfiability; Relaxation (psychology); Mathematics; Quadratically constrained quadratic program; Maximum cut; Linear programming; Large margin nearest neighbor; Mathematical optimization; Combinatorics; Computer science; Discrete mathematics; Quadratic programming; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.06582800461354198,"gpt":0.267172676926121,"spread":0.201344672312579,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.003104571,0.0003861992,0.0004813703,0.0002455281,0.0008729512,0.0007383849,0.0003441378,0.000175177,0.000006417816],"category_scores_gemma":[0.0003469789,0.0003657282,0.000167094,0.0003134718,0.0001930329,0.001172602,0.0001782036,0.0005292012,0.000001979076],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001850429,"about_ca_system_score_gemma":0.0001130838,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002280222,"about_ca_topic_score_gemma":0.00001810404,"domain_scores_codex":[0.9966227,0.0003970181,0.001040276,0.0009541682,0.0005300413,0.0004557668],"domain_scores_gemma":[0.9977394,0.0007522826,0.00036142,0.0004862781,0.0003644474,0.0002961525],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003142436,0.0001167774,0.001042787,0.0001120083,0.00001996917,3.648021e-7,0.0006448,0.6589874,0.00000467898,0.00788103,0.00001247869,0.3311463],"study_design_scores_gemma":[0.0007890452,0.0003078357,0.00172124,0.00008864336,0.00002105004,0.0000537766,0.00006856088,0.7799224,0.000009994783,0.2166334,0.00007330748,0.0003107752],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1543878,0.0002686988,0.8422843,0.001752727,0.0001905688,0.0008232555,0.00001644194,0.0001585694,0.0001176575],"genre_scores_gemma":[0.6001691,0.0000655692,0.399388,0.0001684141,0.0001269667,0.00004862067,0.00001135178,0.0000158922,0.000006163821],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.4457812,"threshold_uncertainty_score":0.9998795,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2159832387","doi":"10.3233/sat190066","title":"Algorithms and Complexity Results for Input and Unit Resolution","year":2009,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"graph theory and CDMA systems","field":"Engineering","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; University of Toronto","keywords":"Resolution (logic); Completeness (order theory); Space (punctuation); PSPACE; Computer science; Algorithm; Mathematics; Computational complexity theory; Artificial intelligence; Mathematical analysis","retraction":null,"screen_n_in":null,"score":{"opus":0.05055338728897672,"gpt":0.2739712533785834,"spread":0.2234178660896066,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001008963,0.0001318229,0.0001879679,0.00009621657,0.0002692484,0.0001194402,0.00003152024,0.00007123756,2.994719e-7],"category_scores_gemma":[0.00005449185,0.0001247186,0.00003237422,0.00006198881,0.00004361889,0.0001588341,0.000007020806,0.0001993644,3.244628e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002868658,"about_ca_system_score_gemma":0.000008055446,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006178107,"about_ca_topic_score_gemma":0.000004532463,"domain_scores_codex":[0.9990735,0.00009727309,0.0003784826,0.0001810823,0.0001207104,0.0001489723],"domain_scores_gemma":[0.9995387,0.0001258158,0.00006341849,0.00007631688,0.00008401088,0.0001117084],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002836838,0.00003078383,0.0002322795,0.0000887408,0.00002503186,0.00000148651,0.0009634199,0.8012514,0.0001661244,0.00537606,0.00008103355,0.1915],"study_design_scores_gemma":[0.0008264806,0.0002664539,0.005257818,0.00007569375,0.00001351378,0.00004651934,0.0001312201,0.8579976,0.00001180945,0.1351947,0.00006119217,0.0001170566],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6590968,0.0002265534,0.3400695,0.0001691244,0.00009028042,0.0001670784,0.00001325663,0.00005667801,0.0001106785],"genre_scores_gemma":[0.9920971,0.00008356517,0.007629533,0.00005946812,0.0001019528,0.000003366772,0.00001137308,0.000008663786,0.000005003777],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3330003,"threshold_uncertainty_score":0.5085875,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2163348426","doi":"10.3233/sat190036","title":"An Extended Semidefinite Relaxation for Satisfiability","year":2007,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Formal Methods in Verification","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada; Technische Universiteit Delft","keywords":"Semidefinite programming; Relaxation (psychology); Satisfiability; Semidefinite embedding; Mathematics; Class (philosophy); Extension (predicate logic); Combinatorics; Mathematical optimization; Discrete mathematics; Computer science; Quadratically constrained quadratic program; Quadratic programming; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.04658219512101994,"gpt":0.3505630059101519,"spread":0.303980810789132,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008756869,0.0002258207,0.0002678969,0.0002542463,0.0005613266,0.0003849054,0.0003404742,0.00015131,0.000001925833],"category_scores_gemma":[0.0007649407,0.000216818,0.0001069429,0.0002800197,0.00006183556,0.001346016,0.00004270234,0.000401108,0.000003561879],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002430641,"about_ca_system_score_gemma":0.00009613349,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001534599,"about_ca_topic_score_gemma":0.000008996042,"domain_scores_codex":[0.9971867,0.0003975032,0.0009466092,0.0005994968,0.0005066118,0.0003630398],"domain_scores_gemma":[0.9976573,0.0005749057,0.000442828,0.0005055254,0.0005575721,0.0002618785],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000171892,0.00016401,0.001304609,0.00004267071,0.00001352155,0.000001459753,0.001008788,0.2156895,0.0005520048,0.0606692,0.00001446216,0.7203679],"study_design_scores_gemma":[0.000481018,0.000564423,0.01891815,0.00003173969,0.00001078908,0.00003735697,0.0001047509,0.7978942,0.0002334539,0.1814891,0.00003804682,0.0001970241],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3822052,0.00002785183,0.6167619,0.0001403396,0.0003215082,0.0003392252,0.000002483617,0.0001144061,0.00008707321],"genre_scores_gemma":[0.5467883,0.000009183787,0.4529127,0.0001564376,0.0001022602,0.0000110955,0.000006675636,0.00001045738,0.000002866965],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7201709,"threshold_uncertainty_score":0.884158,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2580735446","doi":"10.3233/sat190111","title":"Hard satisfiable 3-SAT instances via autocorrelation","year":2016,"lang":"en","type":"article","venue":"Journal on Satisfiability Boolean Modeling and Computation","topic":"Formal Methods in Verification","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Wilfrid Laurier University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Autocorrelation; Boolean satisfiability problem; Reduction (mathematics); Satisfiability; Computer science; Mathematics; Combinatorics; Algorithm; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.03937523132885755,"gpt":0.2920864361888223,"spread":0.2527112048599647,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001966739,0.0001999372,0.0002308786,0.0002030893,0.0004083275,0.0002975635,0.0002951189,0.000105599,0.000007662057],"category_scores_gemma":[0.0002750558,0.0001447168,0.0000760102,0.0002378958,0.00006365775,0.001672451,0.00006934804,0.0002666064,0.000038398],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002258714,"about_ca_system_score_gemma":0.00008540383,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001047202,"about_ca_topic_score_gemma":0.000002190466,"domain_scores_codex":[0.9976899,0.0003891543,0.0006601734,0.0004604818,0.0005283602,0.0002719199],"domain_scores_gemma":[0.9986303,0.0002591109,0.0003335291,0.000342448,0.000272469,0.0001621659],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004360751,0.00004633123,0.002713753,0.00001668453,0.00001451695,0.0000018671,0.0003734172,0.1017086,0.0008182358,0.03190409,0.00004769118,0.8623112],"study_design_scores_gemma":[0.0004570481,0.0002047214,0.0108591,0.0001070751,0.00000692255,0.00004880311,0.00001882394,0.8309575,0.0001339758,0.1569298,0.00009713204,0.0001791366],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2740855,0.00005274404,0.7241929,0.0006630735,0.0004810758,0.0002081059,0.000001389565,0.0001238688,0.0001913398],"genre_scores_gemma":[0.6514786,0.0000388542,0.3482071,0.0001337471,0.00008358255,0.00001253335,0.000001167228,0.000009206082,0.00003518798],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.862132,"threshold_uncertainty_score":0.5901382,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}