{"meta":{"query_hash":"29178c4a3bef","filters":{"venue":"ACM Transactions on Recommender Systems"},"cohort_total":8,"direct_labels_cover":0,"predictions_cover":8,"exported":8,"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/29178c4a3bef","api":"https://metacan.xera.ac/api/v1/cohort?venue=ACM+Transactions+on+Recommender+Systems"},"results":[{"id":"W4366598314","doi":"10.1145/3591109","title":"A Framework and Toolkit for Testing the Correctness of Recommendation Algorithms","year":2023,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Recommender Systems and Techniques","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Test suite; Correctness; Unit testing; Python (programming language); Algorithm; Recommender system; Suite; White-box testing; Implementation; Surprise; Regression testing; Code coverage; Integration testing; Test case; Software engineering; Software; Programming language; Machine learning; Software system; Software construction","score_opus":0.12685383430496341,"score_gpt":0.3261905453913853,"score_spread":0.1993367110864219,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366598314","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.003989222,0.00024382788,0.9137149,0.00044079832,0.00010069781,0.00044489958,0.0007193979,0.0779419,0.0024042856],"genre_scores_gemma":[0.09268761,0.00062346406,0.884574,0.0008130341,0.0001354075,0.0020586806,0.003354991,0.012365044,0.0033877536],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98414683,0.005094115,0.0023002531,0.0019174616,0.0055326796,0.0010086492],"domain_scores_gemma":[0.9704329,0.01544132,0.002168676,0.0078100227,0.0033404534,0.0008066139],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012173374,0.00274486,0.001580146,0.0039447583,0.0010183404,0.0037078254,0.0066931043,0.0032519551,0.00879672],"category_scores_gemma":[0.050164312,0.002606544,0.0046132733,0.001591801,0.0038079584,0.007837695,0.0063396785,0.0065720654,0.0053695827],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001098128,0.001674696,0.027803753,0.0035169853,0.0008399277,0.0026395733,0.0022191324,0.16455211,0.03149631,0.24233195,0.08674587,0.43508154],"study_design_scores_gemma":[0.00039593372,0.000908388,0.0060752435,0.0010429077,0.00022133201,0.0022561324,0.00037791464,0.63307977,0.038163085,0.20807517,0.1089739,0.00043025753],"about_ca_topic_score_codex":0.006409619,"about_ca_topic_score_gemma":0.0042873956,"teacher_disagreement_score":0.012173374,"about_ca_system_score_codex":0.0018536346,"about_ca_system_score_gemma":0.004390904,"threshold_uncertainty_score":0.06437969},"labels":[],"label_agreement":null},{"id":"W4386136194","doi":"10.1145/3616379","title":"Multi-Interest Multi-Round Conversational Recommendation System with Fuzzy Feedback Based User Simulator","year":2023,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Recommender Systems and Techniques","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"National Natural Science Foundation of China","keywords":"Computer science; Preference; Recommender system; Human–computer interaction; Fuzzy logic; Information retrieval; Artificial intelligence","score_opus":0.08600942693861284,"score_gpt":0.2924780118068677,"score_spread":0.20646858486825487,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386136194","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.090523824,0.00044051418,0.8979607,0.00037734842,0.00010909264,0.00046672407,0.0002653998,0.004926362,0.0049300515],"genre_scores_gemma":[0.8887628,0.00013379789,0.10551822,0.00019926779,0.00004044245,0.00045067817,0.00029229635,0.00005861536,0.004543877],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9991098,0.0002161065,0.000088554705,0.00027766096,0.00020461451,0.00010311964],"domain_scores_gemma":[0.9985795,0.0006176059,0.00008252304,0.00015412553,0.00035706247,0.00020917841],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012280652,0.0008394891,0.0013000211,0.0003660186,0.00080469757,0.00089694787,0.0023620483,0.0015732597,0.0048053428],"category_scores_gemma":[0.0027770745,0.00045252958,0.0007844723,0.00022972103,0.0004036335,0.0010837452,0.0013810857,0.0012273513,0.0012997871],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00343044,0.0011883447,0.0097946245,0.0007787548,0.00041810988,0.002103616,0.0020116584,0.63051337,0.06799328,0.01078208,0.0068663782,0.2641193],"study_design_scores_gemma":[0.000038807684,0.00006690445,0.00015905857,0.0000031057032,0.000016880143,0.000064208,0.000022378961,0.9973815,0.0012987319,0.00046169249,0.00047037692,0.000016274318],"about_ca_topic_score_codex":0.011248299,"about_ca_topic_score_gemma":0.008194828,"teacher_disagreement_score":0.011248299,"about_ca_system_score_codex":0.00065671885,"about_ca_system_score_gemma":0.0010316318,"threshold_uncertainty_score":0.02236563},"labels":[],"label_agreement":null},{"id":"W4391436413","doi":"10.1145/3643138","title":"Measuring Commonality in Recommendation of Cultural Content to Strengthen Cultural Citizenship","year":2024,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Recommender Systems and Techniques","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"University College London; Canadian Institute for Advanced Research","keywords":"Citizenship; Content (measure theory); Cultural diversity; Sociology; Political science; Anthropology; Law","score_opus":0.2429439671267938,"score_gpt":0.3216424187380756,"score_spread":0.07869845161128178,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391436413","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88735294,0.00085842074,0.105979666,0.00028976853,0.00004673078,0.00035229785,0.00046645518,0.0005976608,0.0040560933],"genre_scores_gemma":[0.9436584,0.000106496744,0.055344764,0.000048942988,0.000019077832,0.00008207728,0.0004042538,0.000022793187,0.00031312509],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99088913,0.0034804977,0.0008687812,0.0019709638,0.002348917,0.00044168698],"domain_scores_gemma":[0.94155675,0.03564131,0.005908215,0.009466292,0.005721637,0.001705782],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010813337,0.00081691437,0.0015965733,0.0048145503,0.0015019586,0.0035543442,0.0013405387,0.0015121901,0.00077413593],"category_scores_gemma":[0.061822418,0.0003046865,0.0011802331,0.003837105,0.0012029491,0.0043451083,0.0032002556,0.0015975656,0.00030747152],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018258797,0.0016396804,0.50027156,0.0008797685,0.0014873041,0.00026940345,0.004393738,0.07911115,0.016018234,0.010998804,0.0022817196,0.3808227],"study_design_scores_gemma":[0.00018300598,0.002017741,0.24222918,0.0002037677,0.0006725069,0.00063259614,0.003315041,0.7008606,0.022358477,0.021988673,0.0052015227,0.0003367602],"about_ca_topic_score_codex":0.008856094,"about_ca_topic_score_gemma":0.008334986,"teacher_disagreement_score":0.010813337,"about_ca_system_score_codex":0.0014269671,"about_ca_system_score_gemma":0.0012948415,"threshold_uncertainty_score":0.05718708},"labels":[],"label_agreement":null},{"id":"W4405194834","doi":"10.1145/3707693","title":"Modelling Concept Drift in Dynamic Data Streams for Recommender Systems","year":2024,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Data Stream Mining Techniques","field":"Computer Science","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Recommender system; STREAMS; Computer science; Concept drift; Data stream mining; Dynamic data; Data mining; Data science; Information retrieval; Database; Computer network","score_opus":0.10291693760084104,"score_gpt":0.33782853265750123,"score_spread":0.2349115950566602,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405194834","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05190644,0.00046917578,0.94577765,0.00046223478,0.00009463874,0.00014800043,0.00036684662,0.0002154322,0.0005595202],"genre_scores_gemma":[0.7847932,0.00079119386,0.21064094,0.00019919206,0.00013284222,0.00041478843,0.00076975796,0.000068784735,0.0021892446],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9986651,0.000622164,0.000089533736,0.00026460786,0.00025836684,0.00010029809],"domain_scores_gemma":[0.98881197,0.0084766755,0.00084450695,0.0007032905,0.0009299464,0.0002337302],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006007558,0.00082047837,0.0010620048,0.0010904528,0.0005441082,0.0015537033,0.0018563416,0.0016290876,0.0011943511],"category_scores_gemma":[0.025950907,0.00069826865,0.0009965113,0.0011991143,0.00094470644,0.0025771924,0.0012392281,0.002309452,0.00033226167],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015229738,0.00006487606,0.0055879266,0.00010630743,0.00006550278,0.00013988928,0.0002931927,0.9278395,0.0010866199,0.03903069,0.0011059924,0.024527201],"study_design_scores_gemma":[0.000005343537,0.000012502709,0.00018096952,0.000004961779,0.0000036540744,0.000013956015,0.0000124885755,0.99158317,0.00011838688,0.007822329,0.00023752292,0.000004714962],"about_ca_topic_score_codex":0.0077116275,"about_ca_topic_score_gemma":0.0076239444,"teacher_disagreement_score":0.0077116275,"about_ca_system_score_codex":0.0015589286,"about_ca_system_score_gemma":0.0009977624,"threshold_uncertainty_score":0.03177136},"labels":[],"label_agreement":null},{"id":"W4407654895","doi":"10.1145/3718488","title":"Balancing Embedding Spectrum for Recommendation","year":2025,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Recommender Systems and Techniques","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":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Computer science; Embedding; Spectrum (functional analysis); Artificial intelligence; Physics","score_opus":0.02706613366870131,"score_gpt":0.30838186907996673,"score_spread":0.2813157354112654,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407654895","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.024915574,0.0003290989,0.97122127,0.00024236536,0.00003972101,0.000064142325,0.000071768685,0.001058456,0.0020576294],"genre_scores_gemma":[0.5845034,0.00031111628,0.40894592,0.00048413745,0.00012671482,0.00022375061,0.00039896608,0.0002824074,0.004723538],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9978592,0.00071251055,0.00008700161,0.0005947205,0.0005965115,0.00015002827],"domain_scores_gemma":[0.9970337,0.0012244135,0.0001916075,0.00090739084,0.00050915015,0.00013367919],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0022365353,0.00093531626,0.0012144356,0.0008701435,0.0008493621,0.0010804232,0.0017390357,0.0014515636,0.0035049303],"category_scores_gemma":[0.009821524,0.0005391953,0.0005902003,0.0011151057,0.00092815835,0.003485012,0.002208169,0.0019049222,0.0019478681],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000487941,0.0005623318,0.003510305,0.00019863823,0.000116413576,0.000114819675,0.00041926777,0.24198504,0.020649599,0.04343733,0.010380857,0.67813754],"study_design_scores_gemma":[0.000030356554,0.0000799511,0.00044528808,0.000013247032,0.000014450233,0.00008036981,0.00005830212,0.96704525,0.0033238595,0.026196323,0.0026939625,0.000018690684],"about_ca_topic_score_codex":0.0025906912,"about_ca_topic_score_gemma":0.0048268167,"teacher_disagreement_score":0.0035049303,"about_ca_system_score_codex":0.00087209034,"about_ca_system_score_gemma":0.00097532436,"threshold_uncertainty_score":0.011828065},"labels":[],"label_agreement":null},{"id":"W4409168760","doi":"10.1145/3728373","title":"Recall, Robustness, and Lexicographic Evaluation","year":2025,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Topic Modeling","field":"Computer Science","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Microsoft (Canada)","funders":"","keywords":"Lexicographical order; Robustness (evolution); Recall; Computer science; Artificial intelligence; Cognitive psychology; Mathematics; Psychology; Chemistry; Combinatorics","score_opus":0.05255388057789485,"score_gpt":0.2979719475362509,"score_spread":0.24541806695835605,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409168760","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16495243,0.0054069064,0.7915114,0.0025635343,0.00028454335,0.0006626607,0.0014489199,0.00061614776,0.03255344],"genre_scores_gemma":[0.84028304,0.00089369074,0.15382238,0.00060245243,0.00041086,0.0006187239,0.0011767605,0.00018702912,0.0020050071],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9248484,0.043828435,0.006632861,0.0061926143,0.017272625,0.0012250507],"domain_scores_gemma":[0.6673087,0.25700557,0.027848737,0.03025647,0.015693244,0.0018872586],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.053536803,0.0014979509,0.0019205065,0.009679002,0.0014949683,0.0059272046,0.001565361,0.0019623113,0.003206602],"category_scores_gemma":[0.27469733,0.00050745707,0.0016831437,0.008139462,0.0070576677,0.011884118,0.004184558,0.0022118187,0.0006979393],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020776577,0.0005038314,0.09188061,0.0020970646,0.0013566914,0.00036692148,0.0048069493,0.057431325,0.004201259,0.44007894,0.006346952,0.38885173],"study_design_scores_gemma":[0.00023139533,0.0020821628,0.043169305,0.0009665648,0.0005911299,0.0010390318,0.0021197035,0.13899225,0.010999877,0.78046095,0.01878065,0.0005670422],"about_ca_topic_score_codex":0.0016530567,"about_ca_topic_score_gemma":0.0013284549,"teacher_disagreement_score":0.053536803,"about_ca_system_score_codex":0.0030587048,"about_ca_system_score_gemma":0.001593243,"threshold_uncertainty_score":0.28313303},"labels":[],"label_agreement":null},{"id":"W4412841400","doi":"10.1145/3757327","title":"A Comparative Survey Of Algorithmic Feed Recommendation System Designs","year":2025,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Recommender Systems and Techniques","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":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Information retrieval; Data science","score_opus":0.10218141728614505,"score_gpt":0.32943248951364307,"score_spread":0.22725107222749802,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412841400","genre_codex":"empirical","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5552731,0.08571675,0.2813888,0.0038522328,0.00032817724,0.0014703255,0.0043141767,0.004628517,0.06302794],"genre_scores_gemma":[0.568391,0.043294806,0.36729485,0.00090821076,0.00018494221,0.00091223675,0.007144615,0.0010280826,0.01084125],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98084706,0.007262876,0.0031088004,0.0016147906,0.006507628,0.0006588238],"domain_scores_gemma":[0.92415434,0.053449795,0.0032033254,0.006372229,0.012131431,0.00068885734],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.018413102,0.0009478169,0.00092492386,0.007443119,0.0009415357,0.0048285597,0.001441623,0.0018501714,0.0028272793],"category_scores_gemma":[0.06215117,0.0010952903,0.0011644663,0.009359284,0.00089778146,0.0077593247,0.001096036,0.0010700428,0.001671557],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008102468,0.00028224537,0.0521697,0.0071389084,0.00039036915,0.00013917382,0.0046632225,0.0040193684,0.0072601736,0.01781418,0.009122008,0.89619046],"study_design_scores_gemma":[0.0003158457,0.00551206,0.17027533,0.008199023,0.0013394817,0.0050051794,0.012763299,0.057196155,0.04041736,0.014521134,0.68362224,0.00083294924],"about_ca_topic_score_codex":0.002817461,"about_ca_topic_score_gemma":0.0037644175,"teacher_disagreement_score":0.018413102,"about_ca_system_score_codex":0.001444742,"about_ca_system_score_gemma":0.0014811865,"threshold_uncertainty_score":0.09737891},"labels":[],"label_agreement":null},{"id":"W4413167298","doi":"10.1145/3760763","title":"Revisiting LightGCN: Unexpected Inflexibility, Inconsistency, and A Remedy Towards Improved Recommendation","year":2025,"lang":"en","type":"article","venue":"ACM Transactions on Recommender Systems","topic":"Recommender Systems and Techniques","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":false,"ca_institutions":"Kootenay Association for Science & Technology","funders":"","keywords":"Computer science; Psychology","score_opus":0.028871919565577144,"score_gpt":0.2931055492405019,"score_spread":0.26423362967492475,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413167298","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20141597,0.007850984,0.7644972,0.006218673,0.0011327076,0.0004989518,0.0016105432,0.0074913003,0.009283632],"genre_scores_gemma":[0.6241862,0.0014255192,0.36020923,0.0029511005,0.0003599013,0.00023771719,0.0024262466,0.000531307,0.0076728333],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9977956,0.0007698778,0.00013359568,0.000637109,0.00051548064,0.00014837702],"domain_scores_gemma":[0.9919477,0.002937047,0.000308533,0.0026757908,0.0018842347,0.0002467607],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005160345,0.0013031928,0.001443496,0.0011866163,0.0011416662,0.0018863927,0.003944609,0.0025848916,0.001794184],"category_scores_gemma":[0.01804015,0.0008146049,0.0008670027,0.001672904,0.0014976249,0.004471669,0.0027358243,0.0035773388,0.0011333207],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006654718,0.00039518625,0.013592621,0.00071885495,0.0005402228,0.00039584065,0.00082988344,0.2773648,0.009560111,0.025001913,0.03326199,0.6376731],"study_design_scores_gemma":[0.00006717244,0.00015986772,0.0010282737,0.00005559871,0.000081508035,0.00013826533,0.000117813426,0.9736091,0.0023059656,0.01633548,0.006062151,0.00003888217],"about_ca_topic_score_codex":0.02855209,"about_ca_topic_score_gemma":0.06026249,"teacher_disagreement_score":0.02855209,"about_ca_system_score_codex":0.0020238294,"about_ca_system_score_gemma":0.0019236378,"threshold_uncertainty_score":0.056771815},"labels":[],"label_agreement":null}]}