{"meta":{"query_hash":"54a798ac6bc9","filters":{"venue":"AEA Randomized Controlled Trials"},"cohort_total":24,"direct_labels_cover":3,"predictions_cover":24,"exported":24,"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/54a798ac6bc9","api":"https://metacan.xera.ac/api/v1/cohort?venue=AEA+Randomized+Controlled+Trials"},"results":[{"id":"W2580505567","doi":"10.1257/rct.810-3.0","title":"Texting Students to Help Achieve Their Goals","year":2015,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Educational and Psychological Assessments","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science; Mathematics education; World Wide Web; Psychology","score_opus":0.11367487000037,"score_gpt":0.4751930426452955,"score_spread":0.3615181726449255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2580505567","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":[0.0014637774,0.0039623203,0.00009115026,0.0032726144,0.014966674,0.0200609,0.94661045,0.00012304004,0.009449051],"genre_scores_gemma":[0.0022573571,0.00043331948,0.00016708332,0.007197313,0.006216615,0.019480255,0.9467041,0.00009633143,0.01744765],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9685496,0.0220578,0.005452802,0.0014096724,0.0015310454,0.0009990553],"domain_scores_gemma":[0.94979864,0.043966655,0.0035269463,0.0014574894,0.0005674599,0.00068283227],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.05884898,0.0012525242,0.018831845,0.0006399009,0.00021777966,0.00040656346,0.0021550257,0.0010796215,0.01467629],"category_scores_gemma":[0.031127185,0.00068918854,0.00363876,0.00043831623,0.00015032967,0.000080878934,0.00030364876,0.0009650358,0.0077422815],"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":[0.40967664,0.0008161803,0.0000057546463,0.00001005586,0.0061561377,0.000018776082,0.00017604914,0.0000023320829,0.0000058032083,0.00011192779,0.5820942,0.00092614105],"study_design_scores_gemma":[0.5415162,0.00021146562,0.000016465916,0.00009756487,0.0017724643,0.0000069938487,0.00019965638,4.0270388e-7,7.273114e-7,0.0034376057,0.45227125,0.00046920008],"about_ca_topic_score_codex":0.00035753843,"about_ca_topic_score_gemma":0.000030154191,"teacher_disagreement_score":0.13183956,"about_ca_system_score_codex":0.00015434947,"about_ca_system_score_gemma":0.00031575985,"threshold_uncertainty_score":0.99955595},"labels":[],"label_agreement":null},{"id":"W2949163000","doi":"10.1257/rct.3147-1.0","title":"Welfare effects of a non-contributory old age pension: experimental evidence for Ekiti State in Nigeria","year":2019,"lang":"en","type":"preprint","venue":"AEA Randomized Controlled Trials","topic":"Financial Literacy, Pension, Retirement Analysis","field":"Business, Management and Accounting","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":"Carleton University","funders":"","keywords":"Pension; Welfare; Demographic economics; State (computer science); Welfare state; Business; Economics; Political science; Computer science; Finance; Law; Market economy; Algorithm","score_opus":0.03228616867411146,"score_gpt":0.3148393669534935,"score_spread":0.2825531982793821,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2949163000","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9294849,0.012790304,0.0011843775,0.00038070796,0.005841088,0.049730334,0.00007173153,0.000133121,0.00038341404],"genre_scores_gemma":[0.98818654,0.00032986669,0.0002875711,0.00046954997,0.0014807059,0.007819509,0.000397136,0.00011919168,0.0009099391],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9889448,0.0016794124,0.0061043184,0.0013688008,0.0010741137,0.00082849927],"domain_scores_gemma":[0.9707761,0.019807553,0.007222587,0.0011825035,0.00096662773,0.000044639575],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.025078885,0.0011842155,0.020057678,0.0015061708,0.00018398071,0.0005778847,0.00088427996,0.0005667282,0.00043644913],"category_scores_gemma":[0.049309712,0.00088657363,0.0064625824,0.00048090628,0.00017430636,0.0007848396,0.0008392274,0.0004938494,0.00006899189],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.9772772,0.00044326409,0.0025942253,0.0030731696,0.0016567245,0.000118720345,0.00035494403,0.0010023012,0.01078332,0.0012287467,0.0012144028,0.00025293362],"study_design_scores_gemma":[0.97179747,0.000067609646,0.0007424058,0.0048042233,0.0045453184,4.0646466e-7,0.00005492293,0.010425268,0.0026080771,0.0035972134,0.00032147617,0.0010356377],"about_ca_topic_score_codex":0.002282284,"about_ca_topic_score_gemma":0.00026114384,"teacher_disagreement_score":0.05870161,"about_ca_system_score_codex":0.00021082514,"about_ca_system_score_gemma":0.00022363041,"threshold_uncertainty_score":0.9993585},"labels":[],"label_agreement":null},{"id":"W2955944786","doi":"10.1257/rct.2295","title":"Keeping College Options Open: A Field Experiment to Help All High School Seniors Through the College Application Process","year":2017,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Higher Education Research Studies","field":"Social Sciences","cited_by":77,"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 Toronto","funders":"","keywords":"Process (computing); Field (mathematics); Mathematics education; Medical education; Computer science; Psychology; Medicine; Operating system; Mathematics","score_opus":0.06260783106028149,"score_gpt":0.4759728709744864,"score_spread":0.41336503991420487,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2955944786","genre_codex":"dataset","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00009342708,0.003645229,0.0003848971,0.13461292,0.0034246647,0.09507458,0.7560588,0.00012477396,0.0065807025],"genre_scores_gemma":[0.020398712,0.017258847,0.001693882,0.016432852,0.013157944,0.48923266,0.40378654,0.00018985821,0.03784869],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9784902,0.014574018,0.0029116725,0.0010758706,0.0019404835,0.0010077526],"domain_scores_gemma":[0.95374936,0.039532274,0.0031167876,0.0019124118,0.0011865102,0.0005026355],"candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03879248,0.0007234504,0.008688918,0.0003169085,0.00457092,0.0015060975,0.00429645,0.00058343104,0.0018487597],"category_scores_gemma":[0.1243447,0.00044381001,0.0014521831,0.00056370354,0.00066301244,0.0004954908,0.00062872865,0.0006933653,0.00061431824],"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":[0.12928726,0.00021547085,2.454717e-7,0.000029595045,0.0025110776,0.000005228733,0.0020553516,0.000020439511,0.000012672483,0.0052187922,0.86060166,0.00004218867],"study_design_scores_gemma":[0.32180867,0.000055391854,4.520892e-7,0.00013457108,0.0011168462,5.4965426e-7,0.0042758547,0.0000046862165,0.000032648517,0.0024916048,0.66970366,0.00037508484],"about_ca_topic_score_codex":0.023242733,"about_ca_topic_score_gemma":0.005327422,"teacher_disagreement_score":0.3941581,"about_ca_system_score_codex":0.00052141125,"about_ca_system_score_gemma":0.0034299695,"threshold_uncertainty_score":0.99980134},"labels":[],"label_agreement":null},{"id":"W3121396802","doi":"10.1257/rct.5733-1.0","title":"Love in the Time of HIV: Testing as a Signal of Risk","year":2020,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"HIV/AIDS Research and Interventions","field":"Medicine","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":"University of Toronto","funders":"","keywords":"Human immunodeficiency virus (HIV); SIGNAL (programming language); Medicine; Psychology; Biology; Computer science; Virology","score_opus":0.039565613952362966,"score_gpt":0.35559345015817256,"score_spread":0.3160278362058096,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3121396802","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":[0.00027523187,0.0024638912,0.0001234193,0.0011205488,0.000058241916,0.012552445,0.9799268,0.000011581638,0.0034678436],"genre_scores_gemma":[0.0064794896,0.001841532,0.0003493337,0.00034191002,0.0011614241,0.0013514665,0.9847664,0.000056923673,0.003651471],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9790277,0.014574826,0.0045947274,0.00035217617,0.0010924042,0.0003582062],"domain_scores_gemma":[0.8625659,0.13121355,0.004557801,0.000825244,0.00062218675,0.00021536778],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.03275128,0.00039687066,0.014360426,0.00061055756,0.000036065718,0.000037260215,0.00052111136,0.00029321914,0.004390389],"category_scores_gemma":[0.35680416,0.00019810394,0.0044765663,0.00054538494,0.00028709034,0.000043927197,0.000081560014,0.00091203745,0.00094332424],"study_design_candidate":"randomized_trial","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":[0.44851384,0.00024607652,0.0000011104022,0.00013765549,0.0024523565,0.000046708057,0.0000974001,0.000002853657,0.00041759876,0.000007718759,0.5476627,0.00041403584],"study_design_scores_gemma":[0.9610616,0.00163109,0.0000013211301,0.0013598595,0.0075354227,0.000014915824,0.0001258202,0.0011874065,0.0001664658,0.00051601714,0.026252527,0.00014754104],"about_ca_topic_score_codex":0.0009888903,"about_ca_topic_score_gemma":0.000030332065,"teacher_disagreement_score":0.5214101,"about_ca_system_score_codex":0.00003328709,"about_ca_system_score_gemma":0.00068678195,"threshold_uncertainty_score":0.99983454},"labels":[],"label_agreement":null},{"id":"W4241403533","doi":"10.1257/rct.26-6.0","title":"The Impact of Voter Knowledge Initiatives in Sierra Leone","year":2013,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"African Studies and Geopolitics","field":"Social 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":"ImmunoPrecise (Canada)","funders":"","keywords":"Sierra leone; Political science; Geography; Sociology; Socioeconomics","score_opus":0.04746532713651962,"score_gpt":0.39697093110529263,"score_spread":0.349505603968773,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4241403533","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":[0.00049984886,0.023111623,0.000012729915,0.0012227634,0.00193534,0.017047018,0.9144256,0.000032964304,0.041712116],"genre_scores_gemma":[0.26923004,0.18709108,0.000093584255,0.0003884006,0.01399537,0.021068012,0.48733643,0.00020220784,0.020594884],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9741481,0.020591374,0.003498852,0.00034811883,0.0005773639,0.00083620497],"domain_scores_gemma":[0.8868817,0.10862951,0.0032161141,0.0005355463,0.0005705558,0.00016659085],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.037874546,0.0005299992,0.012927558,0.0003059203,0.0005885909,0.00018689597,0.00075913995,0.0004731167,0.0008479022],"category_scores_gemma":[0.100744784,0.0002431241,0.0041246214,0.0003908358,0.0017111708,0.00013016775,0.00013905123,0.00045320234,0.000112831054],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.17171682,0.00012761353,0.0000018347778,0.000012118765,0.0035691322,0.0000024651968,0.0020421662,0.0000033932197,4.703291e-7,0.0021058351,0.82012695,0.00029116837],"study_design_scores_gemma":[0.78806883,0.000088522145,0.000011835838,0.0001103705,0.0013160954,1.8334137e-7,0.0027515993,0.000016562266,5.527423e-7,0.00803292,0.19929294,0.00030961135],"about_ca_topic_score_codex":0.02786875,"about_ca_topic_score_gemma":0.0067759263,"teacher_disagreement_score":0.62083405,"about_ca_system_score_codex":0.00021530155,"about_ca_system_score_gemma":0.0013446709,"threshold_uncertainty_score":0.99143136},"labels":[],"label_agreement":null},{"id":"W4243526881","doi":"10.1257/rct.1402","title":"Team Production in International Labor Markets: Experimental Evidence from the Field","year":2016,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Economic Policies and Impacts","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Rotman School of Management, University of Toronto; Social Sciences and Humanities Research Council of Canada; University of Toronto","keywords":"Production (economics); Field (mathematics); Labour economics; Business; Industrial organization; Economics; Microeconomics; Mathematics","score_opus":0.04269693031442133,"score_gpt":0.30897317993284495,"score_spread":0.2662762496184236,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4243526881","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":[0.0030155275,0.030603006,0.00002571852,0.022499511,0.009916968,0.0054673487,0.92692673,0.00002481911,0.0015203516],"genre_scores_gemma":[0.10725583,0.1544527,0.0001963387,0.023003923,0.02914931,0.014944711,0.6607536,0.00023358394,0.010009974],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99312043,0.0013345964,0.0042443434,0.00076031004,0.00010986652,0.00043046012],"domain_scores_gemma":[0.96317303,0.032172758,0.003628424,0.00088712457,0.000049552495,0.000089102825],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01885756,0.0004934645,0.007363622,0.00038093887,0.00010021831,0.00030945035,0.0009920829,0.00038907712,0.005850863],"category_scores_gemma":[0.1142811,0.0002928127,0.0015440026,0.000110001674,0.0001236709,0.00042948558,0.0001627734,0.0003711302,0.0005940713],"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":[0.20081998,0.00007294433,0.00003130956,0.0000044660333,0.0011544304,0.000001618299,0.00008449436,0.000003020427,0.000015117862,0.0002381247,0.7974051,0.0001693426],"study_design_scores_gemma":[0.46752298,0.0000372417,0.00003335004,0.0003470893,0.00015064781,0.0000013868587,0.000039990497,0.000058964073,0.00011410474,0.0035936288,0.5277552,0.00034541948],"about_ca_topic_score_codex":0.0061181393,"about_ca_topic_score_gemma":0.0002901019,"teacher_disagreement_score":0.26964998,"about_ca_system_score_codex":0.0003230954,"about_ca_system_score_gemma":0.00011229336,"threshold_uncertainty_score":0.9999524},"labels":[],"label_agreement":null},{"id":"W4243642979","doi":"10.1257/rct.1402-3.0","title":"Team Production in International Labor Markets: Experimental Evidence from the Field","year":2016,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Economic Policies and Impacts","field":"Economics, Econometrics and Finance","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada; University of Toronto","keywords":"Production (economics); Field (mathematics); Economics; Business; Labour economics; Macroeconomics; Mathematics","score_opus":0.04269693031442133,"score_gpt":0.30897317993284495,"score_spread":0.2662762496184236,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4243642979","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":[0.0030155275,0.030603006,0.00002571852,0.022499511,0.009916968,0.0054673487,0.92692673,0.00002481911,0.0015203516],"genre_scores_gemma":[0.10725583,0.1544527,0.0001963387,0.023003923,0.02914931,0.014944711,0.6607536,0.00023358394,0.010009974],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99312043,0.0013345964,0.0042443434,0.00076031004,0.00010986652,0.00043046012],"domain_scores_gemma":[0.96317303,0.032172758,0.003628424,0.00088712457,0.000049552495,0.000089102825],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01885756,0.0004934645,0.007363622,0.00038093887,0.00010021831,0.00030945035,0.0009920829,0.00038907712,0.005850863],"category_scores_gemma":[0.1142811,0.0002928127,0.0015440026,0.000110001674,0.0001236709,0.00042948558,0.0001627734,0.0003711302,0.0005940713],"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":[0.20081998,0.00007294433,0.00003130956,0.0000044660333,0.0011544304,0.000001618299,0.00008449436,0.000003020427,0.000015117862,0.0002381247,0.7974051,0.0001693426],"study_design_scores_gemma":[0.46752298,0.0000372417,0.00003335004,0.0003470893,0.00015064781,0.0000013868587,0.000039990497,0.000058964073,0.00011410474,0.0035936288,0.5277552,0.00034541948],"about_ca_topic_score_codex":0.0061181393,"about_ca_topic_score_gemma":0.0002901019,"teacher_disagreement_score":0.26964998,"about_ca_system_score_codex":0.0003230954,"about_ca_system_score_gemma":0.00011229336,"threshold_uncertainty_score":0.9999524},"labels":[],"label_agreement":null},{"id":"W4243803868","doi":"10.1257/rct.3199-1.0","title":"One in a Million: Field Experiments on Perceived Closeness of the Election and Voter Turnout","year":2018,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Electoral Systems and Political Participation","field":"Social 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":"University of Toronto; Social Sciences and Humanities Research Council","funders":"Washington Center for Equitable Growth; National Science Foundation","keywords":"Turnout; Voter turnout; Closeness; Field (mathematics); Political science; Mathematics; Voting; Law; Politics; Mathematical analysis","score_opus":0.057726267231075,"score_gpt":0.38745287009397944,"score_spread":0.32972660286290445,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4243803868","genre_codex":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.44040108,0.005130455,0.000061826526,0.019763427,0.014115466,0.06305206,0.45434105,0.00010102784,0.0030335882],"genre_scores_gemma":[0.9805774,0.0003623659,0.0000058852793,0.0004196375,0.0018174555,0.0013715115,0.0149616795,0.000013174121,0.0004709101],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9854045,0.0118144285,0.0016413672,0.00027052625,0.00055309164,0.0003160734],"domain_scores_gemma":[0.9866634,0.011799552,0.0010336156,0.00027640857,0.0001456627,0.00008138708],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011151822,0.00022676079,0.00427967,0.00012709678,0.00020377699,0.000079347614,0.00020396269,0.0004637367,0.000272933],"category_scores_gemma":[0.022864548,0.0001283833,0.0007187342,0.0001781801,0.00025087583,0.0000552162,0.000025394385,0.00022043886,0.000009241398],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.6770959,0.00027496862,0.000022111262,0.00006408746,0.0008876113,0.0000012138855,0.0014118607,0.0000014341468,0.0006752715,0.0013000991,0.3179617,0.00030372766],"study_design_scores_gemma":[0.95516163,0.00036779753,0.00011487847,0.0008053091,0.0012212652,2.8408627e-7,0.000067165085,0.000023638682,0.0007221835,0.0042119483,0.036974575,0.00032933068],"about_ca_topic_score_codex":0.031954933,"about_ca_topic_score_gemma":0.0113467965,"teacher_disagreement_score":0.5401763,"about_ca_system_score_codex":0.00008859701,"about_ca_system_score_gemma":0.00017922942,"threshold_uncertainty_score":0.9853663},"labels":[],"label_agreement":null},{"id":"W4244497551","doi":"10.1257/rct.1668-1.0","title":"Hiring Discrimination Against Maghrebi Women in the National Capital Region of Quebec, Canada: An Econometric Analysis on Field Experiment Data","year":2016,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Multiculturalism, Politics, Migration, Gender","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Econometric analysis; Field (mathematics); Capital (architecture); Econometric model; Geography; Political science; Economics; Regional science; Econometrics; Archaeology; Mathematics","score_opus":0.16460157577311718,"score_gpt":0.4151256649800849,"score_spread":0.2505240892069677,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4244497551","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":[0.043217108,0.0009025357,0.000065628315,0.012306144,0.001519503,0.01073913,0.93003696,0.00002656231,0.0011864428],"genre_scores_gemma":[0.33716726,0.0012196102,0.000011706238,0.0031428474,0.0014955839,0.0015874044,0.65366215,0.000018670466,0.0016947679],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98452914,0.01026224,0.0025644682,0.0006176457,0.0015089699,0.0005175678],"domain_scores_gemma":[0.97218883,0.024015706,0.002426754,0.0008200153,0.00039523875,0.00015342946],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.026520597,0.00038070139,0.0046884404,0.0011510296,0.0002432512,0.00022731446,0.0013129829,0.00029271463,0.00034885894],"category_scores_gemma":[0.05258107,0.00020766463,0.00071693625,0.0004945018,0.00023853869,0.00039959876,0.000074350624,0.00019986989,0.0000033109322],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.061818697,0.00024518062,0.00000808556,0.000022196422,0.0041850493,0.000008258107,0.011913338,0.00020015915,0.0000049561045,0.0013252833,0.9198819,0.00038693578],"study_design_scores_gemma":[0.79342335,0.00006630288,0.00014636443,0.00011577231,0.0042934306,4.888131e-7,0.023938812,0.000449737,0.000034407443,0.00035650234,0.17632939,0.0008454405],"about_ca_topic_score_codex":0.79324174,"about_ca_topic_score_gemma":0.9633748,"teacher_disagreement_score":0.74355245,"about_ca_system_score_codex":0.0037144015,"about_ca_system_score_gemma":0.0038301884,"threshold_uncertainty_score":0.9713031},"labels":[],"label_agreement":null},{"id":"W4244876586","doi":"10.1257/rct.4111-4.0","title":"Business Boost, the Cavendish Enterprise productivity improvement trial; using leading management practices with young businesses with fewer than 20 employees to improve productivity.","year":2019,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Quality and Supply Management","field":"Business, Management and Accounting","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Economic and Social Research Council; Innovation Growth Lab; University College Cork; University of Leeds; Queen's University; University of Warwick","keywords":"Productivity; Business; Operations management; Industrial organization; Engineering; Economics; Economic growth","score_opus":0.03698270223453807,"score_gpt":0.3007789889833197,"score_spread":0.2637962867487816,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4244876586","genre_codex":"protocol","genre_gemma":"dataset","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"dataset","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.082630344,0.0033621946,0.013086967,0.039480094,0.03536275,0.46253517,0.35705587,0.0012730012,0.0052135927],"genre_scores_gemma":[0.12356934,0.0016840848,0.0018052865,0.006699055,0.047995824,0.08060129,0.7114697,0.0014785087,0.02469691],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9853326,0.0026406543,0.004321461,0.003065483,0.0029693365,0.0016704615],"domain_scores_gemma":[0.976203,0.004852707,0.013527762,0.003547268,0.0017673294,0.000101931735],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":["metaresearch","metaepi_narrow"],"category_scores_codex":[0.037516013,0.0023916895,0.013489946,0.0017307585,0.00097244926,0.004756123,0.0020234268,0.00039431467,0.00022547311],"category_scores_gemma":[0.014729831,0.001253199,0.0016008898,0.0022192749,0.00047072614,0.0032474513,0.0013341828,0.00095123745,0.0002380726],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.77597463,0.0013806848,0.000022721892,0.0029511538,0.010986547,0.00010470314,0.00006633275,0.0012045285,0.00010708269,0.00015588677,0.20624451,0.0008012351],"study_design_scores_gemma":[0.6636423,0.00028153986,0.000011232875,0.0011957719,0.021042615,0.000008572095,0.0003448545,0.00013736825,0.000028622684,0.0001618746,0.31157878,0.00156649],"about_ca_topic_score_codex":0.0066291643,"about_ca_topic_score_gemma":0.003741283,"teacher_disagreement_score":0.3819339,"about_ca_system_score_codex":0.00033875406,"about_ca_system_score_gemma":0.0004307793,"threshold_uncertainty_score":0.99998575},"labels":[],"label_agreement":null},{"id":"W4253173572","doi":"10.1257/rct.5228-1.0","title":"Intergroup interaction and attitudes to migrants","year":2020,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Migration and Labor Dynamics","field":"Social 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":"Toronto Metropolitan University","funders":"Norges Forskningsråd","keywords":"Psychology; Geography; Social psychology; Demographic economics; Sociology; Economics","score_opus":0.03952353624129736,"score_gpt":0.38057300990201093,"score_spread":0.3410494736607136,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4253173572","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":[0.00029453516,0.0010214525,0.0004963098,0.009902915,0.0037233483,0.012110585,0.9715835,0.00012855334,0.0007387923],"genre_scores_gemma":[0.012148357,0.016298711,0.00026799197,0.005958284,0.0036808674,0.0030172146,0.9561779,0.00005253338,0.0023981307],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.986539,0.009773722,0.0021962237,0.00052591535,0.00062704855,0.00033809873],"domain_scores_gemma":[0.9820149,0.015724374,0.0013945688,0.00027226043,0.00021696603,0.0003769376],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.017343475,0.00043715892,0.008165269,0.00035523612,0.0003211062,0.00059543416,0.00039517312,0.00042320325,0.0010821825],"category_scores_gemma":[0.053658463,0.0003005976,0.0013687053,0.00026929748,0.00019679508,0.00016274911,0.00008298384,0.00038244342,0.00035297286],"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":[0.291288,0.000037156966,4.0166552e-7,0.000019653353,0.0010074558,0.0000054818515,0.0012346607,0.000003684246,0.000010323843,0.00064264965,0.7051606,0.0005899512],"study_design_scores_gemma":[0.39082998,0.000032275257,3.5240316e-7,0.00010581386,0.0010556142,4.6790137e-7,0.00032952655,0.000053424224,8.1447035e-7,0.00013331206,0.6072493,0.00020909308],"about_ca_topic_score_codex":0.0031592909,"about_ca_topic_score_gemma":0.030825133,"teacher_disagreement_score":0.099542,"about_ca_system_score_codex":0.00009707459,"about_ca_system_score_gemma":0.00025568652,"threshold_uncertainty_score":0.9999446},"labels":[],"label_agreement":null},{"id":"W4253233741","doi":"10.1257/rct.843","title":"The Effects of Child Care Subsidies on Women’s Economic Opportunities in the Slums of Nairobi","year":2015,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Poverty, Education, and Child Welfare","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"African Population and Health Research Center; Department for International Development; Government of the United Kingdom; Comic Relief; Wellcome Trust; Styrelsen för Internationellt Utvecklingssamarbete; Bill and Melinda Gates Foundation; William and Flora Hewlett Foundation; International Development Research Centre; McGill University; Rockefeller Foundation","keywords":"Subsidy; Child care; Economic growth; Economics; Business; Socioeconomics; Medicine; Pediatrics; Market economy","score_opus":0.024556995036440887,"score_gpt":0.30411249692190495,"score_spread":0.2795555018854641,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4253233741","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":[0.007959304,0.09305337,4.2475008e-7,0.010852023,0.009968274,0.025489226,0.8327469,0.00003862494,0.019891802],"genre_scores_gemma":[0.16573046,0.39597753,0.0000036951662,0.002009557,0.0061254976,0.016450115,0.41086543,0.00009966366,0.0027380714],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9783488,0.017708795,0.0024069177,0.0003046126,0.00075517053,0.00047569987],"domain_scores_gemma":[0.9413677,0.054624557,0.002920006,0.00073727255,0.00024581683,0.00010465203],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.035011265,0.00042885996,0.0068938606,0.00027622867,0.0006334919,0.00014586007,0.0012831023,0.0003534069,0.00012105089],"category_scores_gemma":[0.030472172,0.00020567025,0.0014536238,0.0001161932,0.0008947678,0.00009521565,0.000033588978,0.00031791735,0.000010758213],"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":[0.19527915,0.000109068504,0.0000013433137,0.00012911543,0.0012498358,0.000002133541,0.02351623,0.0000062215986,2.472998e-7,0.0042227725,0.77498865,0.00049525104],"study_design_scores_gemma":[0.389202,0.00009281208,0.0000030509966,0.00021967039,0.00070355175,2.915946e-7,0.023407469,4.0256845e-7,0.000006128297,0.0010796604,0.58512294,0.00016201544],"about_ca_topic_score_codex":0.008093303,"about_ca_topic_score_gemma":0.01639501,"teacher_disagreement_score":0.42188153,"about_ca_system_score_codex":0.00034078336,"about_ca_system_score_gemma":0.002001003,"threshold_uncertainty_score":0.9985119},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"randomized_trial","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"randomized_trial","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4256692176","doi":"10.1257/rct.843-4.0","title":"The Effects of Child Care Subsidies on Women’s Economic Opportunities in the Slums of Nairobi","year":2015,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Poverty, Education, and Child Welfare","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"African Population and Health Research Center; Department for International Development; Government of the United Kingdom; Comic Relief; Wellcome Trust; Styrelsen för Internationellt Utvecklingssamarbete; Bill and Melinda Gates Foundation; William and Flora Hewlett Foundation; International Development Research Centre; McGill University; Rockefeller Foundation","keywords":"Subsidy; Child care; Demographic economics; Business; Economic growth; Socioeconomics; Economics; Medicine; Pediatrics; Market economy","score_opus":0.024556995036440887,"score_gpt":0.30411249692190495,"score_spread":0.2795555018854641,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4256692176","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":[0.007959304,0.09305337,4.2475008e-7,0.010852023,0.009968274,0.025489226,0.8327469,0.00003862494,0.019891802],"genre_scores_gemma":[0.16573046,0.39597753,0.0000036951662,0.002009557,0.0061254976,0.016450115,0.41086543,0.00009966366,0.0027380714],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9783488,0.017708795,0.0024069177,0.0003046126,0.00075517053,0.00047569987],"domain_scores_gemma":[0.9413677,0.054624557,0.002920006,0.00073727255,0.00024581683,0.00010465203],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.035011265,0.00042885996,0.0068938606,0.00027622867,0.0006334919,0.00014586007,0.0012831023,0.0003534069,0.00012105089],"category_scores_gemma":[0.030472172,0.00020567025,0.0014536238,0.0001161932,0.0008947678,0.00009521565,0.000033588978,0.00031791735,0.000010758213],"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":[0.19527915,0.000109068504,0.0000013433137,0.00012911543,0.0012498358,0.000002133541,0.02351623,0.0000062215986,2.472998e-7,0.0042227725,0.77498865,0.00049525104],"study_design_scores_gemma":[0.389202,0.00009281208,0.0000030509966,0.00021967039,0.00070355175,2.915946e-7,0.023407469,4.0256845e-7,0.000006128297,0.0010796604,0.58512294,0.00016201544],"about_ca_topic_score_codex":0.008093303,"about_ca_topic_score_gemma":0.01639501,"teacher_disagreement_score":0.42188153,"about_ca_system_score_codex":0.00034078336,"about_ca_system_score_gemma":0.002001003,"threshold_uncertainty_score":0.9985119},"labels":[],"label_agreement":null},{"id":"W4285076116","doi":"10.1257/rct.3147","title":"Welfare effects of a non-contributory old age pension: experimental evidence for Ekiti State in Nigeria","year":2019,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Financial Literacy, Pension, Retirement Analysis","field":"Business, Management and Accounting","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University","funders":"Department for International Development; International Development Research Centre; Government of Canada","keywords":"Pension; Welfare; Welfare state; Demographic economics; State (computer science); Economics; Political science; Computer science; Law; Finance; Market economy","score_opus":0.03242950834181379,"score_gpt":0.31925894029807694,"score_spread":0.28682943195626315,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285076116","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":[0.1350355,0.03172033,0.00028851238,0.0006122463,0.01679817,0.1414974,0.67366004,0.00022267438,0.0001651502],"genre_scores_gemma":[0.23313487,0.00090486207,0.00011031089,0.0011212862,0.002935731,0.010062502,0.75063354,0.0001621993,0.0009346932],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9891949,0.0016000799,0.005959793,0.0012088423,0.0011636725,0.0008726951],"domain_scores_gemma":[0.9673944,0.023404106,0.007169772,0.0012297228,0.00075628964,0.000045716206],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.021840604,0.001233692,0.020654442,0.0017555121,0.00021130856,0.00048950716,0.0009545063,0.0005689277,0.0009365739],"category_scores_gemma":[0.05748659,0.0008990977,0.0054118834,0.00068445347,0.00018162478,0.0009865649,0.00037184608,0.0004023305,0.00018341182],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.6123917,0.0003189788,0.000099378194,0.0018277505,0.00097645615,0.00012413025,0.00004094762,0.00003900112,0.0014674425,0.00008884495,0.3825745,0.000050898525],"study_design_scores_gemma":[0.93488896,0.00010450303,0.00007561964,0.0034497178,0.006080483,6.4699833e-7,0.00002707986,0.00072732545,0.0004540824,0.00037422805,0.052809186,0.0010081675],"about_ca_topic_score_codex":0.0028793789,"about_ca_topic_score_gemma":0.00058325136,"teacher_disagreement_score":0.3297653,"about_ca_system_score_codex":0.00017458612,"about_ca_system_score_gemma":0.0001794275,"threshold_uncertainty_score":0.9999767},"labels":[],"label_agreement":null},{"id":"W4285582584","doi":"10.1257/rct.6797-1.0","title":"Consumer Surplus of Alternative Payment Methods: Paying Uber with Cash","year":2020,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","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":"Bank of Canada","funders":"","keywords":"Payment; Cash; Economic surplus; Economics; Business; Commerce; Finance; Market economy","score_opus":0.1061276893956362,"score_gpt":0.3669807349822568,"score_spread":0.2608530455866206,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285582584","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":[0.000013769601,0.00989892,0.010560907,0.00452797,0.0018740043,0.012668164,0.9586535,0.000040284332,0.0017624547],"genre_scores_gemma":[0.0037722534,0.039418355,0.010508018,0.0086114295,0.0022706345,0.012400238,0.9213212,0.00024118162,0.0014567074],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9851908,0.0057370104,0.007231829,0.0010064981,0.00019467664,0.0006391761],"domain_scores_gemma":[0.96941936,0.020305846,0.008927903,0.0008889943,0.00017952627,0.00027839787],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.038493358,0.0008099275,0.023051247,0.00076488586,0.00011934869,0.00013524728,0.0006212828,0.00040192658,0.00087901234],"category_scores_gemma":[0.02685314,0.00056282844,0.0026061079,0.0003133708,0.0002344289,0.00011807501,0.00018358276,0.00054926757,0.0005534085],"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":[0.31607836,0.00011802132,0.0000015787415,0.00033915034,0.013712052,0.00003694211,0.00016874484,0.000042460488,0.0000011338948,0.015050824,0.6535426,0.000908135],"study_design_scores_gemma":[0.4673015,0.00009254059,3.0925497e-7,0.00009780547,0.0012829087,0.0000017796791,0.000024825817,0.00014293626,0.000018836747,0.005694618,0.5249779,0.0003640185],"about_ca_topic_score_codex":0.012731179,"about_ca_topic_score_gemma":0.00027629943,"teacher_disagreement_score":0.15122311,"about_ca_system_score_codex":0.00017594495,"about_ca_system_score_gemma":0.00023319745,"threshold_uncertainty_score":0.9996823},"labels":[],"label_agreement":null},{"id":"W4301912413","doi":"10.1257/rct.2299-4.0","title":"Employment Discrimination against Indigenous Peoples in the United States: Evidence from a Field Experiment","year":2017,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Labor market dynamics and wage inequality","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Cancer Institute; National Institutes of Health; University of Victoria; Tulane University","keywords":"Indigenous; Field (mathematics); Political science; Geography; Demographic economics; Economics; Biology; Mathematics; Ecology","score_opus":0.06860678150770705,"score_gpt":0.3297667322504887,"score_spread":0.26115995074278164,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4301912413","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":[0.0633453,0.016133187,0.0006986907,0.0024519654,0.001837547,0.010456806,0.9049259,0.000026827955,0.0001237738],"genre_scores_gemma":[0.07422539,0.055576075,0.00005146875,0.0017744771,0.0005648116,0.0055607203,0.86207426,0.000041147454,0.00013166593],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98808014,0.0046670423,0.0054762578,0.0009165741,0.00029873464,0.00056122156],"domain_scores_gemma":[0.94420755,0.047414448,0.0063542216,0.0018095343,0.00011939975,0.00009482381],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.041765723,0.0007230238,0.010406016,0.00076422247,0.00034943203,0.0010015873,0.0016519206,0.0005255307,0.0003811842],"category_scores_gemma":[0.063658684,0.00046101335,0.0018997821,0.00021530682,0.00011817779,0.0002457027,0.00017497297,0.0005766856,0.000097756],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.58856535,0.0011350979,0.00016482148,0.0002084021,0.005609772,0.0001479343,0.011659149,0.00033908224,0.000006017224,0.0029684475,0.38858408,0.0006118566],"study_design_scores_gemma":[0.88435316,0.00017553885,0.00007410339,0.0009266796,0.0010892759,8.713729e-7,0.00069730723,0.0014784168,0.000008678565,0.030150741,0.079974644,0.0010705715],"about_ca_topic_score_codex":0.050960083,"about_ca_topic_score_gemma":0.0053787557,"teacher_disagreement_score":0.30860943,"about_ca_system_score_codex":0.00027217524,"about_ca_system_score_gemma":0.00019089726,"threshold_uncertainty_score":0.9997842},"labels":[],"label_agreement":null},{"id":"W4319311206","doi":"10.1257/rct.10357-1.2","title":"Third-Party Performance Pay to Improve Local Government Accountability: A Field Experiment in Burkina Faso","year":2022,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Local Government Finance and Decentralization","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":"Impact","funders":"","keywords":"Accountability; Government (linguistics); Public administration; Business; Field (mathematics); Local government; Political science; Law; Mathematics","score_opus":0.01856049604180564,"score_gpt":0.3197619804529984,"score_spread":0.30120148441119277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4319311206","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":[0.0037821033,0.0025864968,0.0016848235,0.0063323793,0.010385536,0.046811964,0.92418534,0.00015188803,0.0040794555],"genre_scores_gemma":[0.24247037,0.05293929,0.00019864808,0.023153672,0.0066179484,0.10632867,0.5549313,0.00026286812,0.013097251],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9839728,0.0070663225,0.0035197388,0.0009980991,0.0034766681,0.00096635905],"domain_scores_gemma":[0.9847807,0.012212014,0.0018348241,0.0008083865,0.00010411976,0.00025990198],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.032192208,0.00066660816,0.008007691,0.00015734967,0.0005197009,0.00030697338,0.00090283,0.00054852373,0.016409626],"category_scores_gemma":[0.021094348,0.00043460118,0.001593637,0.00048404612,0.00020940666,0.00029059124,0.00031104128,0.0006440962,0.000059166756],"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":[0.38741636,0.0003261993,0.000013709141,0.000035022695,0.00037077593,0.000011303407,0.0009292773,0.00014272325,0.000017430908,0.0002496182,0.60750645,0.0029811584],"study_design_scores_gemma":[0.38793987,0.00019385687,0.000002080327,0.000068139336,0.00045725782,2.0481399e-7,0.0009747957,0.0000900813,0.00015728407,0.00012209847,0.6095977,0.00039665244],"about_ca_topic_score_codex":0.0093334,"about_ca_topic_score_gemma":0.0028729301,"teacher_disagreement_score":0.36925408,"about_ca_system_score_codex":0.0022669588,"about_ca_system_score_gemma":0.0008252994,"threshold_uncertainty_score":0.9998106},"labels":[],"label_agreement":null},{"id":"W4321446580","doi":"10.1257/rct.10684-1.0","title":"Dual-Client Workforce Development Model: NPower Canada Randomized Control Trial","year":2023,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Ultrasound in Clinical Applications","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Workforce; Dual (grammatical number); Control (management); Randomized controlled trial; Workforce development; Business; Operations management; Computer science; Engineering; Medicine; Economic growth; Economics; Internal medicine; Artificial intelligence; Art","score_opus":0.05692895259477479,"score_gpt":0.3632131569965749,"score_spread":0.30628420440180015,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4321446580","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":[0.00027813992,0.0022150138,0.0035366926,0.00516375,0.007033921,0.11173378,0.8686819,0.00040781553,0.00094898883],"genre_scores_gemma":[0.0010819059,0.0032434429,0.0008620572,0.004411694,0.0033617045,0.07785276,0.89661,0.00024754752,0.01232884],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.96333474,0.012069977,0.01784653,0.0018945769,0.0033479105,0.0015062873],"domain_scores_gemma":[0.67900765,0.30575594,0.009776198,0.0028884187,0.001299306,0.0012724819],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.07663697,0.0019690427,0.057904504,0.0006913727,0.00045549622,0.00025779931,0.00081618724,0.0016084875,0.0010238809],"category_scores_gemma":[0.27216217,0.0011886937,0.010501163,0.00062492024,0.0006970435,0.00008310651,0.00016948707,0.0018907699,0.0011107672],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.50595784,0.0003721587,3.8967357e-8,0.000050370738,0.012050403,0.00003642291,0.000011228259,0.00026525563,0.0000025980314,0.00013428341,0.4810537,0.00006570741],"study_design_scores_gemma":[0.7025562,0.000102157,8.053276e-8,0.00035124025,0.025526725,0.000009894143,0.000009774305,0.0017235144,0.0000054773272,0.00077203644,0.26816764,0.0007752847],"about_ca_topic_score_codex":0.044574033,"about_ca_topic_score_gemma":0.047573145,"teacher_disagreement_score":0.29368597,"about_ca_system_score_codex":0.00093155087,"about_ca_system_score_gemma":0.012410321,"threshold_uncertainty_score":0.9998893},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"randomized_trial","genre":"dataset","about_ca_system":false,"about_ca_topic":true,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4388262510","doi":"10.1257/rct.4295-3.0","title":"Soft and hard commitment devices to increase HIV testing","year":2019,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"University of Toronto; University of Minnesota","keywords":"Human immunodeficiency virus (HIV); Psychology; Computer science; Virology; Medicine","score_opus":0.06905619605726096,"score_gpt":0.3787510489318139,"score_spread":0.3096948528745529,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388262510","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":[0.00049956545,0.0067447666,0.00034357546,0.0031790247,0.0003449496,0.027956441,0.9591031,0.0000702923,0.0017583013],"genre_scores_gemma":[0.00065716734,0.0010478451,0.0010883025,0.0014289292,0.0013771927,0.006443525,0.9610659,0.000070934526,0.02682019],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98761135,0.0066499067,0.003399154,0.00072787364,0.0009382443,0.00067347725],"domain_scores_gemma":[0.94186664,0.053412307,0.0017228641,0.0012160162,0.0007177617,0.0010644331],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.025942553,0.0006987074,0.015872452,0.0008926563,0.00019657533,0.00030083657,0.0003603414,0.00039192266,0.002170463],"category_scores_gemma":[0.20504335,0.00041766718,0.002776107,0.00027143303,0.00015225832,0.000099797515,0.0002480243,0.0006951064,0.0012732594],"study_design_candidate":"randomized_trial","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":[0.40905124,0.00017048698,0.0000062309236,0.00027276916,0.003766239,0.000046449175,0.000010360598,0.0000011488296,0.00009133996,0.000006013939,0.58505446,0.0015232575],"study_design_scores_gemma":[0.5993959,0.0008739086,0.0000032133764,0.0013478192,0.005818656,0.000022506094,0.000023404353,0.00032689152,0.000010387064,0.000034283476,0.3919032,0.00023985971],"about_ca_topic_score_codex":0.0010390852,"about_ca_topic_score_gemma":0.00012293053,"teacher_disagreement_score":0.19315128,"about_ca_system_score_codex":0.00010983776,"about_ca_system_score_gemma":0.0005305128,"threshold_uncertainty_score":0.9998275},"labels":[],"label_agreement":null},{"id":"W4391765013","doi":"10.1257/rct.977-3.1","title":"Senegal Behavior Change Campaign & Solar Lights Evaluation","year":2015,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"ICT Impact and Policies","field":"Engineering","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":"Impact","funders":"Leibniz-Gemeinschaft","keywords":"Environmental science; Meteorology; Atmospheric sciences; Geography; Physics","score_opus":0.10477470317291927,"score_gpt":0.35873998119630307,"score_spread":0.2539652780233838,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391765013","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":[0.001338154,0.015254369,0.000079619356,0.000096170355,0.007890355,0.029977763,0.94454217,0.00036630637,0.00045506828],"genre_scores_gemma":[0.0021682892,0.0026932042,0.000032577038,0.00016545481,0.0061545772,0.030313121,0.9578724,0.000125332,0.00047506465],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98726326,0.0069472757,0.003254288,0.00026673864,0.0014783038,0.0007901064],"domain_scores_gemma":[0.9889362,0.007923264,0.0012357249,0.0010118322,0.00053964846,0.00035335592],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.037261635,0.0011195486,0.014436586,0.0008015161,0.00014140854,0.00034367267,0.000510887,0.001095448,0.00152253],"category_scores_gemma":[0.017227875,0.00074015855,0.0029166308,0.00023191835,0.00009349334,0.00026372084,0.000053777974,0.0006128291,0.00094063516],"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":[0.14671983,0.00005591741,1.2197495e-7,0.000057293866,0.002460017,0.000004272041,0.00044437108,0.00013436763,0.00006183237,0.0000045221846,0.84940195,0.0006555226],"study_design_scores_gemma":[0.5547775,0.000028705936,3.7873275e-7,0.00009187791,0.012530677,0.000007772373,0.000019393914,0.00028153035,0.00004594984,0.00011724106,0.43156135,0.00053763646],"about_ca_topic_score_codex":0.0007398944,"about_ca_topic_score_gemma":0.00039732386,"teacher_disagreement_score":0.4178406,"about_ca_system_score_codex":0.00031109338,"about_ca_system_score_gemma":0.00035167648,"threshold_uncertainty_score":0.9998372},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"dataset","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4396930369","doi":"10.1257/rct.977-4.0","title":"Senegal Behavior Change Campaign & Solar Lights Evaluation","year":2015,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"ICT Impact and Policies","field":"Engineering","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":"Impact","funders":"","keywords":"Environmental science; Meteorology; Geography","score_opus":0.10477470317291927,"score_gpt":0.35873998119630307,"score_spread":0.2539652780233838,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396930369","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":[0.001338154,0.015254369,0.000079619356,0.000096170355,0.007890355,0.029977763,0.94454217,0.00036630637,0.00045506828],"genre_scores_gemma":[0.0021682892,0.0026932042,0.000032577038,0.00016545481,0.0061545772,0.030313121,0.9578724,0.000125332,0.00047506465],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98726326,0.0069472757,0.003254288,0.00026673864,0.0014783038,0.0007901064],"domain_scores_gemma":[0.9889362,0.007923264,0.0012357249,0.0010118322,0.00053964846,0.00035335592],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.037261635,0.0011195486,0.014436586,0.0008015161,0.00014140854,0.00034367267,0.000510887,0.001095448,0.00152253],"category_scores_gemma":[0.017227875,0.00074015855,0.0029166308,0.00023191835,0.00009349334,0.00026372084,0.000053777974,0.0006128291,0.00094063516],"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":[0.14671983,0.00005591741,1.2197495e-7,0.000057293866,0.002460017,0.000004272041,0.00044437108,0.00013436763,0.00006183237,0.0000045221846,0.84940195,0.0006555226],"study_design_scores_gemma":[0.5547775,0.000028705936,3.7873275e-7,0.00009187791,0.012530677,0.000007772373,0.000019393914,0.00028153035,0.00004594984,0.00011724106,0.43156135,0.00053763646],"about_ca_topic_score_codex":0.0007398944,"about_ca_topic_score_gemma":0.00039732386,"teacher_disagreement_score":0.4178406,"about_ca_system_score_codex":0.00031109338,"about_ca_system_score_gemma":0.00035167648,"threshold_uncertainty_score":0.9998372},"labels":[],"label_agreement":null},{"id":"W4402469220","doi":"10.1257/rct.14295","title":"Consumer Responseto Mortgage Payment Shocks: evidence from Canada","year":2024,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Housing Market and Economics","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Payment; Mortgage underwriting; Business; Shared appreciation mortgage; Commercial mortgage-backed security; Mortgage insurance; Collateralized mortgage obligation; Economics; Financial system; Monetary economics; Finance","score_opus":0.04832301067264219,"score_gpt":0.2721686110028325,"score_spread":0.22384560033019033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402469220","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":[0.0009612749,0.058284592,0.00006822829,0.0015428972,0.011383121,0.0063218786,0.9187023,0.00008581248,0.0026499056],"genre_scores_gemma":[0.0068142833,0.06748182,0.00018798141,0.0018964384,0.002994581,0.0043456405,0.9129975,0.00022915186,0.0030526023],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98473287,0.0030297572,0.009260691,0.0017759693,0.00025767274,0.0009430653],"domain_scores_gemma":[0.9428976,0.049912676,0.0049500163,0.0017767189,0.00010010731,0.0003628413],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.04027988,0.0012451459,0.02358632,0.0007988878,0.00022940793,0.00089943025,0.0011527867,0.00083004526,0.008932424],"category_scores_gemma":[0.05937705,0.0010304276,0.0039451355,0.00026703466,0.00020596471,0.00023675933,0.00028356366,0.0010016938,0.00595106],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.2773454,0.000038070117,0.0000126543755,0.00006902885,0.0053073177,0.0001118297,0.000025126914,0.000029206156,0.000001388776,0.00013959272,0.71678436,0.00013605058],"study_design_scores_gemma":[0.37002364,0.000023084032,0.0000024113376,0.0004080204,0.0021917243,0.0000028542233,0.000015292246,0.00046106393,0.0000054776897,0.006320514,0.619722,0.0008239044],"about_ca_topic_score_codex":0.6252852,"about_ca_topic_score_gemma":0.3980269,"teacher_disagreement_score":0.22725828,"about_ca_system_score_codex":0.0010514647,"about_ca_system_score_gemma":0.001799516,"threshold_uncertainty_score":0.9992146},"labels":[],"label_agreement":null},{"id":"W4402469781","doi":"10.1257/rct.14295-1.0","title":"Consumer Responseto Mortgage Payment Shocks: evidence from Canada","year":2024,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Housing Market and Economics","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Payment; Business; Mortgage underwriting; Commercial mortgage-backed security; Mortgage insurance; Finance","score_opus":0.04832301067264219,"score_gpt":0.2721686110028325,"score_spread":0.22384560033019033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402469781","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":[0.0009612749,0.058284592,0.00006822829,0.0015428972,0.011383121,0.0063218786,0.9187023,0.00008581248,0.0026499056],"genre_scores_gemma":[0.0068142833,0.06748182,0.00018798141,0.0018964384,0.002994581,0.0043456405,0.9129975,0.00022915186,0.0030526023],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98473287,0.0030297572,0.009260691,0.0017759693,0.00025767274,0.0009430653],"domain_scores_gemma":[0.9428976,0.049912676,0.0049500163,0.0017767189,0.00010010731,0.0003628413],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.04027988,0.0012451459,0.02358632,0.0007988878,0.00022940793,0.00089943025,0.0011527867,0.00083004526,0.008932424],"category_scores_gemma":[0.05937705,0.0010304276,0.0039451355,0.00026703466,0.00020596471,0.00023675933,0.00028356366,0.0010016938,0.00595106],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.2773454,0.000038070117,0.0000126543755,0.00006902885,0.0053073177,0.0001118297,0.000025126914,0.000029206156,0.000001388776,0.00013959272,0.71678436,0.00013605058],"study_design_scores_gemma":[0.37002364,0.000023084032,0.0000024113376,0.0004080204,0.0021917243,0.0000028542233,0.000015292246,0.00046106393,0.0000054776897,0.006320514,0.619722,0.0008239044],"about_ca_topic_score_codex":0.6252852,"about_ca_topic_score_gemma":0.3980269,"teacher_disagreement_score":0.22725828,"about_ca_system_score_codex":0.0010514647,"about_ca_system_score_gemma":0.001799516,"threshold_uncertainty_score":0.9992146},"labels":[],"label_agreement":null},{"id":"W7077046891","doi":"10.1257/rct.4111-5.0","title":"Business Boost, the Cavendish Enterprise productivity improvement trial; using leading management practices with young businesses with fewer than 20 employees to improve productivity.","year":2019,"lang":"en","type":"dataset","venue":"AEA Randomized Controlled Trials","topic":"Pregnancy and preeclampsia studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Economic and Social Research Council; University College Cork; University of Leeds; Queen's University; University of Warwick","keywords":"Productivity; Work (physics); Business management; Human resource management; Context (archaeology)","score_opus":0.040245955389527625,"score_gpt":0.32792975628484355,"score_spread":0.2876838008953159,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7077046891","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":[0.058542542,0.020786429,0.0027594808,0.016707275,0.01549214,0.36849216,0.51519644,0.00048890983,0.0015346206],"genre_scores_gemma":[0.20033582,0.015815545,0.004491493,0.002267794,0.025494803,0.13418272,0.5884141,0.0013390367,0.027658703],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.986433,0.0046087326,0.0032720752,0.0023976706,0.0020602676,0.0012282279],"domain_scores_gemma":[0.97782224,0.008589119,0.008795842,0.0031342409,0.0014199617,0.00023856785],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.021261819,0.0019290564,0.017334586,0.000777949,0.000617258,0.00067685527,0.0008084438,0.00038692902,0.000056627076],"category_scores_gemma":[0.022413207,0.00089133467,0.0014660397,0.0010959368,0.0004975183,0.0007335628,0.0005398691,0.0009910926,0.00003552778],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.8826235,0.0012435304,0.000047462254,0.0022466532,0.026301991,0.00016947543,0.00023436095,0.00011182842,0.00048538603,0.000006377711,0.085774556,0.0007548561],"study_design_scores_gemma":[0.8922182,0.0017019775,0.0000277825,0.004814134,0.04569981,0.0000882405,0.00023061413,0.0000135576065,0.00036097466,0.000032903485,0.053687442,0.0011243668],"about_ca_topic_score_codex":0.001351011,"about_ca_topic_score_gemma":0.0011266526,"teacher_disagreement_score":0.23430945,"about_ca_system_score_codex":0.00033446992,"about_ca_system_score_gemma":0.0009914077,"threshold_uncertainty_score":0.9993537},"labels":[],"label_agreement":null}]}