{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":15,"total_is_capped":false,"direct_labels_cover":0,"predictions_cover":15,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"59d05407f7d7","filters":{"venue":"Meta-Psychology"}},"results":[{"id":"W2915455637","doi":"10.15626/mp.2018.843","title":"A Brief Guide to Evaluate Replications","year":2019,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":82,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University","funders":"","keywords":"Replication (statistics); Transparency (behavior); Independence (probability theory); Computer science; Similarity (geometry); Research design; Psychology; Statistics; Artificial intelligence; Mathematics; Computer security","retraction":null,"screen_n_in":null,"score":{"opus":0.8534897186781798,"gpt":0.6528829468946589,"spread":0.2006067717835209,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.1272678,0.0003035868,0.004216787,0.0005475528,0.00009063977,0.000318553,0.002883065,0.0001146175,0.1438041],"category_scores_gemma":[0.02839296,0.0001437781,0.002519505,0.002094109,0.00004468581,0.0001325465,0.0001885429,0.0001528961,0.18516],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001783192,"about_ca_system_score_gemma":0.00004458729,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001586258,"about_ca_topic_score_gemma":0.00002709265,"domain_scores_codex":[0.9754149,0.01016772,0.00841983,0.002148402,0.003450798,0.0003983247],"domain_scores_gemma":[0.9815844,0.00210902,0.002381533,0.01237196,0.001309111,0.0002440127],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000006619124,0.00005388484,0.004139873,0.000004398082,0.001076933,0.000002185167,0.0001223054,0.00004975092,0.001091133,0.004553144,0.9656626,0.02323717],"study_design_scores_gemma":[0.00019003,0.00006159129,0.03375987,0.000001886853,0.001091104,0.00002672239,0.00002549924,0.0002413702,0.00002047161,0.007166452,0.9572689,0.0001460868],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.09121804,0.002328424,0.1187117,0.05806079,0.001858036,0.003226637,0.00004320297,0.00003735765,0.7245158],"genre_scores_gemma":[0.3258395,0.00004037352,0.06468218,0.06392372,0.0002390996,0.0008322992,0.00001922018,0.00005082922,0.5443729],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.2346214,"threshold_uncertainty_score":0.9797913,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3030972095","doi":"10.15626/mp.2018.874","title":"Estimating Population Mean Power Under Conditions of Heterogeneity and Selection for Significance","year":2020,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":81,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"University of Toronto Mississauga; University of Toronto","keywords":"Statistics; Sample size determination; Mathematics; Population; Statistical power; Replication (statistics); Statistical significance; Power (physics); Selection (genetic algorithm); Statistical hypothesis testing; Econometrics; Sample (material); Set (abstract data type); Demography; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.8026212016523445,"gpt":0.5861197031220517,"spread":0.2165014985302928,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01519707,0.00016234,0.002011152,0.0001370942,0.0001093632,0.00008827395,0.0003387096,0.00007721407,0.002298162],"category_scores_gemma":[0.006050619,0.000090949,0.0008094744,0.0006045742,0.0000555111,0.0001296496,0.00002682809,0.00007237101,0.00008706954],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000006554323,"about_ca_system_score_gemma":0.00001004638,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009053762,"about_ca_topic_score_gemma":0.00003861888,"domain_scores_codex":[0.9924232,0.002606895,0.003328995,0.0007265203,0.0007742422,0.0001400884],"domain_scores_gemma":[0.9953901,0.001237656,0.002199555,0.0006162267,0.000463979,0.00009254907],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002567635,0.0007831965,0.4753292,0.0006746714,0.0190953,0.000004246079,0.004327067,0.0290678,0.1376724,0.07189561,0.205902,0.05499173],"study_design_scores_gemma":[0.001410799,0.000702395,0.7059123,0.00001705268,0.006877794,0.00004729047,0.0003158118,0.1421558,0.001132304,0.1083292,0.03244835,0.0006508935],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4299092,0.0003400165,0.5658498,0.002628494,0.0001517481,0.0005997015,0.00004106274,0.000007145998,0.0004728119],"genre_scores_gemma":[0.9592785,0.000002021501,0.03873875,0.00175179,0.00004194951,0.00007240704,0.00001343907,0.000008069647,0.00009305381],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5293693,"threshold_uncertainty_score":0.9986139,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4226236655","doi":"10.15626/mp.2020.2595","title":"Tutorial in Longitudinal Measurement Invariance and Cross-lagged Panel Models Using Lavaan","year":2022,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Psychological Well-being and Life Satisfaction","field":"Psychology","cited_by":76,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Concordia University; Dalhousie University","funders":"","keywords":"Structural equation modeling; Measurement invariance; Latent variable; Computer science; Panel data; Syntax; Econometrics; Process (computing); Test (biology); Interpretation (philosophy); Mathematics; Confirmatory factor analysis; Artificial intelligence; Machine learning; Programming language","retraction":null,"screen_n_in":null,"score":{"opus":0.5074782751927323,"gpt":0.4361204782108346,"spread":0.07135779698189765,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003176819,0.0003683101,0.0007644524,0.000322736,0.0003886387,0.00005494694,0.0003993866,0.0002684139,0.004398425],"category_scores_gemma":[0.0001071676,0.0003484417,0.0001913935,0.0005304298,0.0002300239,0.0001641403,0.00017055,0.0009718327,0.00005849426],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001735952,"about_ca_system_score_gemma":0.00003876567,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009671885,"about_ca_topic_score_gemma":0.0001036688,"domain_scores_codex":[0.995099,0.001579495,0.0007602516,0.001256011,0.0005687669,0.0007364638],"domain_scores_gemma":[0.9985957,0.0001361995,0.0002804858,0.0007552091,0.00008197362,0.0001504828],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01996961,0.01119458,0.5825195,0.0001647338,0.0125258,0.003675908,0.01764991,0.03475853,0.05596066,0.1362318,0.0523241,0.07302482],"study_design_scores_gemma":[0.009542273,0.001160223,0.8798217,0.00001284464,0.0008044144,0.0009580481,0.0002781312,0.001278509,0.0000412002,0.0496664,0.05530092,0.001135348],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9397391,0.003718911,0.008113062,0.001320651,0.01057021,0.0007233852,0.00005563908,0.0001844276,0.0355746],"genre_scores_gemma":[0.9960282,0.0000379091,0.0006900579,0.002058639,0.0004272866,0.0003295528,0.00001106767,0.00004282056,0.0003744565],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2973021,"threshold_uncertainty_score":0.9998968,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4294233607","doi":"10.15626/mp.2021.2720","title":"Z-curve 2.0: Estimating Replication Rates and Discovery Rates","year":2022,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":60,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Replication (statistics); Confidence interval; False discovery rate; Statistics; Selection (genetic algorithm); False positive paradox; Multiple comparisons problem; Computer science; Range (aeronautics); Extension (predicate logic); Statistical power; Econometrics; Mathematics; Artificial intelligence; Biology; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.7276317683307449,"gpt":0.64387429511749,"spread":0.083757473213255,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.007435662,0.0002077712,0.0008794412,0.00008822545,0.0002469225,0.00005371817,0.0003175724,0.00009464981,0.001365496],"category_scores_gemma":[0.06030139,0.0001702773,0.0001639292,0.000266899,0.0002123765,0.00009941067,0.0002914481,0.0004343012,0.00002245408],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002315106,"about_ca_system_score_gemma":0.00002062444,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008719313,"about_ca_topic_score_gemma":0.000001840573,"domain_scores_codex":[0.9951055,0.002602671,0.0009406981,0.0007995249,0.0002542006,0.0002974129],"domain_scores_gemma":[0.9731434,0.0252068,0.0004562248,0.001055938,0.00005755076,0.00008010906],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0009231677,0.001814323,0.008078642,0.0003717468,0.004676923,0.0001154824,0.001116703,0.00007334703,0.009564796,0.7168034,0.102731,0.1537306],"study_design_scores_gemma":[0.0006487608,0.0002309523,0.002684729,0.000003819187,0.0007278182,0.00004508838,0.00002988985,0.0006837421,0.0003296986,0.9889766,0.005443143,0.0001957464],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.06307341,0.0007893696,0.9164829,0.01145749,0.002408108,0.001109586,0.000275743,0.0002459409,0.004157405],"genre_scores_gemma":[0.05193904,0.00002354443,0.9439685,0.002491086,0.0002230139,0.0005782188,0.00001088853,0.00004708385,0.0007186732],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2721733,"threshold_uncertainty_score":0.9995474,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3135512811","doi":"10.15626/mp.2019.1645","title":"The Validation Crisis in Psychology","year":2021,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Mental Health Research Topics","field":"Psychology","cited_by":52,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"Social Sciences and Humanities Research Council of Canada; University of Toronto Mississauga; University of Toronto","keywords":"Nomological network; Construct (python library); Construct validity; Cronbach's alpha; Psychology; Criterion validity; External validity; Metatheory; Predictive validity; Incremental validity; Psychometrics; Structural equation modeling; Social psychology; Computer science; Developmental psychology; Machine learning","retraction":null,"screen_n_in":null,"score":{"opus":0.3490952054780505,"gpt":0.5715458269845495,"spread":0.222450621506499,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00268299,0.0002732209,0.000600559,0.0002567616,0.0002463453,0.0000464754,0.0007062502,0.0003985816,0.01260069],"category_scores_gemma":[0.0003163026,0.0002060631,0.000238379,0.0009354934,0.0002432322,0.00008549315,0.0001422984,0.0009088328,0.003012215],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006502139,"about_ca_system_score_gemma":0.00008001326,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001413965,"about_ca_topic_score_gemma":0.000461715,"domain_scores_codex":[0.9935184,0.003061426,0.0008945133,0.001031356,0.0003631493,0.001131205],"domain_scores_gemma":[0.9969552,0.0006096183,0.0001838156,0.00189002,0.000158655,0.0002027408],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000752087,0.00170711,0.01309052,0.0000481255,0.001651534,0.001289769,0.001497832,0.000003218763,0.001660726,0.04383012,0.6127722,0.3216968],"study_design_scores_gemma":[0.001739493,0.0002286281,0.1052832,0.000003240993,0.0001296941,0.000268319,0.0003044461,0.000003977334,0.0003047508,0.01810754,0.8734129,0.0002138094],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1573514,0.04168805,0.00397775,0.3177763,0.01597501,0.001923883,0.00007296463,0.000275213,0.4609595],"genre_scores_gemma":[0.8507349,0.006304145,0.0038998,0.1071386,0.001162129,0.00262625,0.0002032004,0.0002006988,0.02773028],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6933835,"threshold_uncertainty_score":0.9977641,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4308725914","doi":"10.15626/mp.2021.2803","title":"Meta-Analytic Findings of the Self-Controlled Motor Learning Literature: Underpowered, Biased, and Lacking Evidential Value","year":2022,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; University of Ottawa","funders":"","keywords":"Moderation; Psychology; Selection bias; Robustness (evolution); Publication bias; Meta-analysis; Statistics; Social psychology; Medicine; Mathematics; Internal medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.3099148603923441,"gpt":0.382829465661358,"spread":0.07291460526901389,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002402215,0.0004894516,0.00206788,0.0002551721,0.0006836542,0.00006344779,0.0007929711,0.0002453268,0.01156774],"category_scores_gemma":[0.000164446,0.0002841697,0.001784882,0.0009208495,0.0003716699,0.00008405006,0.000472313,0.001310481,0.00006055085],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003671083,"about_ca_system_score_gemma":0.00001643748,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008331549,"about_ca_topic_score_gemma":0.00001133075,"domain_scores_codex":[0.9936996,0.003131375,0.0009856711,0.0010358,0.0005196895,0.0006278249],"domain_scores_gemma":[0.9977656,0.0006325032,0.0005742101,0.0008204072,0.00009875677,0.0001085404],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"observational","study_design_scores_codex":[0.02122183,0.01231317,0.1450148,0.0002913201,0.5487448,0.001397498,0.03389212,0.0005583474,0.02690221,0.08916464,0.1023322,0.01816707],"study_design_scores_gemma":[0.02683629,0.005389849,0.4294201,0.00002385674,0.1957334,0.001176626,0.001314215,0.00007472163,0.0001259165,0.01822552,0.3196614,0.002018117],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9009727,0.06380782,0.0001961542,0.01430095,0.004847427,0.002043672,0.0001441145,0.0002588007,0.01342838],"genre_scores_gemma":[0.9879964,0.0001420162,0.0002125563,0.004071607,0.0001184555,0.0006041008,0.000009872935,0.0000430781,0.006801935],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3530114,"threshold_uncertainty_score":0.999961,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3198867954","doi":"10.15626/mp.2020.2506","title":"What to make of equivalence testing with a post-specified margin?","year":2021,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":13,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Margin (machine learning); Equivalence (formal languages); Frequentist inference; Post hoc; Statistical hypothesis testing; Computer science; Econometrics; Mathematics; Statistics; Machine learning; Artificial intelligence; Bayesian probability; Medicine; Bayesian inference","retraction":null,"screen_n_in":null,"score":{"opus":0.8258386333211938,"gpt":0.6156700248907838,"spread":0.21016860843041,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003040569,0.0002581723,0.001420514,0.00009937153,0.00004216823,0.00004431875,0.0004210755,0.0001825769,0.00222439],"category_scores_gemma":[0.08739571,0.0001902313,0.0002050503,0.0007113859,0.0002178189,0.00006997745,0.0001964437,0.0003223,0.0001008042],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001453027,"about_ca_system_score_gemma":0.00006747173,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003647338,"about_ca_topic_score_gemma":0.00001671734,"domain_scores_codex":[0.9959667,0.001346258,0.001095718,0.000740532,0.0004071979,0.0004435752],"domain_scores_gemma":[0.962277,0.03530516,0.0003559842,0.001203404,0.0006723671,0.0001860901],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.003851331,0.003917559,0.002380487,0.00119419,0.007436152,0.002372291,0.001574214,0.00001956944,0.09283659,0.3302986,0.01271499,0.5414041],"study_design_scores_gemma":[0.002188158,0.001571376,0.01187005,0.0002496763,0.002215721,0.0002587711,0.0002684219,0.00001415352,0.008924169,0.963473,0.008414783,0.0005516959],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09944557,0.001388737,0.8428853,0.02166892,0.004187721,0.001649632,0.0002247349,0.0002600442,0.02828941],"genre_scores_gemma":[0.01248875,0.0000508535,0.9814989,0.004756562,0.0001658217,0.00008772543,0.000001480232,0.00004990491,0.0009000191],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6331744,"threshold_uncertainty_score":0.9986877,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4387116974","doi":"10.15626/mp.2023.4020","title":"A Plea to Psychology Professional Societies that Publish Journals: Assess Computational Reproducibility","year":2023,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"scientometrics and bibliometrics research","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Victoria","funders":"","keywords":"Plea; Publication; Reproducibility; Psychology; Applied psychology; Political science; Law; Mathematics; Statistics","retraction":null,"screen_n_in":null,"score":{"opus":0.9415478893122667,"gpt":0.728999741796331,"spread":0.2125481475159358,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","bibliometrics","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","bibliometrics","insufficient_payload"],"category_scores_codex":[0.1193723,0.0003480006,0.001125819,0.04136764,0.0006105331,0.001987144,0.004528896,0.0004197921,0.009483469],"category_scores_gemma":[0.08018442,0.0002315939,0.0006321585,0.1477998,0.0005005564,0.0009260555,0.00153737,0.001000575,0.004256992],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007845007,"about_ca_system_score_gemma":0.0003328977,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001135638,"about_ca_topic_score_gemma":0.00001147424,"domain_scores_codex":[0.9793342,0.002743615,0.001768374,0.004085931,0.01056652,0.001501359],"domain_scores_gemma":[0.982884,0.007389652,0.0005935253,0.003422389,0.004931236,0.0007791533],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00006824321,0.0003366904,0.08249592,0.000004334824,0.0002512134,0.00003557392,0.0002051178,0.0001491609,0.0003121792,0.0004008647,0.8735426,0.0421981],"study_design_scores_gemma":[0.0005116293,0.000147787,0.687478,0.000002845296,0.00003824623,0.00005017834,0.000201985,0.0003502109,0.00004318396,0.09205174,0.2189028,0.0002213908],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6122456,0.001972009,0.04627454,0.3057218,0.01253341,0.001270137,0.0002940788,0.0003504773,0.01933804],"genre_scores_gemma":[0.9314603,0.0003247319,0.01451368,0.03094783,0.0006617408,0.0003560936,0.00007481842,0.0000619905,0.02159877],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6546398,"threshold_uncertainty_score":0.9990489,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4383908607","doi":"10.15626/mp.2021.2938","title":"A critical re-analysis of six implicit learning papers","year":2023,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Pairwise comparison; Contrast (vision); Statistics; Psychology; Mathematics; Computer science; Econometrics; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.1946829444336663,"gpt":0.5180829281699013,"spread":0.323399983736235,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.006625067,0.0002983583,0.001372006,0.001562325,0.0001736918,0.00001654758,0.0004329791,0.0003477466,0.01621494],"category_scores_gemma":[0.002976456,0.0002558595,0.0008209698,0.004549948,0.0003610576,0.00004358906,0.00008628904,0.001189086,0.0009506562],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000176285,"about_ca_system_score_gemma":0.00001804445,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001392104,"about_ca_topic_score_gemma":0.00001467908,"domain_scores_codex":[0.992562,0.004777729,0.0007869974,0.000852397,0.0002800937,0.0007408064],"domain_scores_gemma":[0.9961832,0.00235755,0.000268991,0.0009048732,0.0001848245,0.0001005354],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.001158719,0.001460942,0.1822246,0.0001165863,0.1559225,0.00064245,0.03469511,0.002518829,0.1418988,0.2065382,0.07105106,0.2017722],"study_design_scores_gemma":[0.0005982931,0.0004296021,0.7303863,0.00000405883,0.01087194,0.00001849719,0.0007940551,0.0001586707,0.00007343597,0.001148222,0.2551504,0.0003664994],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5112022,0.0006817307,0.02156178,0.008209346,0.002357896,0.0002551428,0.00003843309,0.0009512055,0.4547423],"genre_scores_gemma":[0.98584,0.00001168429,0.002756414,0.002369905,0.0001264431,0.00009033915,0.00006094445,0.00005748184,0.008686846],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5481617,"threshold_uncertainty_score":0.9999894,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3039001205","doi":"10.15626/mp.2019.1992","title":"Multiplicity Control vs Replication","year":2020,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Probability and Risk Models","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Multiplicity (mathematics); Replication (statistics); Multiplicity of infection; Computer science; Mathematics; Biology; Statistics; Genetics; In vitro; Geometry","retraction":null,"screen_n_in":null,"score":{"opus":0.5084974778146327,"gpt":0.4953623604817329,"spread":0.01313511733289979,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.003748457,0.0001578532,0.0006636151,0.00008356468,0.0001222161,0.00006004158,0.001073185,0.0001712529,0.001364636],"category_scores_gemma":[0.005690158,0.0001007714,0.0003421183,0.0005534536,0.0001829265,0.0001954541,0.00008985644,0.0002447056,0.002397184],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000009657638,"about_ca_system_score_gemma":0.0000246619,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001664045,"about_ca_topic_score_gemma":0.00001483453,"domain_scores_codex":[0.9964249,0.0007450539,0.000778068,0.00121408,0.0005666148,0.0002712701],"domain_scores_gemma":[0.996685,0.0008164716,0.0002661774,0.001786886,0.0002436652,0.0002017449],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003041605,0.00120699,0.04601913,0.00002085766,0.001929868,0.00004599281,0.00370982,0.002876735,0.03780145,0.0379956,0.3567019,0.50865],"study_design_scores_gemma":[0.002036,0.0003432388,0.0954628,7.634142e-7,0.0003549581,0.00001650885,0.00002328992,0.01551184,0.0006978235,0.1802301,0.7050272,0.0002954433],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1015299,0.0007747326,0.67997,0.2054771,0.0005163689,0.0007251553,0.00007194364,0.0002126698,0.01072214],"genre_scores_gemma":[0.950846,0.00002053354,0.002844457,0.04582332,0.0001315901,0.00006949522,0.00000326757,0.00000887135,0.0002524334],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8493161,"threshold_uncertainty_score":0.9995483,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4394955471","doi":"10.15626/mp.2021.2918","title":"Associations between Goal Orientation and Self-Regulated Learning Strategies are Stable across Course Types, Underrepresented Minority Status, and Gender","year":2024,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Innovative Teaching and Learning Methods","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Goal orientation; Psychology; Underrepresented Minority; Orientation (vector space); Future orientation; Social psychology; Medical education; Mathematics; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.1354320734427921,"gpt":0.4753749570623645,"spread":0.3399428836195724,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002988695,0.0002848503,0.0005822243,0.0001720313,0.0004283559,0.0002253193,0.0001094693,0.0003473815,0.0004042285],"category_scores_gemma":[0.0002062765,0.000258609,0.00008275441,0.0005877947,0.0002087955,0.0002321207,0.0000655633,0.001131483,0.00005855654],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005383307,"about_ca_system_score_gemma":0.00005896536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004582939,"about_ca_topic_score_gemma":0.00006252159,"domain_scores_codex":[0.9952369,0.002678794,0.0004384076,0.000824847,0.0001966248,0.000624385],"domain_scores_gemma":[0.9985014,0.000685197,0.0002359066,0.0003151806,0.000157001,0.0001053727],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007291733,0.0003003643,0.9189941,0.00008093171,0.008890096,0.00005601568,0.03090991,0.00008871355,0.0004786456,0.01482843,0.005310762,0.01998913],"study_design_scores_gemma":[0.0006484341,0.000122183,0.9652616,0.000005547126,0.001066965,0.0000187965,0.001978786,0.00009634889,0.000006973162,0.003257003,0.02729148,0.0002458268],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9671137,0.005570312,0.01479543,0.0007306794,0.0009779978,0.0002464875,0.0001456828,0.0005598537,0.009859835],"genre_scores_gemma":[0.9948059,0.0000512114,0.002119591,0.0001580299,0.000214308,0.00003081833,0.0002643232,0.00004334102,0.002312493],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04626758,"threshold_uncertainty_score":0.9999866,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4415699443","doi":"10.15626/mp.2020.2643","title":"A Multi-faceted Mess: A Review of Statistical Power Analysis in Psychology Journal Articles","year":2025,"lang":"","type":"article","venue":"Meta-Psychology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"York University","funders":"","keywords":"Sample size determination; Statistical power; Sample (material); Power (physics); Identification (biology); A priori and a posteriori; Order (exchange)","retraction":null,"screen_n_in":null,"score":{"opus":0.5836364185346339,"gpt":0.5957669098937192,"spread":0.01213049135908528,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","bibliometrics","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04914304,0.0008301772,0.006470586,0.008634497,0.0002528507,0.0001829496,0.003305682,0.0007446437,0.02497615],"category_scores_gemma":[0.1725473,0.0006042687,0.002092863,0.03374037,0.001344049,0.0002935338,0.0005277948,0.001892086,0.0001847166],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007631195,"about_ca_system_score_gemma":0.0003497608,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000615627,"about_ca_topic_score_gemma":0.00006799946,"domain_scores_codex":[0.969979,0.01649867,0.007722181,0.002513368,0.001655506,0.001631202],"domain_scores_gemma":[0.9601957,0.03181493,0.002923715,0.003115198,0.00149824,0.0004522214],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001187766,0.00475889,0.1420352,0.0009413207,0.02969597,0.0007385132,0.0006139278,0.0001007374,0.002211496,0.001870848,0.08914176,0.7267036],"study_design_scores_gemma":[0.005721979,0.001042615,0.8509552,0.00112498,0.01939992,0.0002852894,0.0003962886,0.001733708,0.00004832836,0.03823589,0.08018063,0.0008751069],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01241078,0.3186862,0.6416471,0.01557313,0.003615562,0.0007683611,0.00014808,0.00003831303,0.007112452],"genre_scores_gemma":[0.3586845,0.1443581,0.4541382,0.04150514,0.00009510526,0.0001412069,0.00001483041,0.0000613438,0.001001575],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7258285,"threshold_uncertainty_score":0.9996409,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4409214529","doi":"10.15626/mp.2025.4708","title":"Reality Television Shows Focusing on Sexual Relationships and College Students Engagement in One-Night Stands: A Replication Report","year":2025,"lang":"en","type":"preprint","venue":"Meta-Psychology","topic":"Asian Culture and Media Studies","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Replication (statistics); Psychology; Reality television; Sociology; Media studies; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.3444403399668227,"gpt":0.4876013957980693,"spread":0.1431610558312466,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.009454237,0.0002819882,0.0008275418,0.000280041,0.0006875523,0.0000592169,0.0003918957,0.000560816,0.00005777815],"category_scores_gemma":[0.002881278,0.0002472654,0.00011314,0.0003626421,0.0002498766,0.00007567108,0.0004545634,0.001184727,0.000006256942],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002579425,"about_ca_system_score_gemma":0.0002045899,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000440362,"about_ca_topic_score_gemma":0.009340819,"domain_scores_codex":[0.9943678,0.002439438,0.0008266006,0.001331806,0.0006911181,0.0003432748],"domain_scores_gemma":[0.9976504,0.0004632458,0.0004923513,0.001114169,0.0001779826,0.0001018136],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001158351,0.00426579,0.3254382,0.00067187,0.008096906,0.0008778832,0.3490379,0.00005829526,0.0002173615,0.04790333,0.1623488,0.09992529],"study_design_scores_gemma":[0.00116386,0.0001849964,0.4113051,0.0002110828,0.001690926,0.00000718808,0.01804368,0.000003188796,0.00003151522,0.01220362,0.5545492,0.0006056797],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.529823,0.007063459,0.004104645,0.0825609,0.004860833,0.009272752,0.0005518091,0.0004366034,0.361326],"genre_scores_gemma":[0.9833825,0.001736644,0.001847969,0.001203824,0.0004554281,0.0007867201,0.0001355643,0.00001877045,0.0104326],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4535595,"threshold_uncertainty_score":0.999998,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4398239542","doi":"10.15626/mp.2018.1479","title":"How Close to the Mark Might Published Heritability Estimates Be?","year":2024,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Heritability; Genealogy; Econometrics; History; Evolutionary biology; Biology; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.105602424643944,"gpt":0.3824294459025446,"spread":0.2768270212586006,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.002020786,0.0004037001,0.0006450255,0.0001966358,0.0002049078,0.0004896419,0.0006938156,0.0002250055,0.01461723],"category_scores_gemma":[0.001908885,0.0002539802,0.0004679396,0.0008185837,0.0003308153,0.0002385208,0.0002154104,0.0006247551,0.001017451],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003760276,"about_ca_system_score_gemma":0.00003512723,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001691325,"about_ca_topic_score_gemma":0.0002492752,"domain_scores_codex":[0.9964503,0.0006656881,0.0004797163,0.001345479,0.0002458708,0.0008129677],"domain_scores_gemma":[0.9959289,0.002104033,0.00007613881,0.001513803,0.0001882315,0.0001889142],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002586814,0.0005929667,0.009611666,0.0001200168,0.004094529,0.0003555515,0.004975018,0.000001439311,0.0008029853,0.0344201,0.723352,0.221415],"study_design_scores_gemma":[0.0003124803,0.0002541329,0.08932316,0.00001030562,0.0007447978,0.0001701119,0.0004945659,0.00002044305,0.00004056695,0.01773647,0.8905748,0.0003181989],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2838894,0.01765941,0.003963957,0.5056866,0.0109322,0.001777421,0.0003431909,0.001040025,0.1747077],"genre_scores_gemma":[0.9701037,0.0000129411,0.002092012,0.01475897,0.0007449023,0.0009320133,0.00003914463,0.0000839046,0.01123243],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6862142,"threshold_uncertainty_score":0.9999912,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4406183016","doi":"10.15626/mp.2021.2757","title":"The Devil is Mainly in the Nuisance Parameters: Performance of Structural Fit Indices Under Misspecified Structural Models in SEM","year":2024,"lang":"en","type":"article","venue":"Meta-Psychology","topic":"Multi-Criteria Decision Making","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Nuisance; Structural equation modeling; Econometrics; Nuisance parameter; Statistics; Mathematics; Economics; Political science; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.5075904091821208,"gpt":0.5103925816882979,"spread":0.002802172506177092,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005876569,0.0003483044,0.0007766191,0.0006238572,0.0001808545,0.0005403741,0.002876029,0.0002002807,0.0008528411],"category_scores_gemma":[0.0005607475,0.0001638961,0.0002999334,0.001959485,0.0004882704,0.0006704945,0.0002214236,0.0006736026,0.000106024],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004522091,"about_ca_system_score_gemma":0.00006188781,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004949495,"about_ca_topic_score_gemma":0.0003164854,"domain_scores_codex":[0.9939961,0.001241336,0.001650731,0.001010978,0.001489028,0.0006117947],"domain_scores_gemma":[0.993257,0.004688425,0.0002917951,0.001563485,0.000138552,0.00006071762],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002826408,0.0002018263,0.03875948,0.000144646,0.001709542,0.0005105162,0.05985721,0.04518171,0.005234692,0.04662339,0.0432307,0.7557199],"study_design_scores_gemma":[0.001352689,0.0002059183,0.4439391,0.00007889979,0.0001483081,0.0002368267,0.002623424,0.214578,0.0004528799,0.3156569,0.02014486,0.000582121],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9852225,0.005031411,0.0006887632,0.006162263,0.00137648,0.000396767,0.00003529678,0.0000236197,0.001062939],"genre_scores_gemma":[0.9935976,0.0001179919,0.002474198,0.003467675,0.00006616281,0.00004706988,0.000001908337,0.00002229832,0.0002050757],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7551377,"threshold_uncertainty_score":0.9338012,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}