{"meta":{"query_hash":"59d05407f7d7","filters":{"venue":"Meta-Psychology"},"cohort_total":15,"direct_labels_cover":0,"predictions_cover":15,"exported":15,"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/59d05407f7d7","api":"https://metacan.xera.ac/api/v1/cohort?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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.8534897186781798,"score_gpt":0.6528829468946589,"score_spread":0.20060677178352093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2915455637","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0017593097,0.013958311,0.72610825,0.012036649,0.008997864,0.15930155,0.019328373,0.012040484,0.04646926],"genre_scores_gemma":[0.0022003183,0.0034039014,0.86784106,0.0023080918,0.00072196487,0.11289403,0.00230658,0.0012792038,0.007044866],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.66410184,0.24873155,0.05471181,0.005708472,0.02503463,0.0017116043],"domain_scores_gemma":[0.32951292,0.45094952,0.030299714,0.055678055,0.12949972,0.004060103],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25644526,0.00543806,0.004464301,0.014060706,0.0038345652,0.0119990315,0.007896178,0.00924505,0.11316248],"category_scores_gemma":[0.5819283,0.0059131794,0.006058019,0.011929536,0.004356264,0.0076638362,0.0047350656,0.01198602,0.063146606],"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":[0.0012415589,0.00053984375,0.0011742509,0.028894676,0.0006069713,0.0005159992,0.0049641877,0.0031986288,0.003176921,0.066086955,0.41821846,0.4713816],"study_design_scores_gemma":[0.0015640834,0.0009176944,0.002091548,0.02456959,0.000375884,0.0004611259,0.0013903585,0.0047262814,0.0029377483,0.087754756,0.8727494,0.0004615681],"about_ca_topic_score_codex":0.004048339,"about_ca_topic_score_gemma":0.0068232734,"teacher_disagreement_score":0.7435547,"about_ca_system_score_codex":0.005528864,"about_ca_system_score_gemma":0.028374786,"threshold_uncertainty_score":0.9169354},"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,"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 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","score_opus":0.8026212016523445,"score_gpt":0.5861197031220517,"score_spread":0.21650149853029277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3030972095","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020924907,0.010034241,0.95155716,0.0031499295,0.00078727625,0.006067709,0.0016554879,0.00077790464,0.00504536],"genre_scores_gemma":[0.5166556,0.003479107,0.43778455,0.0028237181,0.0004361479,0.036305465,0.0013702483,0.0004390094,0.0007060615],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.5896006,0.32694477,0.028463582,0.027393293,0.025735108,0.0018626264],"domain_scores_gemma":[0.27841642,0.64753854,0.020065982,0.042468198,0.010925696,0.00058521284],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36960584,0.002039442,0.0076275854,0.008719582,0.0017589972,0.006169249,0.0043433285,0.0070008775,0.0035705515],"category_scores_gemma":[0.7589355,0.0016004925,0.013205133,0.008644607,0.005236844,0.009234727,0.0049927314,0.005011002,0.00062061986],"study_design_candidate":"theoretical_or_conceptual","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":[0.004245606,0.00045354798,0.10734858,0.03271939,0.086835116,0.0023995454,0.0073478236,0.0634827,0.0039347047,0.22878577,0.021076497,0.44137076],"study_design_scores_gemma":[0.0026086785,0.0024700544,0.042323682,0.008073075,0.027372364,0.0019455375,0.0013846747,0.10432273,0.008226349,0.7653504,0.03525709,0.0006653801],"about_ca_topic_score_codex":0.0020391275,"about_ca_topic_score_gemma":0.0011782146,"teacher_disagreement_score":0.63039416,"about_ca_system_score_codex":0.0028887487,"about_ca_system_score_gemma":0.0046383417,"threshold_uncertainty_score":0.7773882},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.5084974778146327,"score_gpt":0.49536236048173293,"score_spread":0.013135117332899793,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3039001205","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007728989,0.012286591,0.9338692,0.02082651,0.0044907206,0.001957073,0.0002261042,0.00050467585,0.018110162],"genre_scores_gemma":[0.39092636,0.006991849,0.56785524,0.01271328,0.006518814,0.009946158,0.00015717592,0.00054863846,0.004342441],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.356358,0.5560315,0.016728226,0.0276935,0.041231707,0.0019570393],"domain_scores_gemma":[0.22087081,0.63172233,0.028273256,0.101076946,0.016575962,0.0014806838],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.39094025,0.0030416385,0.0050988244,0.006142861,0.0046174643,0.008915246,0.007666187,0.009679404,0.008289803],"category_scores_gemma":[0.67860717,0.0019456112,0.0047033695,0.0064947647,0.03535598,0.02020174,0.009596033,0.013199423,0.0015348464],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047366004,0.0000705638,0.0021023038,0.002309886,0.0011027056,0.00022851421,0.0032307326,0.002143707,0.00048222937,0.90761745,0.004122815,0.07611547],"study_design_scores_gemma":[0.00032767656,0.000467404,0.000981698,0.0014145324,0.00045178804,0.0005400987,0.00055913173,0.0061502117,0.0013773134,0.9621881,0.025380056,0.00016194988],"about_ca_topic_score_codex":0.0014061888,"about_ca_topic_score_gemma":0.00073589745,"teacher_disagreement_score":0.60905975,"about_ca_system_score_codex":0.0048592063,"about_ca_system_score_gemma":0.0064860657,"threshold_uncertainty_score":0.7510791},"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,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_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","score_opus":0.3490952054780505,"score_gpt":0.5715458269845495,"score_spread":0.22245062150649897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3135512811","genre_codex":"commentary","genre_gemma":"review","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011873734,0.10941585,0.08535903,0.74554825,0.0075988676,0.00020316955,0.00023123034,0.00022453586,0.039545376],"genre_scores_gemma":[0.5526962,0.051523704,0.10770513,0.2532234,0.025327783,0.0020125515,0.00034644594,0.00079422083,0.006370576],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.76238227,0.15863034,0.011990512,0.0217745,0.041876234,0.0033461328],"domain_scores_gemma":[0.39427492,0.52912116,0.0071480554,0.032535933,0.032021616,0.004898328],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.33139,0.0014885752,0.0036111786,0.008090427,0.0069271186,0.013719106,0.004174522,0.0133046815,0.0032648877],"category_scores_gemma":[0.39921442,0.0013724988,0.002501436,0.00434892,0.082320996,0.034652624,0.011869686,0.027428135,0.0010668265],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010963342,0.00009010471,0.0024626902,0.0006860772,0.00012256144,0.000110864086,0.006038695,0.0004385504,0.000148662,0.88049185,0.025236262,0.084064156],"study_design_scores_gemma":[0.000089368936,0.00009204319,0.0019221493,0.002377305,0.00002978287,0.00021025642,0.0017225222,0.0013257766,0.00022659289,0.9297298,0.062208742,0.00006575748],"about_ca_topic_score_codex":0.0053236084,"about_ca_topic_score_gemma":0.0019301992,"teacher_disagreement_score":0.66861,"about_ca_system_score_codex":0.012295046,"about_ca_system_score_gemma":0.014421723,"threshold_uncertainty_score":0.82451516},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.8258386333211938,"score_gpt":0.6156700248907838,"score_spread":0.21016860843040996,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3198867954","genre_codex":"commentary","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0041133105,0.108983375,0.397398,0.44567809,0.031008974,0.0027849087,0.000607067,0.0005545775,0.008871814],"genre_scores_gemma":[0.1408626,0.03154605,0.6381667,0.14218117,0.03314096,0.011902601,0.00034666172,0.00046107554,0.0013921637],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.40682593,0.49462363,0.05199464,0.013546406,0.031726,0.0012833892],"domain_scores_gemma":[0.19151035,0.7323112,0.02780389,0.029078323,0.01782705,0.0014691825],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5300969,0.002364435,0.010444736,0.0058139535,0.0035100603,0.014747245,0.012559695,0.016888916,0.0037944126],"category_scores_gemma":[0.82923084,0.002035492,0.0069722096,0.006589803,0.03624928,0.030280838,0.0057876497,0.022474103,0.0017491682],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030356224,0.00021928678,0.0031996276,0.038481742,0.0060826764,0.00083623285,0.0065178536,0.0035190873,0.0010103818,0.54117066,0.061589774,0.3343372],"study_design_scores_gemma":[0.0011132323,0.000586189,0.001340741,0.019918324,0.0014074247,0.00058643956,0.0014247912,0.0045582056,0.0012383441,0.9129342,0.05465475,0.00023739446],"about_ca_topic_score_codex":0.0016963101,"about_ca_topic_score_gemma":0.0015487956,"teacher_disagreement_score":0.4699031,"about_ca_system_score_codex":0.0065379115,"about_ca_system_score_gemma":0.009961091,"threshold_uncertainty_score":0.5794742},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.5074782751927323,"score_gpt":0.4361204782108346,"score_spread":0.07135779698189765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4226236655","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00092636555,0.007833837,0.9370284,0.003797256,0.0015789816,0.00043212634,0.010478298,0.021702372,0.016222436],"genre_scores_gemma":[0.013280401,0.009465318,0.928889,0.0024618092,0.0013147793,0.0029654915,0.010713833,0.014112742,0.01679668],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9965113,0.002150799,0.00027235865,0.00038924103,0.00059722795,0.00007907341],"domain_scores_gemma":[0.97023296,0.025569044,0.0006361856,0.0016368994,0.001667078,0.0002578154],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009693327,0.0024228585,0.0012906276,0.002645291,0.0005180953,0.0016057128,0.0019032385,0.0012512971,0.09320891],"category_scores_gemma":[0.04308141,0.0016000851,0.0025910498,0.0031413245,0.0006509961,0.0042767767,0.0019638173,0.004092133,0.037535045],"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":[0.00008193079,0.00014305413,0.0018078246,0.0020628213,0.00023582559,0.0003501843,0.0007457031,0.005573826,0.00092603435,0.10099195,0.5708799,0.31620112],"study_design_scores_gemma":[0.00009866241,0.000091996015,0.0031453804,0.0012482187,0.000088443616,0.0005601127,0.00014050907,0.020712662,0.0013947075,0.22064076,0.75170887,0.00016958574],"about_ca_topic_score_codex":0.0037942503,"about_ca_topic_score_gemma":0.0054735043,"teacher_disagreement_score":0.09320891,"about_ca_system_score_codex":0.00087723584,"about_ca_system_score_gemma":0.0017746346,"threshold_uncertainty_score":0.31181484},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.7276317683307449,"score_gpt":0.64387429511749,"score_spread":0.083757473213255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4294233607","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0047774473,0.0029278097,0.9509302,0.0016315574,0.0011339323,0.008203363,0.011417441,0.014141777,0.0048364964],"genre_scores_gemma":[0.042684298,0.0012073066,0.90139836,0.00080495625,0.0003008787,0.038496956,0.0047522685,0.008236422,0.0021185223],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.680928,0.2535161,0.022200124,0.02070647,0.020715142,0.0019341075],"domain_scores_gemma":[0.3353841,0.5348835,0.028330637,0.0709689,0.028852554,0.0015803466],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34216496,0.005295318,0.0067011397,0.014175891,0.0023630217,0.008945033,0.0077935453,0.00563793,0.03242816],"category_scores_gemma":[0.71770686,0.00570673,0.015248408,0.016808713,0.0051807743,0.007370513,0.009409021,0.00908967,0.010996387],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00703854,0.00043506135,0.054363906,0.025642676,0.032733425,0.0011059226,0.007051914,0.040634368,0.004889577,0.14043072,0.14200294,0.543671],"study_design_scores_gemma":[0.006353006,0.0018341432,0.037675265,0.0071693165,0.016116112,0.0020594765,0.00090459344,0.17765033,0.015879357,0.36402258,0.36882767,0.0015081151],"about_ca_topic_score_codex":0.004232168,"about_ca_topic_score_gemma":0.0034771566,"teacher_disagreement_score":0.657835,"about_ca_system_score_codex":0.0030227352,"about_ca_system_score_gemma":0.010620141,"threshold_uncertainty_score":0.81122774},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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","score_opus":0.3099148603923441,"score_gpt":0.382829465661358,"score_spread":0.07291460526901389,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4308725914","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009137166,0.9554598,0.024140691,0.0037224428,0.0015017025,0.0019188924,0.0011588039,0.00022397067,0.0027365645],"genre_scores_gemma":[0.6088126,0.31747383,0.049326938,0.012444796,0.0022862845,0.0061301016,0.001995005,0.0005605539,0.0009699997],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.68888867,0.18662144,0.07461776,0.020516323,0.027365945,0.0019897402],"domain_scores_gemma":[0.24425514,0.68750215,0.032387752,0.021849638,0.012845055,0.0011602405],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.31608957,0.0039006115,0.014961674,0.015638627,0.0014159051,0.011767485,0.0054077823,0.0042724293,0.005114566],"category_scores_gemma":[0.5962925,0.0030175804,0.040650107,0.01299676,0.005592751,0.00915664,0.0051762788,0.004925286,0.000791073],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018074192,0.000049043065,0.014943181,0.25407934,0.67876655,0.00042993005,0.0008270251,0.0012125751,0.0004067409,0.002507015,0.0017292417,0.043241948],"study_design_scores_gemma":[0.0009846847,0.00049628684,0.010044895,0.08922376,0.87353426,0.00060344825,0.00046917307,0.00093940547,0.0006597514,0.010973172,0.011964719,0.00010640823],"about_ca_topic_score_codex":0.0035712367,"about_ca_topic_score_gemma":0.0083115995,"teacher_disagreement_score":0.98503834,"about_ca_system_score_codex":0.0047021443,"about_ca_system_score_gemma":0.0061381795,"threshold_uncertainty_score":0.8433833},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Pairwise comparison; Contrast (vision); Statistics; Psychology; Mathematics; Computer science; Econometrics; Artificial intelligence","score_opus":0.19468294443366635,"score_gpt":0.5180829281699013,"score_spread":0.32339998373623496,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4383908607","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12320811,0.39150316,0.07832327,0.15214297,0.13064522,0.046874847,0.046213474,0.003417409,0.02767162],"genre_scores_gemma":[0.49952397,0.094291694,0.1683008,0.08112361,0.03464063,0.06436048,0.029911628,0.004630283,0.023216825],"study_design_codex":"systematic_review","study_design_gemma":"observational","domain_scores_codex":[0.6966436,0.0695649,0.1150392,0.019450871,0.09457715,0.004724315],"domain_scores_gemma":[0.117232054,0.5189799,0.07711533,0.045991797,0.23835775,0.002323176],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20379171,0.0035895603,0.006080847,0.06882834,0.004155721,0.010468616,0.00564732,0.0062216255,0.008857053],"category_scores_gemma":[0.68980837,0.0025601608,0.008273343,0.028013928,0.004497114,0.008747269,0.006508126,0.005853317,0.0025839242],"study_design_candidate":"observational","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":[0.007998708,0.0005705814,0.015474293,0.36713862,0.028943148,0.008036971,0.029541278,0.0013507034,0.014138706,0.015818924,0.18849085,0.32249716],"study_design_scores_gemma":[0.0022582826,0.0011994884,0.026875166,0.17969942,0.035159353,0.0037051518,0.009026909,0.0013390861,0.024094144,0.0249449,0.6904707,0.0012273253],"about_ca_topic_score_codex":0.0019916298,"about_ca_topic_score_gemma":0.0038528866,"teacher_disagreement_score":0.79620826,"about_ca_system_score_codex":0.010943096,"about_ca_system_score_gemma":0.014258143,"threshold_uncertainty_score":0.9818666},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Plea; Publication; Reproducibility; Psychology; Applied psychology; Political science; Law; Mathematics; Statistics","score_opus":0.9415478893122667,"score_gpt":0.728999741796331,"score_spread":0.21254814751593576,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4387116974","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16950063,0.011853951,0.30242503,0.40428355,0.011116609,0.005280228,0.0046818834,0.0045113065,0.08634672],"genre_scores_gemma":[0.6713202,0.0017863726,0.27316284,0.028763667,0.004095686,0.0086593665,0.002677768,0.0019311648,0.0076029627],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.5630577,0.33005822,0.029419817,0.02065004,0.054247007,0.0025671422],"domain_scores_gemma":[0.07288624,0.6141971,0.04933104,0.19149719,0.06354548,0.008542958],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.356274,0.0009302802,0.0021699641,0.019737074,0.0068056304,0.01593093,0.004371066,0.007262245,0.014644451],"category_scores_gemma":[0.87048244,0.0018282988,0.0042012124,0.027060464,0.011156007,0.019235782,0.01233319,0.0062864637,0.0072284676],"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":[0.0019658024,0.0009019883,0.34546542,0.0020257882,0.003931458,0.00023790973,0.010100324,0.0021351844,0.0018126192,0.05235286,0.163737,0.41533354],"study_design_scores_gemma":[0.0014891045,0.0018717266,0.24560155,0.0031106987,0.0026030699,0.001470241,0.009667027,0.028212842,0.004261259,0.5341464,0.16677444,0.0007915418],"about_ca_topic_score_codex":0.0017608259,"about_ca_topic_score_gemma":0.0022650536,"teacher_disagreement_score":0.643726,"about_ca_system_score_codex":0.0035043494,"about_ca_system_score_gemma":0.006893505,"threshold_uncertainty_score":0.7938287},"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,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Goal orientation; Psychology; Underrepresented Minority; Orientation (vector space); Future orientation; Social psychology; Medical education; Mathematics; Medicine","score_opus":0.13543207344279207,"score_gpt":0.4753749570623645,"score_spread":0.3399428836195724,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4394955471","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9959436,0.0007470958,0.0010621087,0.000119603625,0.000025166555,0.00008726628,0.00023501553,0.000014641619,0.0017655396],"genre_scores_gemma":[0.99846786,0.000114155235,0.00083260966,0.00003731236,0.0000049483688,0.00006180688,0.00017574128,0.000008367349,0.00029709435],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.997738,0.0008594261,0.00020982313,0.0005514518,0.00047055716,0.00017076713],"domain_scores_gemma":[0.9751221,0.009494511,0.003888278,0.00714182,0.0035804443,0.000772872],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009748483,0.00036748094,0.00067179237,0.0013496599,0.0008972137,0.0017071904,0.00087832985,0.0003953616,0.0023891407],"category_scores_gemma":[0.025200993,0.00025711083,0.0013878247,0.0012507131,0.00096161634,0.0009719478,0.00097387814,0.0006699035,0.00022066462],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006628973,0.0003277456,0.9571666,0.0002946189,0.0013366217,0.00007011874,0.006151375,0.00013300197,0.0026273099,0.00045921034,0.00032633395,0.030444037],"study_design_scores_gemma":[0.00003070909,0.00043039076,0.9941685,0.00007924447,0.0005895476,0.00004518126,0.0021490506,0.0002136084,0.0009266862,0.00047452326,0.0008779778,0.000014593808],"about_ca_topic_score_codex":0.022289464,"about_ca_topic_score_gemma":0.031774897,"teacher_disagreement_score":0.022289464,"about_ca_system_score_codex":0.0008981498,"about_ca_system_score_gemma":0.0016779613,"threshold_uncertainty_score":0.051555574},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Heritability; Genealogy; Econometrics; History; Evolutionary biology; Biology; Mathematics","score_opus":0.10560242464394395,"score_gpt":0.3824294459025446,"score_spread":0.2768270212586006,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398239542","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015615948,0.20610555,0.15260398,0.5233346,0.07499492,0.00014474335,0.0017620808,0.0016180546,0.023820102],"genre_scores_gemma":[0.43785408,0.09344448,0.21031058,0.17186558,0.068774275,0.00085636217,0.0020940176,0.0023357647,0.012464869],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.93268144,0.039540816,0.007563326,0.01083164,0.008338926,0.0010438868],"domain_scores_gemma":[0.7458558,0.15428217,0.016557643,0.029357828,0.051388267,0.0025582833],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09953856,0.0018697672,0.0040410934,0.007731473,0.0030816244,0.015968058,0.0060439687,0.009649697,0.0049103424],"category_scores_gemma":[0.42445597,0.00121768,0.0018316661,0.0044029,0.016198602,0.031110886,0.00589773,0.017000897,0.0049985405],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00092589617,0.00017602154,0.012152743,0.0037719703,0.0019695642,0.00081648666,0.005515471,0.0049136523,0.000895397,0.4339413,0.173614,0.3613074],"study_design_scores_gemma":[0.0001524464,0.00020871659,0.008821437,0.007301426,0.00055171084,0.0008378758,0.0036202015,0.0037535913,0.001736301,0.7277437,0.24484785,0.00042474436],"about_ca_topic_score_codex":0.007169322,"about_ca_topic_score_gemma":0.0025586977,"teacher_disagreement_score":0.90046144,"about_ca_system_score_codex":0.004564101,"about_ca_system_score_gemma":0.0028189418,"threshold_uncertainty_score":0.52641654},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Nuisance; Structural equation modeling; Econometrics; Nuisance parameter; Statistics; Mathematics; Economics; Political science; Law","score_opus":0.5075904091821208,"score_gpt":0.5103925816882979,"score_spread":0.002802172506177092,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406183016","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2797802,0.0032063068,0.7004101,0.0023166682,0.0006644072,0.0005316611,0.000761283,0.0014773401,0.010852012],"genre_scores_gemma":[0.86639446,0.0006515077,0.12970304,0.00049181434,0.00013367503,0.00055913336,0.0006618886,0.00044076322,0.00096368033],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.86834747,0.089114115,0.01061768,0.015027175,0.016015837,0.00087779085],"domain_scores_gemma":[0.39256546,0.5434437,0.012693148,0.04046114,0.009818693,0.001017854],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.14937305,0.0022446432,0.0036852448,0.005248744,0.0013974758,0.005211217,0.0032250246,0.0033562896,0.0020941633],"category_scores_gemma":[0.49594125,0.0013981103,0.0028476908,0.0047106207,0.007449504,0.007509721,0.0049092593,0.005543773,0.00079293305],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016850879,0.00047202414,0.37966663,0.001977812,0.006781644,0.0016209922,0.007012066,0.14045559,0.0036769498,0.17214999,0.0062893378,0.27821198],"study_design_scores_gemma":[0.0003718493,0.0026036634,0.14433445,0.0014900976,0.0014555114,0.0030020727,0.0023218119,0.45432872,0.01267621,0.36357793,0.013094315,0.0007433614],"about_ca_topic_score_codex":0.0017278526,"about_ca_topic_score_gemma":0.002406391,"teacher_disagreement_score":0.85062695,"about_ca_system_score_codex":0.0015875379,"about_ca_system_score_gemma":0.002442894,"threshold_uncertainty_score":0.7899697},"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,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Replication (statistics); Psychology; Reality television; Sociology; Media studies; Medicine","score_opus":0.34444033996682266,"score_gpt":0.48760139579806927,"score_spread":0.1431610558312466,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409214529","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99241924,0.0026900163,0.0010259133,0.00017330708,0.00008586411,0.00046979703,0.0006921621,0.000019776015,0.0024239856],"genre_scores_gemma":[0.9967313,0.00067897077,0.0010203879,0.00010504832,0.000031971696,0.00046102927,0.00039013088,0.000014980231,0.0005662081],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99467105,0.0030959449,0.00063844456,0.0007562743,0.00057391415,0.0002644464],"domain_scores_gemma":[0.9696228,0.0119595695,0.0056176246,0.009291624,0.002994932,0.0005135784],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.012422679,0.00057528465,0.00065714173,0.0011329639,0.0012208575,0.0016268912,0.0011113378,0.00064336206,0.0046783704],"category_scores_gemma":[0.031667896,0.00047232423,0.003190103,0.0017162764,0.001097089,0.0012193983,0.001446706,0.0011933412,0.0003228598],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012537234,0.0012196383,0.9020842,0.003728532,0.00739791,0.00026390565,0.025185358,0.00016491752,0.0019439976,0.0011779618,0.0014458941,0.05413401],"study_design_scores_gemma":[0.00023866663,0.0018166604,0.963442,0.0010309612,0.006866572,0.00025310743,0.017782649,0.0002948444,0.0017448398,0.0008156076,0.0056423633,0.00007159227],"about_ca_topic_score_codex":0.018985642,"about_ca_topic_score_gemma":0.02017088,"teacher_disagreement_score":0.9875773,"about_ca_system_score_codex":0.0008376924,"about_ca_system_score_gemma":0.002055395,"threshold_uncertainty_score":0.06569821},"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,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_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)","score_opus":0.5836364185346339,"score_gpt":0.5957669098937192,"score_spread":0.012130491359085283,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415699443","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00044234068,0.9765347,0.012001572,0.0072287726,0.0013216538,0.00051628874,0.00017179051,0.00008165261,0.0017012187],"genre_scores_gemma":[0.013552843,0.951958,0.02476717,0.0052960846,0.0016340733,0.0019690339,0.00021185062,0.00018609266,0.0004248605],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.8369749,0.081109405,0.04307577,0.006094943,0.031767804,0.0009772063],"domain_scores_gemma":[0.31990162,0.6137779,0.026103511,0.008556624,0.03048325,0.0011770386],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.17082271,0.0018919782,0.0065615047,0.035423957,0.0023690003,0.009688934,0.0045020026,0.0050979825,0.00576815],"category_scores_gemma":[0.46479124,0.0027720265,0.007937127,0.031415887,0.0071525993,0.011987365,0.0048297103,0.005630169,0.0015708467],"study_design_candidate":"systematic_review","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":[0.00016565209,0.00006934371,0.0015327088,0.22278215,0.002909984,0.00017493145,0.0021529119,0.00044597077,0.000272377,0.016221998,0.02141628,0.73185575],"study_design_scores_gemma":[0.00013673588,0.00035929002,0.004998758,0.60862625,0.008706057,0.0013755234,0.0020899998,0.0009438581,0.000776096,0.043695245,0.32803273,0.00025944505],"about_ca_topic_score_codex":0.0027169953,"about_ca_topic_score_gemma":0.0051903236,"teacher_disagreement_score":0.82917726,"about_ca_system_score_codex":0.007082623,"about_ca_system_score_gemma":0.018509163,"threshold_uncertainty_score":0.90340763},"labels":[],"label_agreement":null}]}