{"id":"W7111990554","doi":"","title":"Replication Games Report on Altmann et al. 2022","year":2023,"lang":"","type":"other","venue":"OSF Preprints (OSF Preprints)","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reproducibility","study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reproducibility","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005287936,0.001641943,0.001211075,0.002900469,0.003349064,0.005348299,0.003295944,0.002412906,0.3641798],"category_scores_gemma":[0.03787567,0.000794121,0.001041104,0.002926044,0.001271685,0.00620629,0.003814177,0.003656015,0.1514599],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006255714,"about_ca_system_score_gemma":0.0087531,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.1298693,"about_ca_topic_score_gemma":0.1522827,"domain_scores_codex":[0.9959381,0.001030146,0.0002454728,0.0004996303,0.00171039,0.0005763043],"domain_scores_gemma":[0.9803573,0.005870296,0.0005949028,0.005249624,0.006275935,0.001652015],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0002871522,0.0001846889,0.0004551616,0.0001184196,0.0000239537,0.00003201953,0.0002370531,0.0009485685,0.0003454112,0.02735782,0.9419614,0.0280484],"study_design_scores_gemma":[0.0002389147,0.00007051235,0.002657086,0.0001953395,0.00004112384,0.00003777614,0.0004865222,0.002533239,0.001920205,0.0334143,0.9582888,0.0001162348],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.01216937,0.002222835,0.04739925,0.02500681,0.008784197,0.002836965,0.2036165,0.01070296,0.687261],"genre_scores_gemma":[0.09452353,0.001801109,0.04943867,0.005297611,0.0009713954,0.004326881,0.1218608,0.008190129,0.7135898],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9947121,"threshold_uncertainty_score":0.9069203,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02346462226025473,"score_gpt":0.3095317678688447,"score_spread":0.28606714560859,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}