{"id":"W6906660697","doi":"10.17605/osf.io/sdpem","title":"10-month-old infants' social evaluation of true versus negligent accidents: A direct replication attempt","year":2017,"lang":"en","type":"other","venue":"Open Science Framework","topic":"Child and Animal Learning Development","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Replication (statistics); Evaluation methods; Process (computing)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reproducibility","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":[],"domain":null,"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"],"consensus_categories":[],"category_scores_codex":[0.002266513,0.0004087263,0.0003706229,0.0009073214,0.0005719417,0.00129654,0.0006682666,0.0006701008,0.003620949],"category_scores_gemma":[0.01045296,0.0002129146,0.0003613512,0.0002819849,0.001362063,0.0008705657,0.001647073,0.001004917,0.0006401254],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005066267,"about_ca_system_score_gemma":0.0008048943,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0083009,"about_ca_topic_score_gemma":0.01131612,"domain_scores_codex":[0.9990593,0.0002332076,0.00005380358,0.0002003869,0.0003557816,0.00009753324],"domain_scores_gemma":[0.9946932,0.002123838,0.0008535078,0.001177706,0.0008160921,0.0003357193],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00186849,0.002389133,0.6273414,0.0006060072,0.0002491702,0.004371385,0.1008268,0.0004864012,0.09214415,0.01247446,0.004357821,0.1528847],"study_design_scores_gemma":[0.0000157574,0.0007337415,0.9651214,0.0001052422,0.0001258371,0.001039866,0.01609049,0.000360235,0.009592839,0.002749443,0.004013314,0.00005181415],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9897899,0.000163482,0.0005261293,0.00007123221,0.00002582073,0.00001911172,0.0001393113,0.00001550459,0.00924935],"genre_scores_gemma":[0.9943358,0.0002248682,0.0007968967,0.00006924207,0.000009783575,0.00006371339,0.0001724568,0.00002399436,0.004303094],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9977335,"threshold_uncertainty_score":0.01650518,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08657738528796921,"score_gpt":0.4427051590937182,"score_spread":0.356127773805749,"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."}}