{"id":"W4412619104","doi":"10.1162/opmi.a.2","title":"When Success Is Surprising: Children’s Ability to Use Surprise to Infer Competence","year":2025,"lang":"en","type":"article","venue":"Open Mind","topic":"Child and Animal Learning Development","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"The Scarborough Hospital; University of Toronto","funders":"James S. McDonnell Foundation","keywords":"Surprise; Psychology; Competence (human resources); Social psychology; Developmental psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0005431318,0.0002048942,0.0003171631,0.0001363356,0.0001674625,0.0004325613,0.0008554103,0.0000927197,0.0114439],"category_scores_gemma":[0.0002006098,0.0001960911,0.00005916421,0.0003580909,0.00004294915,0.0001690336,0.0007344945,0.0002212624,0.004123393],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008609483,"about_ca_system_score_gemma":0.0001323176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003067476,"about_ca_topic_score_gemma":0.0007344929,"domain_scores_codex":[0.9981415,0.0001338935,0.0003541288,0.0008163634,0.0001702147,0.0003838982],"domain_scores_gemma":[0.9987785,0.0001676331,0.00005643673,0.0006820097,0.000105019,0.0002104036],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000199256,0.00022753,0.9036056,0.000004032966,0.00006661696,0.00001318856,0.008135668,0.000009399734,0.0003175175,0.0003170758,0.03104645,0.05605766],"study_design_scores_gemma":[0.0003011989,0.00005080058,0.7467495,0.00006285115,0.00001170976,0.000002408332,0.0000811187,8.183655e-7,0.0004722764,0.00003559168,0.2520376,0.0001941131],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9521022,0.00002388675,0.0002177328,0.004329437,0.0003480174,0.0008481171,0.00004370939,0.000008436118,0.0420785],"genre_scores_gemma":[0.9659008,0.000001130872,0.007601718,0.002787739,0.00003630869,0.00003294343,0.00001681034,0.00001492309,0.02360757],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2209911,"threshold_uncertainty_score":0.996652,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04176386054266638,"score_gpt":0.3557976766846702,"score_spread":0.3140338161420038,"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."}}