{"id":"W4411793100","doi":"10.31234/osf.io/mvche_v1","title":"Flexible Behavior or Flexible Methods? A Cross-Taxon Review of Experimental Designs in Reversal Learning","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Behavioral and Psychological Studies","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada; Austrian Science Fund; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; Université Catholique de Louvain; European Commission; Leverhulme Trust; Open Philanthropy Project; National Science Foundation","keywords":"Computer science","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1494393,0.001863478,0.005998712,0.007262056,0.000995414,0.003492731,0.003681577,0.002417556,0.003951926],"category_scores_gemma":[0.2903621,0.001862688,0.005845578,0.00789777,0.005488212,0.003983388,0.003362807,0.002786222,0.0009013733],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004077605,"about_ca_system_score_gemma":0.006846277,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002988458,"about_ca_topic_score_gemma":0.004884866,"domain_scores_codex":[0.8715631,0.08378768,0.02247543,0.01070422,0.01088205,0.0005875718],"domain_scores_gemma":[0.5259298,0.4124261,0.02644442,0.01853781,0.0160317,0.0006302613],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.001148531,0.0001366143,0.002376913,0.1447621,0.008501826,0.00008696031,0.001071972,0.001299595,0.001473699,0.0226458,0.004147381,0.8123486],"study_design_scores_gemma":[0.001170521,0.004279186,0.03058856,0.453657,0.03017607,0.0009946369,0.001255552,0.003666395,0.006434748,0.08697882,0.3801098,0.0006886771],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00195406,0.9660171,0.02805895,0.001266934,0.0006747182,0.0004522184,0.0001890616,0.00008370807,0.001303115],"genre_scores_gemma":[0.04177918,0.838876,0.1081259,0.004291816,0.001054783,0.004302226,0.0004687024,0.0003056719,0.0007958425],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8505607,"threshold_uncertainty_score":0.7903198,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5077044918703139,"score_gpt":0.5550947476409621,"score_spread":0.04739025577064826,"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."}}