{"id":"W4411937686","doi":"10.31234/osf.io/mvche_v2","title":"Flexible behavior or flexible methods? A cross-taxon review of experimental designs in reversal learning","year":2025,"lang":"en","type":"review","venue":"","topic":"Psychological and Educational Research Studies","field":"Psychology","cited_by":1,"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":"Taxon; Computer science; Psychology; Data science; Artificial intelligence; Paleontology; Biology","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":[0.1442987,0.001837045,0.005950745,0.007226593,0.0009761606,0.003459694,0.003629606,0.002416219,0.003938225],"category_scores_gemma":[0.2831457,0.00183153,0.005738331,0.007891533,0.005404585,0.004020465,0.003298283,0.002786528,0.0009041818],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004052659,"about_ca_system_score_gemma":0.006836276,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003001323,"about_ca_topic_score_gemma":0.004909951,"domain_scores_codex":[0.877369,0.07957084,0.02159109,0.01033569,0.01056177,0.000571634],"domain_scores_gemma":[0.5364022,0.4040774,0.02559487,0.017616,0.0156855,0.0006240876],"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":[0.00110248,0.0001317731,0.002292418,0.1445258,0.008093266,0.00008658767,0.001042514,0.001264308,0.001438879,0.02239671,0.004147092,0.8134782],"study_design_scores_gemma":[0.001110613,0.004083273,0.02931108,0.4548579,0.02887438,0.0009858472,0.001227002,0.003518447,0.00618069,0.08503287,0.3841563,0.0006615438],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001819409,0.9685175,0.02588535,0.001220888,0.0006462224,0.0004086127,0.0001749468,0.0000783959,0.001248786],"genre_scores_gemma":[0.03912659,0.8497827,0.1004083,0.004252797,0.001033208,0.003889909,0.0004399839,0.0002917105,0.0007748261],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8557013,"threshold_uncertainty_score":0.7631336,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4671961358542834,"score_gpt":0.6571473535270884,"score_spread":0.189951217672805,"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."}}