{"id":"W4407427851","doi":"10.2139/ssrn.5135862","title":"A Protocol for Structured Robustness Reproductions and Replicability Assessments","year":2025,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Software Reliability and Analysis Research","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Ottawa","funders":"Deutsche Forschungsgemeinschaft","keywords":"Robustness (evolution); Protocol (science); Computer science; Medicine; Biology; Pathology","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05307489,0.003906254,0.003362142,0.00578861,0.004821858,0.003258591,0.003593743,0.006847461,0.1434122],"category_scores_gemma":[0.1679592,0.003735035,0.002933624,0.002755618,0.003091007,0.002246884,0.005276606,0.008002949,0.08167206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002569057,"about_ca_system_score_gemma":0.01079526,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001022933,"about_ca_topic_score_gemma":0.002348365,"domain_scores_codex":[0.943714,0.03314158,0.01042951,0.003683191,0.007100906,0.001930881],"domain_scores_gemma":[0.7699926,0.09085105,0.006717111,0.07642836,0.05282775,0.003183045],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.01818373,0.009008914,0.002484825,0.011039,0.0004941258,0.003065346,0.009914145,0.007894534,0.06563266,0.05470521,0.2798366,0.5377409],"study_design_scores_gemma":[0.01486639,0.008231825,0.006140566,0.00722239,0.000435562,0.001655044,0.001763234,0.01924019,0.05806261,0.07172659,0.8092781,0.001377546],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"protocol","genre_scores_codex":[0.009593935,0.000533756,0.3721606,0.001281529,0.00200455,0.5647371,0.01635148,0.01172436,0.02161263],"genre_scores_gemma":[0.00953465,0.0002344223,0.22904,0.0006476016,0.0002479107,0.7444164,0.002830936,0.0007942579,0.01225387],"genre_candidate":"protocol","genre_consensus":"protocol","teacher_disagreement_score":0.9469251,"threshold_uncertainty_score":0.4797615,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02425805445412338,"score_gpt":0.3878035321090887,"score_spread":0.3635454776549653,"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."}}