{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003628137,0.0001139352,0.0001875811,0.0001625938,0.0004828908,0.0002554826,0.0006072155,0.00006461031,0.000004242115],"category_scores_gemma":[0.0006498746,0.0000917518,0.0001144984,0.0006250221,0.00007218348,0.0003472195,0.0001658458,0.0009310001,4.637319e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005427133,"about_ca_system_score_gemma":0.002608091,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001909601,"about_ca_topic_score_gemma":0.00009798691,"domain_scores_codex":[0.9975808,0.0001605287,0.0003078112,0.0005769721,0.0002449267,0.001129014],"domain_scores_gemma":[0.9986409,0.0001546789,0.00009884867,0.0007188185,0.0003194952,0.00006725996],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001767036,0.0003439482,0.01229982,0.0001737108,0.0004162752,0.00000114861,0.00006420748,0.0009437416,0.000422822,0.4812356,0.0005120732,0.5034099],"study_design_scores_gemma":[0.001326786,0.0003098099,0.003261053,0.00003068712,0.00002372046,0.000109043,0.0001305535,0.02553784,0.0003985085,0.9614484,0.007259018,0.000164604],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.005637874,0.00005459625,0.9666086,0.003466907,0.00006332933,0.02402423,7.992048e-7,0.00006427068,0.00007932253],"genre_scores_gemma":[0.5743824,0.0002654426,0.1226255,0.0003605957,0.0007448906,0.2914259,0.000006319789,0.00004939492,0.01013964],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8439832,"threshold_uncertainty_score":0.4626642,"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."}}