{"id":"W7117116563","doi":"10.2139/ssrn.5960744","title":"Retrieval at Risk: Systematic Failures in Embedding-Based Retrieval for Nuclear Regulatory Compliance","year":2025,"lang":"","type":"preprint","venue":"SSRN Electronic Journal","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University","funders":"","keywords":"Reliability (semiconductor); False positive paradox; Domain (mathematical analysis); Similarity (geometry); Proxy (statistics); Scram; Precision and recall","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":["metaepi_narrow","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.007770878,0.001005289,0.001811925,0.0004157272,0.0008648509,0.0002059256,0.001825476,0.002275452,0.00003276774],"category_scores_gemma":[0.00363847,0.0009554239,0.001190591,0.0004056231,0.0006299194,0.00001007218,0.0008198098,0.005923028,0.0000172953],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003350342,"about_ca_system_score_gemma":0.007385008,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004312095,"about_ca_topic_score_gemma":0.0005722321,"domain_scores_codex":[0.990136,0.001105662,0.002070731,0.001553536,0.0007577694,0.004376334],"domain_scores_gemma":[0.9955081,0.0004122512,0.002018351,0.001283671,0.0004716321,0.0003059958],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.2047431,0.009210093,0.01636907,0.3936809,0.04618874,0.0006069956,0.006006821,0.06219883,0.1250377,0.04630902,0.02286079,0.06678798],"study_design_scores_gemma":[0.08495291,0.03753034,0.006607431,0.2031017,0.01311046,0.005158003,0.03054422,0.2294135,0.0488408,0.2041568,0.1155213,0.02106257],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8248828,0.1204378,0.04323284,0.002931496,0.003554878,0.00412055,0.0004201525,0.0001234911,0.000295999],"genre_scores_gemma":[0.9791111,0.01401854,0.001738693,0.0002274441,0.0007666882,0.00003135933,0.0001641138,0.0001147294,0.003827366],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1905792,"threshold_uncertainty_score":0.9992896,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01565000766946264,"score_gpt":0.2903834509394865,"score_spread":0.2747334432700239,"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."}}