{"id":"W4416023900","doi":"10.48550/arxiv.2511.04053","title":"Interpreting Multi-Attribute Confounding through Numerical Attributes in Large Language Models","year":2025,"lang":"","type":"preprint","venue":"ArXiv.org","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Core Research for Evolutional Science and Technology; Institute for Catastrophic Loss Reduction","keywords":"Context (archaeology); Vulnerability (computing); Linear subspace; ENCODE; Language model; Hierarchy; Correlation; Numerical modeling","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","open_science"],"category_scores_codex":[0.003293583,0.001495477,0.00205384,0.0006820965,0.0008217056,0.0009776538,0.005482231,0.001135149,0.0002569111],"category_scores_gemma":[0.002378681,0.001705417,0.0007460575,0.002702566,0.0003513886,0.002732204,0.01023047,0.003805559,0.0009615202],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001414426,"about_ca_system_score_gemma":0.0009957838,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004375121,"about_ca_topic_score_gemma":0.000674861,"domain_scores_codex":[0.9881945,0.001178163,0.002923516,0.003613855,0.0009799731,0.003109947],"domain_scores_gemma":[0.9928992,0.001813948,0.001085121,0.003093024,0.0007054959,0.0004031464],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001430482,0.002335582,0.6804133,0.001154421,0.0005186831,0.001639843,0.1267931,0.1163456,0.002596157,0.06221028,0.0004394449,0.005410583],"study_design_scores_gemma":[0.0008269922,0.0001391903,0.006052582,0.003653514,0.00008402782,0.0000240835,0.008476486,0.9479513,0.02626161,0.003062906,0.00145196,0.002015367],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2887947,0.001817077,0.7018207,0.00155783,0.003089264,0.001206619,0.0001557141,0.0003719219,0.001186221],"genre_scores_gemma":[0.974957,0.0005187409,0.02011343,0.001804563,0.0003200208,0.0002626696,0.0001066402,0.00008126177,0.001835676],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8316057,"threshold_uncertainty_score":0.9998986,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1102617408806471,"score_gpt":0.3656238299806145,"score_spread":0.2553620890999674,"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."}}