{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WMFHIL7VUGJN2Z3RXLIJ2IM4CY","short_pith_number":"pith:WMFHIL7V","schema_version":"1.0","canonical_sha256":"b30a742ff5a192dd6771bad09d219c1609e8a359721c6efc80ac66bb41b933de","source":{"kind":"arxiv","id":"2605.27996","version":1},"attestation_state":"computed","paper":{"title":"Reward Bias Substitution: Single-Axis Bias Mitigations Redirect Optimization Pressure","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Andreas Haupt, Daniel Fein, Marcel Hussing, Max Lamparth, Mykel J. Kochenderfer","submitted_at":"2026-05-27T05:40:22Z","abstract_excerpt":"Single-axis mitigations of reward-model biases (e.g., reducing proxy reliance on length, sycophancy, or style) can rotate optimization pressure onto correlated proxies rather than eliminate it, a failure mode we call reward bias substitution. The failure is enabled by a measurement-versus-optimization gap between audit and policy-induced distributions during mitigation evaluation and policy training. We formalize mitigation outcomes into a regime taxonomy and prove that successful mitigation, bias substitution, and overcorrection produce identical observables under any audit-distribution scori"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.27996","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-27T05:40:22Z","cross_cats_sorted":[],"title_canon_sha256":"a82fe9fc24db1cbb34a2c61975eb8ead7ddd0f7d8daf9c72239fd80d723d2f02","abstract_canon_sha256":"ce9dbd7c0ef454f4ae641de317082b3fc1ac5ad4efe77a412909cc1923cdeea7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:55.578727Z","signature_b64":"C/baupogX9GWnzehl23eTHfsWN5kP6tGA4IeqRyHdL4EFNpyDfvAfuCrnL68EFZy5tyBoGmRFm1VKzzcY9eKBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b30a742ff5a192dd6771bad09d219c1609e8a359721c6efc80ac66bb41b933de","last_reissued_at":"2026-05-28T01:04:55.578254Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:55.578254Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reward Bias Substitution: Single-Axis Bias Mitigations Redirect Optimization Pressure","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Andreas Haupt, Daniel Fein, Marcel Hussing, Max Lamparth, Mykel J. Kochenderfer","submitted_at":"2026-05-27T05:40:22Z","abstract_excerpt":"Single-axis mitigations of reward-model biases (e.g., reducing proxy reliance on length, sycophancy, or style) can rotate optimization pressure onto correlated proxies rather than eliminate it, a failure mode we call reward bias substitution. The failure is enabled by a measurement-versus-optimization gap between audit and policy-induced distributions during mitigation evaluation and policy training. We formalize mitigation outcomes into a regime taxonomy and prove that successful mitigation, bias substitution, and overcorrection produce identical observables under any audit-distribution scori"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.27996","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.27996/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.27996","created_at":"2026-05-28T01:04:55.578323+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.27996v1","created_at":"2026-05-28T01:04:55.578323+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.27996","created_at":"2026-05-28T01:04:55.578323+00:00"},{"alias_kind":"pith_short_12","alias_value":"WMFHIL7VUGJN","created_at":"2026-05-28T01:04:55.578323+00:00"},{"alias_kind":"pith_short_16","alias_value":"WMFHIL7VUGJN2Z3R","created_at":"2026-05-28T01:04:55.578323+00:00"},{"alias_kind":"pith_short_8","alias_value":"WMFHIL7V","created_at":"2026-05-28T01:04:55.578323+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY","json":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY.json","graph_json":"https://pith.science/api/pith-number/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/graph.json","events_json":"https://pith.science/api/pith-number/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/events.json","paper":"https://pith.science/paper/WMFHIL7V"},"agent_actions":{"view_html":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY","download_json":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY.json","view_paper":"https://pith.science/paper/WMFHIL7V","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.27996&json=true","fetch_graph":"https://pith.science/api/pith-number/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/graph.json","fetch_events":"https://pith.science/api/pith-number/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/action/storage_attestation","attest_author":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/action/author_attestation","sign_citation":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/action/citation_signature","submit_replication":"https://pith.science/pith/WMFHIL7VUGJN2Z3RXLIJ2IM4CY/action/replication_record"}},"created_at":"2026-05-28T01:04:55.578323+00:00","updated_at":"2026-05-28T01:04:55.578323+00:00"}