{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZPP7DBSKGE2R5ZQIRBFBYLNK3I","short_pith_number":"pith:ZPP7DBSK","schema_version":"1.0","canonical_sha256":"cbdff1864a31351ee608884a1c2daada20c77e75e298e6c32ecd995e9891e938","source":{"kind":"arxiv","id":"2603.08145","version":2},"attestation_state":"computed","paper":{"title":"DARC: Disagreement-Aware Alignment via Risk-Constrained Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jiaxiang Chen, Junfan Li, Langzhang Liang, Mingxi Zou, Qifan Wang, Xu Yinghui, Zenglin Xu","submitted_at":"2026-03-09T09:21:29Z","abstract_excerpt":"Preference-based alignment methods (e.g., RLHF, DPO) typically optimize a single scalar objective, implicitly averaging over heterogeneous human preferences. In practice, systematic annotator and user-group disagreement makes mean-reward maximization brittle and susceptible to proxy over-optimization. We propose **Disagreement-Aware Alignment via Risk-Constrained Decoding (DARC)**, a retraining-free inference-time method that frames response selection as distributionally robust, risk-sensitive decision making. Given multiple preference samples or scalable disagreement proxies, DARC reranks can"},"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":"2603.08145","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-03-09T09:21:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"f58128004520d61c93d0aeea7b5b093fd9c1d697a72d88f2f2cef0ecedbe0909","abstract_canon_sha256":"e7fb982655acf16cc36ef3aa3f6fee30cdf25acefbb08ac1e50d8bb86c74c068"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:08.040918Z","signature_b64":"xMe6A+LZopziY1puRNKvaEKJqlC+LFa5lHBlONCs5lMnUE8JzkLnkmPAm6I1wSy39dIfEie/S4qD52McfisLAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cbdff1864a31351ee608884a1c2daada20c77e75e298e6c32ecd995e9891e938","last_reissued_at":"2026-05-20T00:03:08.040008Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:08.040008Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DARC: Disagreement-Aware Alignment via Risk-Constrained Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jiaxiang Chen, Junfan Li, Langzhang Liang, Mingxi Zou, Qifan Wang, Xu Yinghui, Zenglin Xu","submitted_at":"2026-03-09T09:21:29Z","abstract_excerpt":"Preference-based alignment methods (e.g., RLHF, DPO) typically optimize a single scalar objective, implicitly averaging over heterogeneous human preferences. In practice, systematic annotator and user-group disagreement makes mean-reward maximization brittle and susceptible to proxy over-optimization. We propose **Disagreement-Aware Alignment via Risk-Constrained Decoding (DARC)**, a retraining-free inference-time method that frames response selection as distributionally robust, risk-sensitive decision making. Given multiple preference samples or scalable disagreement proxies, DARC reranks can"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.08145","kind":"arxiv","version":2},"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/2603.08145/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":"2603.08145","created_at":"2026-05-20T00:03:08.040131+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.08145v2","created_at":"2026-05-20T00:03:08.040131+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.08145","created_at":"2026-05-20T00:03:08.040131+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZPP7DBSKGE2R","created_at":"2026-05-20T00:03:08.040131+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZPP7DBSKGE2R5ZQI","created_at":"2026-05-20T00:03:08.040131+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZPP7DBSK","created_at":"2026-05-20T00:03:08.040131+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/ZPP7DBSKGE2R5ZQIRBFBYLNK3I","json":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I.json","graph_json":"https://pith.science/api/pith-number/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/graph.json","events_json":"https://pith.science/api/pith-number/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/events.json","paper":"https://pith.science/paper/ZPP7DBSK"},"agent_actions":{"view_html":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I","download_json":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I.json","view_paper":"https://pith.science/paper/ZPP7DBSK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.08145&json=true","fetch_graph":"https://pith.science/api/pith-number/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/graph.json","fetch_events":"https://pith.science/api/pith-number/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/action/storage_attestation","attest_author":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/action/author_attestation","sign_citation":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/action/citation_signature","submit_replication":"https://pith.science/pith/ZPP7DBSKGE2R5ZQIRBFBYLNK3I/action/replication_record"}},"created_at":"2026-05-20T00:03:08.040131+00:00","updated_at":"2026-05-20T00:03:08.040131+00:00"}