{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7QDI7YHUG7WBGHLATEAWJA55BF","short_pith_number":"pith:7QDI7YHU","schema_version":"1.0","canonical_sha256":"fc068fe0f437ec131d6099016483bd094f8021c4ada9fd7f929a2daa0707a988","source":{"kind":"arxiv","id":"2608.10126","version":1},"attestation_state":"computed","paper":{"title":"Procedural Fairness Failures in RLHF from Preference Averaging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"John Joshua, Karthik Kamuju, Kummari Avinash, M P V S Gopinadh, Srinivasa Raju Rudraraju","submitted_at":"2026-08-10T18:38:16Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) aggregates heterogeneous preferences into a single reward model, assuming preference homogeneity. When preferences are heterogeneous, this aggregation induces a procedural fairness failure where majority preference groups dominate reward learning while minority preferences are systematically under-represented. This work defines procedural fairness in alignment as preserving distinct preference signals during reward modeling and shows that standard RLHF violates this via preference averaging. Preference-Aware RLHF (PA-RLHF) is introduced, separa"},"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":"2608.10126","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-08-10T18:38:16Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"164d24b18bc13cdec7285933fc0a76c0d400dccc5a22687e70bc0562b1f42651","abstract_canon_sha256":"5ffc56feb19978b9e60c3e0b9a8b5894dd2074ebf5f959270f2e44aa66a3f71e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-12T00:22:52.157759Z","signature_b64":"G2X4UCFvab417BEXXWo+jdJkb+Fkjc/Fdj2istaLpAeqa76SF4fBX744XI3brCYz+2VKiiVhZN5vkf+DuKu4Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fc068fe0f437ec131d6099016483bd094f8021c4ada9fd7f929a2daa0707a988","last_reissued_at":"2026-08-12T00:22:52.156291Z","signature_status":"signed_v1","first_computed_at":"2026-08-12T00:22:52.156291Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Procedural Fairness Failures in RLHF from Preference Averaging","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"John Joshua, Karthik Kamuju, Kummari Avinash, M P V S Gopinadh, Srinivasa Raju Rudraraju","submitted_at":"2026-08-10T18:38:16Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) aggregates heterogeneous preferences into a single reward model, assuming preference homogeneity. When preferences are heterogeneous, this aggregation induces a procedural fairness failure where majority preference groups dominate reward learning while minority preferences are systematically under-represented. This work defines procedural fairness in alignment as preserving distinct preference signals during reward modeling and shows that standard RLHF violates this via preference averaging. Preference-Aware RLHF (PA-RLHF) is introduced, separa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.10126","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/2608.10126/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":"2608.10126","created_at":"2026-08-12T00:22:52.158944+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.10126v1","created_at":"2026-08-12T00:22:52.158944+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.10126","created_at":"2026-08-12T00:22:52.158944+00:00"},{"alias_kind":"pith_short_12","alias_value":"7QDI7YHUG7WB","created_at":"2026-08-12T00:22:52.158944+00:00"},{"alias_kind":"pith_short_16","alias_value":"7QDI7YHUG7WBGHLA","created_at":"2026-08-12T00:22:52.158944+00:00"},{"alias_kind":"pith_short_8","alias_value":"7QDI7YHU","created_at":"2026-08-12T00:22:52.158944+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/7QDI7YHUG7WBGHLATEAWJA55BF","json":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF.json","graph_json":"https://pith.science/api/pith-number/7QDI7YHUG7WBGHLATEAWJA55BF/graph.json","events_json":"https://pith.science/api/pith-number/7QDI7YHUG7WBGHLATEAWJA55BF/events.json","paper":"https://pith.science/paper/7QDI7YHU"},"agent_actions":{"view_html":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF","download_json":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF.json","view_paper":"https://pith.science/paper/7QDI7YHU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.10126&json=true","fetch_graph":"https://pith.science/api/pith-number/7QDI7YHUG7WBGHLATEAWJA55BF/graph.json","fetch_events":"https://pith.science/api/pith-number/7QDI7YHUG7WBGHLATEAWJA55BF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF/action/storage_attestation","attest_author":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF/action/author_attestation","sign_citation":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF/action/citation_signature","submit_replication":"https://pith.science/pith/7QDI7YHUG7WBGHLATEAWJA55BF/action/replication_record"}},"created_at":"2026-08-12T00:22:52.158944+00:00","updated_at":"2026-08-12T00:22:52.158944+00:00"}