{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:E35PDHT44WTTR2YGKYUBEVC5KP","short_pith_number":"pith:E35PDHT4","schema_version":"1.0","canonical_sha256":"26faf19e7ce5a738eb06562812545d53facb9fe93779bd4cf57851d1eb3f7180","source":{"kind":"arxiv","id":"2607.18470","version":1},"attestation_state":"computed","paper":{"title":"RRPO: Reference-Relative Policy Optimization with Stratified Conditional Rollouts","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jingbo Shang, Julian McAuley, Junda Wu, Nikki Lijing Kuang, Rohan Surana, Ryan A. Rossi, Sheldon Yu, Tong Yu, Xintong Li, Xunyi Jiang, Yuxin Xiong","submitted_at":"2026-07-20T19:40:25Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has shown strong effectiveness in reinforcement learning from verifiable feedback, where sampled rollouts can be compared within a group using task-provided correctness signals. However, extending group-relative optimization beyond verifiable settings is challenging because success in many tasks is not captured by a single correctness criterion. We propose \\textbf{Reference-Relative Policy Optimization (RRPO)}, which generalizes GRPO by replacing direct correctness-based advantage construction with reference-relative contrastive comparisons. RRPO first"},"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":"2607.18470","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-20T19:40:25Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9ae80526c7104b311426804afe6ba8eca34deef471f7385c704a9650e7499845","abstract_canon_sha256":"a153a4b87dd0a315a34601da5d26f98cbd6f5351c6a4f5a78ae8bcac104afa2b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-22T00:22:50.294027Z","signature_b64":"269Qrxe9IWrYhtfXu5nbp1ZzehgLPPo87uD8yzDf7n6S5st9/MuSQjROpAz66rUZWQo6tA88zbznLEC/8Ol0Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"26faf19e7ce5a738eb06562812545d53facb9fe93779bd4cf57851d1eb3f7180","last_reissued_at":"2026-07-22T00:22:50.293045Z","signature_status":"signed_v1","first_computed_at":"2026-07-22T00:22:50.293045Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RRPO: Reference-Relative Policy Optimization with Stratified Conditional Rollouts","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jingbo Shang, Julian McAuley, Junda Wu, Nikki Lijing Kuang, Rohan Surana, Ryan A. Rossi, Sheldon Yu, Tong Yu, Xintong Li, Xunyi Jiang, Yuxin Xiong","submitted_at":"2026-07-20T19:40:25Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) has shown strong effectiveness in reinforcement learning from verifiable feedback, where sampled rollouts can be compared within a group using task-provided correctness signals. However, extending group-relative optimization beyond verifiable settings is challenging because success in many tasks is not captured by a single correctness criterion. We propose \\textbf{Reference-Relative Policy Optimization (RRPO)}, which generalizes GRPO by replacing direct correctness-based advantage construction with reference-relative contrastive comparisons. RRPO first"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.18470","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/2607.18470/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":"2607.18470","created_at":"2026-07-22T00:22:50.293524+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.18470v1","created_at":"2026-07-22T00:22:50.293524+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.18470","created_at":"2026-07-22T00:22:50.293524+00:00"},{"alias_kind":"pith_short_12","alias_value":"E35PDHT44WTT","created_at":"2026-07-22T00:22:50.293524+00:00"},{"alias_kind":"pith_short_16","alias_value":"E35PDHT44WTTR2YG","created_at":"2026-07-22T00:22:50.293524+00:00"},{"alias_kind":"pith_short_8","alias_value":"E35PDHT4","created_at":"2026-07-22T00:22:50.293524+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/E35PDHT44WTTR2YGKYUBEVC5KP","json":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP.json","graph_json":"https://pith.science/api/pith-number/E35PDHT44WTTR2YGKYUBEVC5KP/graph.json","events_json":"https://pith.science/api/pith-number/E35PDHT44WTTR2YGKYUBEVC5KP/events.json","paper":"https://pith.science/paper/E35PDHT4"},"agent_actions":{"view_html":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP","download_json":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP.json","view_paper":"https://pith.science/paper/E35PDHT4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.18470&json=true","fetch_graph":"https://pith.science/api/pith-number/E35PDHT44WTTR2YGKYUBEVC5KP/graph.json","fetch_events":"https://pith.science/api/pith-number/E35PDHT44WTTR2YGKYUBEVC5KP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP/action/storage_attestation","attest_author":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP/action/author_attestation","sign_citation":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP/action/citation_signature","submit_replication":"https://pith.science/pith/E35PDHT44WTTR2YGKYUBEVC5KP/action/replication_record"}},"created_at":"2026-07-22T00:22:50.293524+00:00","updated_at":"2026-07-22T00:22:50.293524+00:00"}