{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YMV7HAO3KGQ6SJ2YUIUCPYZKKU","short_pith_number":"pith:YMV7HAO3","schema_version":"1.0","canonical_sha256":"c32bf381db51a1e92758a22827e32a553f4e2dcb4dfd9fb66b20bad11d7cc465","source":{"kind":"arxiv","id":"2601.11960","version":3},"attestation_state":"computed","paper":{"title":"R$^2$PO: Decoupling Rollout and Inference Policies for LLM Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bingbing Xu, Bin Xie, Dan Zhang, Huawei Shen, Jingchu Wang, Xiaoqian Sun, Yige Yuan","submitted_at":"2026-01-17T08:30:50Z","abstract_excerpt":"Existing reinforcement learning methods for LLM reasoning implicitly assume that the policy generating training trajectories should coincide with the one producing inference responses. We argue that this is a misleading inductive bias: the optimization-optimal trajectory distribution favors informative gradients, whereas the inference-optimal response distribution emphasizes accuracy and consistency. Forcing both into a single policy entangles their gradients and suppresses exploration. We propose R$^2$PO (Residual Rollout Policy Optimization), which attaches a lightweight Residual Rollout-Hea"},"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":"2601.11960","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-01-17T08:30:50Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"758d75a3c59079e0d1ba6c19499b597bd33189c6f38511b6d1b73fe7ebe65c6a","abstract_canon_sha256":"93363cb2a2beafb6cd17d82eff26dc18cf8b3779177c0582ddb87897118ecb87"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-07T01:16:04.238005Z","signature_b64":"ATkStdSWhZOxZlI6dhThKibcG84i8eHI8zp0kFh2wha2OfLpA6bnFC2/h8UTj+ub+l/ozEHMyMjYtal9oYi4CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c32bf381db51a1e92758a22827e32a553f4e2dcb4dfd9fb66b20bad11d7cc465","last_reissued_at":"2026-07-07T01:16:04.237060Z","signature_status":"signed_v1","first_computed_at":"2026-07-07T01:16:04.237060Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"R$^2$PO: Decoupling Rollout and Inference Policies for LLM Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bingbing Xu, Bin Xie, Dan Zhang, Huawei Shen, Jingchu Wang, Xiaoqian Sun, Yige Yuan","submitted_at":"2026-01-17T08:30:50Z","abstract_excerpt":"Existing reinforcement learning methods for LLM reasoning implicitly assume that the policy generating training trajectories should coincide with the one producing inference responses. We argue that this is a misleading inductive bias: the optimization-optimal trajectory distribution favors informative gradients, whereas the inference-optimal response distribution emphasizes accuracy and consistency. Forcing both into a single policy entangles their gradients and suppresses exploration. We propose R$^2$PO (Residual Rollout Policy Optimization), which attaches a lightweight Residual Rollout-Hea"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.11960","kind":"arxiv","version":3},"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/2601.11960/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":"2601.11960","created_at":"2026-07-07T01:16:04.237167+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.11960v3","created_at":"2026-07-07T01:16:04.237167+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.11960","created_at":"2026-07-07T01:16:04.237167+00:00"},{"alias_kind":"pith_short_12","alias_value":"YMV7HAO3KGQ6","created_at":"2026-07-07T01:16:04.237167+00:00"},{"alias_kind":"pith_short_16","alias_value":"YMV7HAO3KGQ6SJ2Y","created_at":"2026-07-07T01:16:04.237167+00:00"},{"alias_kind":"pith_short_8","alias_value":"YMV7HAO3","created_at":"2026-07-07T01:16:04.237167+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/YMV7HAO3KGQ6SJ2YUIUCPYZKKU","json":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU.json","graph_json":"https://pith.science/api/pith-number/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/graph.json","events_json":"https://pith.science/api/pith-number/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/events.json","paper":"https://pith.science/paper/YMV7HAO3"},"agent_actions":{"view_html":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU","download_json":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU.json","view_paper":"https://pith.science/paper/YMV7HAO3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.11960&json=true","fetch_graph":"https://pith.science/api/pith-number/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/graph.json","fetch_events":"https://pith.science/api/pith-number/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/action/storage_attestation","attest_author":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/action/author_attestation","sign_citation":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/action/citation_signature","submit_replication":"https://pith.science/pith/YMV7HAO3KGQ6SJ2YUIUCPYZKKU/action/replication_record"}},"created_at":"2026-07-07T01:16:04.237167+00:00","updated_at":"2026-07-07T01:16:04.237167+00:00"}