{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:QDRCQ3OSEALCE7DLXINMPSIQ2P","short_pith_number":"pith:QDRCQ3OS","schema_version":"1.0","canonical_sha256":"80e2286dd22016227c6bba1ac7c910d3d45772e9e4cf2c379247223e1a0684b6","source":{"kind":"arxiv","id":"2605.28014","version":1},"attestation_state":"computed","paper":{"title":"ROSD: Reflective On-Policy Self-Distillation for Language Model Reasoning across Domains","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Daiting Shi, Jingzhou He, Liu Yang, Xiao-Ming Wu, Xin Xin, Xinyu Ma, Yujie Feng, Zhaochun Ren, Ziqi Zhao","submitted_at":"2026-05-27T06:09:29Z","abstract_excerpt":"On-policy self-distillation (OPSD) improves the reasoning performance of large language models (LLMs) by providing dense token-level supervision for on-policy rollouts. However, existing OPSD methods often yield limited gains on in-domain reasoning and generalize poorly to out-of-domain problems. We identify two key causes: conditioning the self-teacher on a verified solution encourages imitation of training-domain reference trajectories rather than error-specific correction, and applying distillation to the full response can overwrite valid reasoning prefixes and reinforce overfitting.\n  We p"},"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":"2605.28014","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-27T06:09:29Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"30c404646821034cc125895a153e205e5fba914aa26973381365bf85447b1c05","abstract_canon_sha256":"6e84775626cb69c734ae7fbee91a5684618aa168deb25e10a3e8eb378e244bff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:56.038961Z","signature_b64":"0+I1/oV3yFiuB1rhbVWij/VzW98MnDMtOrajq01f26Zq/CAXSz/nXmXb36HMG+l7Pcka1ApDkNq41CkAC3jKCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"80e2286dd22016227c6bba1ac7c910d3d45772e9e4cf2c379247223e1a0684b6","last_reissued_at":"2026-05-28T01:04:56.038579Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:56.038579Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ROSD: Reflective On-Policy Self-Distillation for Language Model Reasoning across Domains","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Daiting Shi, Jingzhou He, Liu Yang, Xiao-Ming Wu, Xin Xin, Xinyu Ma, Yujie Feng, Zhaochun Ren, Ziqi Zhao","submitted_at":"2026-05-27T06:09:29Z","abstract_excerpt":"On-policy self-distillation (OPSD) improves the reasoning performance of large language models (LLMs) by providing dense token-level supervision for on-policy rollouts. However, existing OPSD methods often yield limited gains on in-domain reasoning and generalize poorly to out-of-domain problems. We identify two key causes: conditioning the self-teacher on a verified solution encourages imitation of training-domain reference trajectories rather than error-specific correction, and applying distillation to the full response can overwrite valid reasoning prefixes and reinforce overfitting.\n  We p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28014","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/2605.28014/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":"2605.28014","created_at":"2026-05-28T01:04:56.038629+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28014v1","created_at":"2026-05-28T01:04:56.038629+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28014","created_at":"2026-05-28T01:04:56.038629+00:00"},{"alias_kind":"pith_short_12","alias_value":"QDRCQ3OSEALC","created_at":"2026-05-28T01:04:56.038629+00:00"},{"alias_kind":"pith_short_16","alias_value":"QDRCQ3OSEALCE7DL","created_at":"2026-05-28T01:04:56.038629+00:00"},{"alias_kind":"pith_short_8","alias_value":"QDRCQ3OS","created_at":"2026-05-28T01:04:56.038629+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/QDRCQ3OSEALCE7DLXINMPSIQ2P","json":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P.json","graph_json":"https://pith.science/api/pith-number/QDRCQ3OSEALCE7DLXINMPSIQ2P/graph.json","events_json":"https://pith.science/api/pith-number/QDRCQ3OSEALCE7DLXINMPSIQ2P/events.json","paper":"https://pith.science/paper/QDRCQ3OS"},"agent_actions":{"view_html":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P","download_json":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P.json","view_paper":"https://pith.science/paper/QDRCQ3OS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28014&json=true","fetch_graph":"https://pith.science/api/pith-number/QDRCQ3OSEALCE7DLXINMPSIQ2P/graph.json","fetch_events":"https://pith.science/api/pith-number/QDRCQ3OSEALCE7DLXINMPSIQ2P/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P/action/storage_attestation","attest_author":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P/action/author_attestation","sign_citation":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P/action/citation_signature","submit_replication":"https://pith.science/pith/QDRCQ3OSEALCE7DLXINMPSIQ2P/action/replication_record"}},"created_at":"2026-05-28T01:04:56.038629+00:00","updated_at":"2026-05-28T01:04:56.038629+00:00"}