{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NHFCSFEZVIN6IRC4DDIRI3MOUJ","short_pith_number":"pith:NHFCSFEZ","schema_version":"1.0","canonical_sha256":"69ca291499aa1be4445c18d1146d8ea26716fd90b8c7cc2f038b28045766b076","source":{"kind":"arxiv","id":"2605.17295","version":1},"attestation_state":"computed","paper":{"title":"DISA: Offline Importance Sampling for Distribution-Matching LLM-RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Dayiheng Liu, Huiqiang Jiang, Linfeng Zhang, Shaobo Wang, Sihang Li, Wenjie Qiu, Xingzhang Ren, Xuming Hu, Yafeng Sun, Yucheng Li, Yujie Chen, Zhihui Xie","submitted_at":"2026-05-17T07:14:44Z","abstract_excerpt":"Modern reasoning agents are increasingly evaluated on their ability to generate multiple valid solution paths, plans, or tool-use traces for a given input. Standard reward-maximizing RL tends to collapse onto the most easily reinforced high-reward mode, whereas distribution-matching RL aims to allocate probability mass across the entire reward-shaped solution set. Achieving this objective requires computing a prompt-dependent partition function over the trajectory space. Because existing distribution-matching methods learn this partition function online alongside the policy, calibration errors"},"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":"2605.17295","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-17T07:14:44Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"f29fd610c3e259c302f1b719e1bff0263e61b5f0fa3cd081e1d2280290310f91","abstract_canon_sha256":"013c560828e343e65420b293d4555dd944a7d8a855dd1b96f5ac880ac17246ba"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:50.737758Z","signature_b64":"sR0EgWF+u52F9FbIIRzFYcMgd24PnAfG4pJc8Pofm80rI1qzu6JG4JxOlShaj8iFAqar/vWdyf3bVxZEsW/qBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"69ca291499aa1be4445c18d1146d8ea26716fd90b8c7cc2f038b28045766b076","last_reissued_at":"2026-05-20T00:03:50.736857Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:50.736857Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DISA: Offline Importance Sampling for Distribution-Matching LLM-RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Dayiheng Liu, Huiqiang Jiang, Linfeng Zhang, Shaobo Wang, Sihang Li, Wenjie Qiu, Xingzhang Ren, Xuming Hu, Yafeng Sun, Yucheng Li, Yujie Chen, Zhihui Xie","submitted_at":"2026-05-17T07:14:44Z","abstract_excerpt":"Modern reasoning agents are increasingly evaluated on their ability to generate multiple valid solution paths, plans, or tool-use traces for a given input. Standard reward-maximizing RL tends to collapse onto the most easily reinforced high-reward mode, whereas distribution-matching RL aims to allocate probability mass across the entire reward-shaped solution set. Achieving this objective requires computing a prompt-dependent partition function over the trajectory space. Because existing distribution-matching methods learn this partition function online alongside the policy, calibration errors"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.17295","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/2605.17295/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-19T22:01:57.809570Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.763895Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"f431f76d5be6e2c9584cd28390f2ace3c73e55d459858a30c3ce1eb442bd25f9"},"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":"2605.17295","created_at":"2026-05-20T00:03:50.737009+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.17295v1","created_at":"2026-05-20T00:03:50.737009+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.17295","created_at":"2026-05-20T00:03:50.737009+00:00"},{"alias_kind":"pith_short_12","alias_value":"NHFCSFEZVIN6","created_at":"2026-05-20T00:03:50.737009+00:00"},{"alias_kind":"pith_short_16","alias_value":"NHFCSFEZVIN6IRC4","created_at":"2026-05-20T00:03:50.737009+00:00"},{"alias_kind":"pith_short_8","alias_value":"NHFCSFEZ","created_at":"2026-05-20T00:03:50.737009+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/NHFCSFEZVIN6IRC4DDIRI3MOUJ","json":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ.json","graph_json":"https://pith.science/api/pith-number/NHFCSFEZVIN6IRC4DDIRI3MOUJ/graph.json","events_json":"https://pith.science/api/pith-number/NHFCSFEZVIN6IRC4DDIRI3MOUJ/events.json","paper":"https://pith.science/paper/NHFCSFEZ"},"agent_actions":{"view_html":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ","download_json":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ.json","view_paper":"https://pith.science/paper/NHFCSFEZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.17295&json=true","fetch_graph":"https://pith.science/api/pith-number/NHFCSFEZVIN6IRC4DDIRI3MOUJ/graph.json","fetch_events":"https://pith.science/api/pith-number/NHFCSFEZVIN6IRC4DDIRI3MOUJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ/action/storage_attestation","attest_author":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ/action/author_attestation","sign_citation":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ/action/citation_signature","submit_replication":"https://pith.science/pith/NHFCSFEZVIN6IRC4DDIRI3MOUJ/action/replication_record"}},"created_at":"2026-05-20T00:03:50.737009+00:00","updated_at":"2026-05-20T00:03:50.737009+00:00"}