{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MSMKKHDLUWLGALO2X6ZWHW43IN","short_pith_number":"pith:MSMKKHDL","schema_version":"1.0","canonical_sha256":"6498a51c6ba596602ddabfb363db9b4351a6312127a54e3e9ca23e3ee7962b20","source":{"kind":"arxiv","id":"2605.30251","version":1},"attestation_state":"computed","paper":{"title":"Same Evidence, Different Answers: Canonical-Context On-Policy Distillation for Multi-Turn Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chao Zhang, Jingtao Xu, Jinsheng Quan, Quanling Liu, Xing Shi, Yawei Luo, Yifan Zhu, Zhihui Li, Zizhuo Lin","submitted_at":"2026-05-28T17:14:29Z","abstract_excerpt":"Large language models (LLMs) often solve a task when all instructions are given in a single prompt, but fail when the same information is revealed gradually across turns. When a clean FULL prompt and a RAW-SHARDED conversation contain the same complete user evidence, the model should still arrive at the same answer. We argue that a key reason for this gap is self-anchored drift: responses produced under partial information introduce unsupported assumptions, and those assumptions later distort the final answer. To reduce this effect, we propose Canonical-Context On-Policy Distillation (CCOPD). "},"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":"2605.30251","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-28T17:14:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"37bc0da38cef79452f4d7d5034650fc4f50e70da36fb9827cebe73d0e49c4cfb","abstract_canon_sha256":"ccb08ed3ecd8f2d0fb8c815e64d9b5492096cadc9e76be0a6ac411f3bb4c3ae0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T02:06:14.037125Z","signature_b64":"hnddWpf5yQUgdhUKJexN1uoTF7VmWt3DlLwbJ59FCcCwruPRvt77TwFgnr0wztociM/lTQQFLzuKP/5oTQ/2Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6498a51c6ba596602ddabfb363db9b4351a6312127a54e3e9ca23e3ee7962b20","last_reissued_at":"2026-05-29T02:06:14.036638Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T02:06:14.036638Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Same Evidence, Different Answers: Canonical-Context On-Policy Distillation for Multi-Turn Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chao Zhang, Jingtao Xu, Jinsheng Quan, Quanling Liu, Xing Shi, Yawei Luo, Yifan Zhu, Zhihui Li, Zizhuo Lin","submitted_at":"2026-05-28T17:14:29Z","abstract_excerpt":"Large language models (LLMs) often solve a task when all instructions are given in a single prompt, but fail when the same information is revealed gradually across turns. When a clean FULL prompt and a RAW-SHARDED conversation contain the same complete user evidence, the model should still arrive at the same answer. We argue that a key reason for this gap is self-anchored drift: responses produced under partial information introduce unsupported assumptions, and those assumptions later distort the final answer. To reduce this effect, we propose Canonical-Context On-Policy Distillation (CCOPD). "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.30251","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/2605.30251/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":"2605.30251","created_at":"2026-05-29T02:06:14.036709+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.30251v1","created_at":"2026-05-29T02:06:14.036709+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.30251","created_at":"2026-05-29T02:06:14.036709+00:00"},{"alias_kind":"pith_short_12","alias_value":"MSMKKHDLUWLG","created_at":"2026-05-29T02:06:14.036709+00:00"},{"alias_kind":"pith_short_16","alias_value":"MSMKKHDLUWLGALO2","created_at":"2026-05-29T02:06:14.036709+00:00"},{"alias_kind":"pith_short_8","alias_value":"MSMKKHDL","created_at":"2026-05-29T02:06:14.036709+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/MSMKKHDLUWLGALO2X6ZWHW43IN","json":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN.json","graph_json":"https://pith.science/api/pith-number/MSMKKHDLUWLGALO2X6ZWHW43IN/graph.json","events_json":"https://pith.science/api/pith-number/MSMKKHDLUWLGALO2X6ZWHW43IN/events.json","paper":"https://pith.science/paper/MSMKKHDL"},"agent_actions":{"view_html":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN","download_json":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN.json","view_paper":"https://pith.science/paper/MSMKKHDL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.30251&json=true","fetch_graph":"https://pith.science/api/pith-number/MSMKKHDLUWLGALO2X6ZWHW43IN/graph.json","fetch_events":"https://pith.science/api/pith-number/MSMKKHDLUWLGALO2X6ZWHW43IN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN/action/storage_attestation","attest_author":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN/action/author_attestation","sign_citation":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN/action/citation_signature","submit_replication":"https://pith.science/pith/MSMKKHDLUWLGALO2X6ZWHW43IN/action/replication_record"}},"created_at":"2026-05-29T02:06:14.036709+00:00","updated_at":"2026-05-29T02:06:14.036709+00:00"}