{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:263E7XYFKIIS5LEMCQVDO37T6T","short_pith_number":"pith:263E7XYF","schema_version":"1.0","canonical_sha256":"d7b64fdf0552112eac8c142a376ff3f4ce4e87619000517207cdfa2eb2d3c6d7","source":{"kind":"arxiv","id":"2605.20201","version":1},"attestation_state":"computed","paper":{"title":"Long-Context Reasoning Through Proxy-Based Chain-of-Thought Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander Gurung, Irina Saparina, Miao Li, Mirella Lapata","submitted_at":"2026-04-06T16:44:17Z","abstract_excerpt":"Recent large language models support inputs of up to 10 million tokens, yet they perform poorly on long-context tasks that require complex reasoning. Such tasks can be solved using only a subset of the input -- a proxy context -- rather than the full sequence. Despite sharing the same underlying reasoning process, models exhibit a significant performance disparity between proxy and full contexts. To improve long-context reasoning, we propose ProxyCoT, a novel training framework that transfers reasoning capabilities from short proxy contexts to full long contexts. Specifically, we first obtain "},"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":"2605.20201","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-06T16:44:17Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"b19841224a6ccf2d7dd822133546a5f2912e9b9af2dc0db7a29d88aa57539dfe","abstract_canon_sha256":"64eebf9ed5fd5aa01b3f8a56219defd04cc4dead82fb78c9d8945f001c594b19"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T00:04:20.835934Z","signature_b64":"AEMMRjYk8KmNNqnb56OLhaTKjRZL+MZzgXjT535+iS4TOZfySL87qx9wvTsreAyb3AIavVzSbv+YGeA3kWcTCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d7b64fdf0552112eac8c142a376ff3f4ce4e87619000517207cdfa2eb2d3c6d7","last_reissued_at":"2026-05-21T00:04:20.835115Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T00:04:20.835115Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Long-Context Reasoning Through Proxy-Based Chain-of-Thought Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander Gurung, Irina Saparina, Miao Li, Mirella Lapata","submitted_at":"2026-04-06T16:44:17Z","abstract_excerpt":"Recent large language models support inputs of up to 10 million tokens, yet they perform poorly on long-context tasks that require complex reasoning. Such tasks can be solved using only a subset of the input -- a proxy context -- rather than the full sequence. Despite sharing the same underlying reasoning process, models exhibit a significant performance disparity between proxy and full contexts. To improve long-context reasoning, we propose ProxyCoT, a novel training framework that transfers reasoning capabilities from short proxy contexts to full long contexts. Specifically, we first obtain "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.20201","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/2605.20201/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":"2605.20201","created_at":"2026-05-21T00:04:20.835240+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.20201v1","created_at":"2026-05-21T00:04:20.835240+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.20201","created_at":"2026-05-21T00:04:20.835240+00:00"},{"alias_kind":"pith_short_12","alias_value":"263E7XYFKIIS","created_at":"2026-05-21T00:04:20.835240+00:00"},{"alias_kind":"pith_short_16","alias_value":"263E7XYFKIIS5LEM","created_at":"2026-05-21T00:04:20.835240+00:00"},{"alias_kind":"pith_short_8","alias_value":"263E7XYF","created_at":"2026-05-21T00:04:20.835240+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/263E7XYFKIIS5LEMCQVDO37T6T","json":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T.json","graph_json":"https://pith.science/api/pith-number/263E7XYFKIIS5LEMCQVDO37T6T/graph.json","events_json":"https://pith.science/api/pith-number/263E7XYFKIIS5LEMCQVDO37T6T/events.json","paper":"https://pith.science/paper/263E7XYF"},"agent_actions":{"view_html":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T","download_json":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T.json","view_paper":"https://pith.science/paper/263E7XYF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.20201&json=true","fetch_graph":"https://pith.science/api/pith-number/263E7XYFKIIS5LEMCQVDO37T6T/graph.json","fetch_events":"https://pith.science/api/pith-number/263E7XYFKIIS5LEMCQVDO37T6T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T/action/storage_attestation","attest_author":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T/action/author_attestation","sign_citation":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T/action/citation_signature","submit_replication":"https://pith.science/pith/263E7XYFKIIS5LEMCQVDO37T6T/action/replication_record"}},"created_at":"2026-05-21T00:04:20.835240+00:00","updated_at":"2026-05-21T00:04:20.835240+00:00"}