{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:C5OAQIDVWRS7VK2DBSQEJMJBGH","short_pith_number":"pith:C5OAQIDV","canonical_record":{"source":{"id":"2603.20843","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-21T14:59:45Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"283d5556757bb7a51c6d023cf85c1c8dcaadf2a31005fbc27930de42889a31cd","abstract_canon_sha256":"7ae76e4229bc6e0770fbbc369b33ef81bc41c55f251c807dde3bae32087cebaf"},"schema_version":"1.0"},"canonical_sha256":"175c082075b465faab430ca044b12131dbcf915f56fb313f0b90c171df111d1c","source":{"kind":"arxiv","id":"2603.20843","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.20843","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"arxiv_version","alias_value":"2603.20843v3","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.20843","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"pith_short_12","alias_value":"C5OAQIDVWRS7","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"pith_short_16","alias_value":"C5OAQIDVWRS7VK2D","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"pith_short_8","alias_value":"C5OAQIDV","created_at":"2026-07-21T01:21:43Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:C5OAQIDVWRS7VK2DBSQEJMJBGH","target":"record","payload":{"canonical_record":{"source":{"id":"2603.20843","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-21T14:59:45Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"283d5556757bb7a51c6d023cf85c1c8dcaadf2a31005fbc27930de42889a31cd","abstract_canon_sha256":"7ae76e4229bc6e0770fbbc369b33ef81bc41c55f251c807dde3bae32087cebaf"},"schema_version":"1.0"},"canonical_sha256":"175c082075b465faab430ca044b12131dbcf915f56fb313f0b90c171df111d1c","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-21T01:21:43.806241Z","signature_b64":"9Bsoj2p9loLKwnSbrfXAbC9DdReJrYn8Y4xLh5vUaRB7DwUcDaSuJJcf4ifGKNoNyQIrM3XZQI3EvK+yrVJmCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"175c082075b465faab430ca044b12131dbcf915f56fb313f0b90c171df111d1c","last_reissued_at":"2026-07-21T01:21:43.805232Z","signature_status":"signed_v1","first_computed_at":"2026-07-21T01:21:43.805232Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2603.20843","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-21T01:21:43Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0WkvPhqQrTfwzT/TOn8v8OeUJJYS+U5wguhtqLfjjzwNSxIYxUF8Ed7CygIjha4/pHVBXMS71CaYDnDuoOAHBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T05:01:06.313392Z"},"content_sha256":"2e82e51ad6163e13bc1b5f0f61fc75b25b8c07608544cd2c2ca33f7d282c640c","schema_version":"1.0","event_id":"sha256:2e82e51ad6163e13bc1b5f0f61fc75b25b8c07608544cd2c2ca33f7d282c640c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:C5OAQIDVWRS7VK2DBSQEJMJBGH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"HiCI: Hierarchical Construction-Integration for Long-Context Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"HiCI builds explicit hierarchical attention by constructing segment representations, integrating them globally, and broadcasting the result to condition lower-level attention.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chang Xu, Qi Xu, Xiangyu Zeng, Yunke Wang","submitted_at":"2026-03-21T14:59:45Z","abstract_excerpt":"Long-context language modeling is commonly framed as a scalability challenge of token-level attention, yet local-to-global information structuring remains largely implicit in existing approaches. Drawing on cognitive theories of discourse comprehension, we propose HiCI (Hierarchical Construction--Integration), a hierarchical attention module that constructs segment-level representations, integrates them into a shared global context, and broadcasts both to condition segment-level attention. We validate HiCI through parameter-efficient adaptation of LLaMA-2 with only <5.5% additional parameters,"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the explicit hierarchical construction-integration structure, rather than other implementation details or training choices, is the primary driver of the reported gains when adapting LLaMA-2.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HiCI is a new hierarchical attention module that extends LLaMA-2 context from 4K to 100K tokens using under 5.5% extra parameters and improves results on retrieval and code tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"HiCI builds explicit hierarchical attention by constructing segment representations, integrating them globally, and broadcasting the result to condition lower-level attention.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"88144639629e90ca910e24d799ece410994af4ae39e754d54f6f9905079ab08d"},"source":{"id":"2603.20843","kind":"arxiv","version":3},"verdict":{"id":"b7a35ee6-3559-44f4-8004-e7426bedb02e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:44:14.610046Z","strongest_claim":"HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension.","one_line_summary":"HiCI is a new hierarchical attention module that extends LLaMA-2 context from 4K to 100K tokens using under 5.5% extra parameters and improves results on retrieval and code tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the explicit hierarchical construction-integration structure, rather than other implementation details or training choices, is the primary driver of the reported gains when adapting LLaMA-2.","pith_extraction_headline":"HiCI builds explicit hierarchical attention by constructing segment representations, integrating them globally, and broadcasting the result to condition lower-level attention."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.20843/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":2,"snapshot_sha256":"6238df0d13e6f17a8f5945e2aed3b65d0f3fd94deb81ab81aca15ebe98ba8956"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"b7a35ee6-3559-44f4-8004-e7426bedb02e"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-21T01:21:43Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"DhVLvBLTw9pLiSOUqos1hbNtRPbNzFArpPCNzl7bdu9aRjifeLU9v2fqGm5ab/YrOg9vvYUSjkOg/vNmxKzZBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T05:01:06.314541Z"},"content_sha256":"ee84ab9641c6b4358b5ca855d3ff821d8f60efc0a5b686333971822dd28ae699","schema_version":"1.0","event_id":"sha256:ee84ab9641c6b4358b5ca855d3ff821d8f60efc0a5b686333971822dd28ae699"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH/bundle.json","state_url":"https://pith.science/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-18T05:01:06Z","links":{"resolver":"https://pith.science/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH","bundle":"https://pith.science/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH/bundle.json","state":"https://pith.science/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/C5OAQIDVWRS7VK2DBSQEJMJBGH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:C5OAQIDVWRS7VK2DBSQEJMJBGH","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"7ae76e4229bc6e0770fbbc369b33ef81bc41c55f251c807dde3bae32087cebaf","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-21T14:59:45Z","title_canon_sha256":"283d5556757bb7a51c6d023cf85c1c8dcaadf2a31005fbc27930de42889a31cd"},"schema_version":"1.0","source":{"id":"2603.20843","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.20843","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"arxiv_version","alias_value":"2603.20843v3","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.20843","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"pith_short_12","alias_value":"C5OAQIDVWRS7","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"pith_short_16","alias_value":"C5OAQIDVWRS7VK2D","created_at":"2026-07-21T01:21:43Z"},{"alias_kind":"pith_short_8","alias_value":"C5OAQIDV","created_at":"2026-07-21T01:21:43Z"}],"graph_snapshots":[{"event_id":"sha256:ee84ab9641c6b4358b5ca855d3ff821d8f60efc0a5b686333971822dd28ae699","target":"graph","created_at":"2026-07-21T01:21:43Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the explicit hierarchical construction-integration structure, rather than other implementation details or training choices, is the primary driver of the reported gains when adapting LLaMA-2."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"HiCI is a new hierarchical attention module that extends LLaMA-2 context from 4K to 100K tokens using under 5.5% extra parameters and improves results on retrieval and code tasks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"HiCI builds explicit hierarchical attention by constructing segment representations, integrating them globally, and broadcasting the result to condition lower-level attention."}],"snapshot_sha256":"88144639629e90ca910e24d799ece410994af4ae39e754d54f6f9905079ab08d"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6238df0d13e6f17a8f5945e2aed3b65d0f3fd94deb81ab81aca15ebe98ba8956"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2603.20843/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Long-context language modeling is commonly framed as a scalability challenge of token-level attention, yet local-to-global information structuring remains largely implicit in existing approaches. Drawing on cognitive theories of discourse comprehension, we propose HiCI (Hierarchical Construction--Integration), a hierarchical attention module that constructs segment-level representations, integrates them into a shared global context, and broadcasts both to condition segment-level attention. We validate HiCI through parameter-efficient adaptation of LLaMA-2 with only <5.5% additional parameters,","authors_text":"Chang Xu, Qi Xu, Xiangyu Zeng, Yunke Wang","cross_cats":["cs.AI","cs.LG"],"headline":"HiCI builds explicit hierarchical attention by constructing segment representations, integrating them globally, and broadcasting the result to condition lower-level attention.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-21T14:59:45Z","title":"HiCI: Hierarchical Construction-Integration for Long-Context Attention"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.20843","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T06:44:14.610046Z","id":"b7a35ee6-3559-44f4-8004-e7426bedb02e","model_set":{"reader":"grok-4.3"},"one_line_summary":"HiCI is a new hierarchical attention module that extends LLaMA-2 context from 4K to 100K tokens using under 5.5% extra parameters and improves results on retrieval and code tasks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"HiCI builds explicit hierarchical attention by constructing segment representations, integrating them globally, and broadcasting the result to condition lower-level attention.","strongest_claim":"HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension.","weakest_assumption":"That the explicit hierarchical construction-integration structure, rather than other implementation details or training choices, is the primary driver of the reported gains when adapting LLaMA-2."}},"verdict_id":"b7a35ee6-3559-44f4-8004-e7426bedb02e"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:2e82e51ad6163e13bc1b5f0f61fc75b25b8c07608544cd2c2ca33f7d282c640c","target":"record","created_at":"2026-07-21T01:21:43Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"7ae76e4229bc6e0770fbbc369b33ef81bc41c55f251c807dde3bae32087cebaf","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-21T14:59:45Z","title_canon_sha256":"283d5556757bb7a51c6d023cf85c1c8dcaadf2a31005fbc27930de42889a31cd"},"schema_version":"1.0","source":{"id":"2603.20843","kind":"arxiv","version":3}},"canonical_sha256":"175c082075b465faab430ca044b12131dbcf915f56fb313f0b90c171df111d1c","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"175c082075b465faab430ca044b12131dbcf915f56fb313f0b90c171df111d1c","first_computed_at":"2026-07-21T01:21:43.805232Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-21T01:21:43.805232Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"9Bsoj2p9loLKwnSbrfXAbC9DdReJrYn8Y4xLh5vUaRB7DwUcDaSuJJcf4ifGKNoNyQIrM3XZQI3EvK+yrVJmCw==","signature_status":"signed_v1","signed_at":"2026-07-21T01:21:43.806241Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.20843","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:2e82e51ad6163e13bc1b5f0f61fc75b25b8c07608544cd2c2ca33f7d282c640c","sha256:ee84ab9641c6b4358b5ca855d3ff821d8f60efc0a5b686333971822dd28ae699"],"state_sha256":"379fcf5d9270d369f8728a86567f71010952b2b918a05ae83a33f7111095387a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mvC1Y3e8DnHTg1MVeuy8/hTemGXFolCANQ6jrVA0tyHMIhfWNR24Hnt8p+qe6Nd9q7rAa6zkIZD87BtArnImDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-18T05:01:06.321021Z","bundle_sha256":"cfb8f5ff1e1aa0b49922118817fed622739420d4c54ea594d411110e186e5131"}}