{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:DWA4HPO55GPCEBJ4FGL62TDYBX","short_pith_number":"pith:DWA4HPO5","canonical_record":{"source":{"id":"2605.16598","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.MA","submitted_at":"2026-05-15T19:59:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6a675f605aec9879c6309d02a584fcd235c7ce793ea5105999ba43284abfee11","abstract_canon_sha256":"1831d9b3d02a69af3812db20e6f0c03792bd11c7b3ee7a1f535ff41598727ef7"},"schema_version":"1.0"},"canonical_sha256":"1d81c3bddde99e22053c2997ed4c780dd66ad77d2a59c6749a871c9dee74d00d","source":{"kind":"arxiv","id":"2605.16598","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.16598","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"arxiv_version","alias_value":"2605.16598v1","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16598","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"pith_short_12","alias_value":"DWA4HPO55GPC","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"pith_short_16","alias_value":"DWA4HPO55GPCEBJ4","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"pith_short_8","alias_value":"DWA4HPO5","created_at":"2026-05-20T00:02:31Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:DWA4HPO55GPCEBJ4FGL62TDYBX","target":"record","payload":{"canonical_record":{"source":{"id":"2605.16598","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.MA","submitted_at":"2026-05-15T19:59:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6a675f605aec9879c6309d02a584fcd235c7ce793ea5105999ba43284abfee11","abstract_canon_sha256":"1831d9b3d02a69af3812db20e6f0c03792bd11c7b3ee7a1f535ff41598727ef7"},"schema_version":"1.0"},"canonical_sha256":"1d81c3bddde99e22053c2997ed4c780dd66ad77d2a59c6749a871c9dee74d00d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:02:31.834475Z","signature_b64":"XwsdSXB8FqmHpOrg5XONT3QA911GZngWQo4wY5GRF3sggQr3PcAocB/NBvk5oIvpHzSRwnY082kiSoQtxohICw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1d81c3bddde99e22053c2997ed4c780dd66ad77d2a59c6749a871c9dee74d00d","last_reissued_at":"2026-05-20T00:02:31.833520Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:02:31.833520Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.16598","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:02:31Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"xGIrzaGhQMIcPKlueaRpy/kp3qxnpTznqqg1w/OBYFMWt4C5iMmm6/2b3DegoeWSL7Hx8HkUuYMXWcxLA+Q7Cg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T12:30:21.913946Z"},"content_sha256":"76a2adbfde54fc33a12285b77c20f1d8131ee36d944d5df3db653ae0845c4406","schema_version":"1.0","event_id":"sha256:76a2adbfde54fc33a12285b77c20f1d8131ee36d944d5df3db653ae0845c4406"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:DWA4HPO55GPCEBJ4FGL62TDYBX","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"GRASP: Graph Agentic Search over Propositions for Multi-hop Question Answering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"GRASP achieves top accuracy on multi-hop QA benchmarks while using 40 to 50 percent fewer tokens through hierarchical graph search.","cross_cats":["cs.AI"],"primary_cat":"cs.MA","authors_text":"Junjie Hu, Ramya Korlakai Vinayak, Stockton Jenkins","submitted_at":"2026-05-15T19:59:35Z","abstract_excerpt":"Agentic retrieval improves multi-hop question answering by giving language models autonomy to iteratively gather evidence. Recent work augments these systems with knowledge graphs for structured traversal, but this combination introduces significant cost: expensive graph construction at index time and compounding token usage at inference time. We introduce Graph Agentic Search over Propositions (GRASP), an agentic system that simultaneously optimizes for high accuracy and minimal token usage in multi-hop question answering. Rather than executing a rigid, singular query, GRASP actively coordina"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"GRASP achieves the highest QA accuracy in the open retrieval setting on MuSiQue and 2Wiki while using 40-50 percent fewer tokens than IRCoT+HippoRAG2, and leads on EM and F1 across all three datasets in the LongBench setting while using 30 percent fewer tokens than the next most accurate method.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The paper assumes that the novel three-layer hierarchical graph of entities, propositions, and passages can be built and traversed at index and inference time without introducing prohibitive construction costs or retrieval noise that would erase the reported token savings.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GRASP introduces a hierarchical graph-based agentic retrieval method that achieves top accuracy on MuSiQue, 2WikiMultihopQA, and HotpotQA while using 30-50% fewer tokens than strong baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GRASP achieves top accuracy on multi-hop QA benchmarks while using 40 to 50 percent fewer tokens through hierarchical graph search.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a72b1845f752b90b48a414af78fa809478af4e990e18620823b3597f9c7ebaac"},"source":{"id":"2605.16598","kind":"arxiv","version":1},"verdict":{"id":"1b53e57b-b0d4-4220-87dd-c14587741ab3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:44:27.748889Z","strongest_claim":"GRASP achieves the highest QA accuracy in the open retrieval setting on MuSiQue and 2Wiki while using 40-50 percent fewer tokens than IRCoT+HippoRAG2, and leads on EM and F1 across all three datasets in the LongBench setting while using 30 percent fewer tokens than the next most accurate method.","one_line_summary":"GRASP introduces a hierarchical graph-based agentic retrieval method that achieves top accuracy on MuSiQue, 2WikiMultihopQA, and HotpotQA while using 30-50% fewer tokens than strong baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The paper assumes that the novel three-layer hierarchical graph of entities, propositions, and passages can be built and traversed at index and inference time without introducing prohibitive construction costs or retrieval noise that would erase the reported token savings.","pith_extraction_headline":"GRASP achieves top accuracy on multi-hop QA benchmarks while using 40 to 50 percent fewer tokens through hierarchical graph search."},"integrity":{"clean":false,"summary":{"advisory":1,"critical":0,"by_detector":{"doi_compliance":{"total":1,"advisory":1,"critical":0,"informational":0}},"informational":0},"endpoint":"/pith/2605.16598/integrity.json","findings":[{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1162/tacla) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":2,"audited_at":"2026-05-19T20:51:23.751169Z","detected_doi":"10.1162/tacla","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.330696Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:51:23.751169Z","status":"completed","version":"1.0.0","findings_count":1},{"name":"claim_evidence","ran_at":"2026-05-19T19:21:56.819470Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.602561Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"74e9a3d7d3a7199b8cfa2916199893753cb8a744d689c8c7b39c1536fc8ae82f"},"references":{"count":39,"sample":[{"doi":"10.18653/v1/2023.findings-acl.565","year":1904,"title":"Understanding Dataset Design Choices for Multi-hop Reasoning","work_id":"38d92463-996e-4540-a8ab-f492c5df8328","ref_index":1,"cited_arxiv_id":"1904.12106","is_internal_anchor":true},{"doi":"10.1162/tacl","year":2025,"title":"Lost in the Middle: How Language Models Use Long Contexts","work_id":"37c05e13-4a24-44f8-a1c4-da1bbe7223aa","ref_index":2,"cited_arxiv_id":"2307.03172","is_internal_anchor":true},{"doi":"","year":2023,"title":"Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions","work_id":"2f30a088-53fc-4701-bbc0-c67bb54bf883","ref_index":3,"cited_arxiv_id":"2212.10509","is_internal_anchor":true},{"doi":"","year":null,"title":"Write arational plan--- a 1--3 sentence outline of the reasoning chain and key facts needed","work_id":"54b77551-b15e-455d-ba04-fcdb2432ed62","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Each will be answered by a research agent searching the knowledge graph","work_id":"e384dd1b-eb5f-4bcd-9a9d-2f1d124371eb","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":39,"snapshot_sha256":"12bf867daac9581fad644330cad9938347b0d64bc44a26f82892737663352eed","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7fd77a071a5159721bd0be1625ecc0e2100517798c34c3b1db8668dff28cb66b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"1b53e57b-b0d4-4220-87dd-c14587741ab3"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:02:31Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"wBXI2JPyzYSXE5IVaRwfY/1RLhUzdjb7pcPqFC33w1KnVpUVlQ8aGoz1/ohf8ziyxz2B7M5KMawBbBKyF9CVDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T12:30:21.914878Z"},"content_sha256":"37bbf203b0eaf4625c2bcc52c4a4dc1eba8201ba2ac64cb8749f8f06f5893dbf","schema_version":"1.0","event_id":"sha256:37bbf203b0eaf4625c2bcc52c4a4dc1eba8201ba2ac64cb8749f8f06f5893dbf"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:DWA4HPO55GPCEBJ4FGL62TDYBX","target":"integrity","payload":{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1162/tacla) was visible in the surrounding text but could not be confirmed against doi.org as printed.","snippet":"URLhttps://arxiv.org/abs/2408.08172. Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang. Lightrag: Simple and fast retrieval-augmented generation, 2025. URLhttps://arxiv.org/abs/2410.05779. Bernal Jim´enez Guti´errez, Yiheng Shu, Wei","arxiv_id":"2605.16598","detector":"doi_compliance","evidence":{"ref_index":2,"verdict_class":"incontrovertible","resolved_title":null,"printed_excerpt":"URLhttps://arxiv.org/abs/2408.08172. Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang. Lightrag: Simple and fast retrieval-augmented generation, 2025. URLhttps://arxiv.org/abs/2410.05779. Bernal Jim´enez Guti´errez, Yiheng Shu, Wei","reconstructed_doi":"10.1162/tacla"},"severity":"advisory","ref_index":2,"audited_at":"2026-05-19T20:51:23.751169Z","event_type":"pith.integrity.v1","detected_doi":"10.1162/tacla","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"recoverable_identifier","evidence_hash":"698808e5c492a8cde4e16ef86fd7f213d8d31c9fff1bde1255a506ae4f2a765c","paper_version":1,"verdict_class":"incontrovertible","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":3000,"payload_sha256":"aa578e69ee2b58094fcb8ed2dbb61eacaf31ac4d64335f22d11e5a3c3240f093","signature_b64":"mo34W9pN5gOXhTH5O0RoLlMdXlWosbeqe9mJuNtcR4pK7t0a932J2i2HXj8MEJRXW2kd6QBcFLj7kMT05SpvCw==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T20:52:22Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EezF8iDItsvlgUPQPIpPCAefxmx/BIjUdkd5zoxvoTzdtwsXeECWORiE/dPaUAVtTxaHEMYtGaXGz5Qqb5r9CA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-21T12:30:21.915780Z"},"content_sha256":"906c5292d76e8258c585988def40825b6efed503a377ced55d065704e40ba8ed","schema_version":"1.0","event_id":"sha256:906c5292d76e8258c585988def40825b6efed503a377ced55d065704e40ba8ed"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/DWA4HPO55GPCEBJ4FGL62TDYBX/bundle.json","state_url":"https://pith.science/pith/DWA4HPO55GPCEBJ4FGL62TDYBX/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/DWA4HPO55GPCEBJ4FGL62TDYBX/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":"2026-05-21T12:30:21Z","links":{"resolver":"https://pith.science/pith/DWA4HPO55GPCEBJ4FGL62TDYBX","bundle":"https://pith.science/pith/DWA4HPO55GPCEBJ4FGL62TDYBX/bundle.json","state":"https://pith.science/pith/DWA4HPO55GPCEBJ4FGL62TDYBX/state.json","well_known_bundle":"https://pith.science/.well-known/pith/DWA4HPO55GPCEBJ4FGL62TDYBX/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:DWA4HPO55GPCEBJ4FGL62TDYBX","merge_version":"pith-open-graph-merge-v1","event_count":3,"valid_event_count":3,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"1831d9b3d02a69af3812db20e6f0c03792bd11c7b3ee7a1f535ff41598727ef7","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.MA","submitted_at":"2026-05-15T19:59:35Z","title_canon_sha256":"6a675f605aec9879c6309d02a584fcd235c7ce793ea5105999ba43284abfee11"},"schema_version":"1.0","source":{"id":"2605.16598","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.16598","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"arxiv_version","alias_value":"2605.16598v1","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16598","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"pith_short_12","alias_value":"DWA4HPO55GPC","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"pith_short_16","alias_value":"DWA4HPO55GPCEBJ4","created_at":"2026-05-20T00:02:31Z"},{"alias_kind":"pith_short_8","alias_value":"DWA4HPO5","created_at":"2026-05-20T00:02:31Z"}],"graph_snapshots":[{"event_id":"sha256:37bbf203b0eaf4625c2bcc52c4a4dc1eba8201ba2ac64cb8749f8f06f5893dbf","target":"graph","created_at":"2026-05-20T00:02:31Z","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":"GRASP achieves the highest QA accuracy in the open retrieval setting on MuSiQue and 2Wiki while using 40-50 percent fewer tokens than IRCoT+HippoRAG2, and leads on EM and F1 across all three datasets in the LongBench setting while using 30 percent fewer tokens than the next most accurate method."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The paper assumes that the novel three-layer hierarchical graph of entities, propositions, and passages can be built and traversed at index and inference time without introducing prohibitive construction costs or retrieval noise that would erase the reported token savings."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GRASP introduces a hierarchical graph-based agentic retrieval method that achieves top accuracy on MuSiQue, 2WikiMultihopQA, and HotpotQA while using 30-50% fewer tokens than strong baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"GRASP achieves top accuracy on multi-hop QA benchmarks while using 40 to 50 percent fewer tokens through hierarchical graph search."}],"snapshot_sha256":"a72b1845f752b90b48a414af78fa809478af4e990e18620823b3597f9c7ebaac"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7fd77a071a5159721bd0be1625ecc0e2100517798c34c3b1db8668dff28cb66b"},"integrity":{"available":true,"clean":false,"detectors_run":[{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.330696Z","status":"completed","version":"1.0.0"},{"findings_count":1,"name":"doi_compliance","ran_at":"2026-05-19T20:51:23.751169Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T19:21:56.819470Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.602561Z","status":"skipped","version":"1.0.0"}],"endpoint":"/pith/2605.16598/integrity.json","findings":[{"audited_at":"2026-05-19T20:51:23.751169Z","detected_arxiv_id":null,"detected_doi":"10.1162/tacla","detector":"doi_compliance","finding_type":"recoverable_identifier","note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1162/tacla) was visible in the surrounding text but could not be confirmed against doi.org as printed.","ref_index":2,"severity":"advisory","verdict_class":"incontrovertible"}],"snapshot_sha256":"74e9a3d7d3a7199b8cfa2916199893753cb8a744d689c8c7b39c1536fc8ae82f","summary":{"advisory":1,"by_detector":{"doi_compliance":{"advisory":1,"critical":0,"informational":0,"total":1}},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Agentic retrieval improves multi-hop question answering by giving language models autonomy to iteratively gather evidence. Recent work augments these systems with knowledge graphs for structured traversal, but this combination introduces significant cost: expensive graph construction at index time and compounding token usage at inference time. We introduce Graph Agentic Search over Propositions (GRASP), an agentic system that simultaneously optimizes for high accuracy and minimal token usage in multi-hop question answering. Rather than executing a rigid, singular query, GRASP actively coordina","authors_text":"Junjie Hu, Ramya Korlakai Vinayak, Stockton Jenkins","cross_cats":["cs.AI"],"headline":"GRASP achieves top accuracy on multi-hop QA benchmarks while using 40 to 50 percent fewer tokens through hierarchical graph search.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.MA","submitted_at":"2026-05-15T19:59:35Z","title":"GRASP: Graph Agentic Search over Propositions for Multi-hop Question Answering"},"references":{"count":39,"internal_anchors":3,"resolved_work":39,"sample":[{"cited_arxiv_id":"1904.12106","doi":"10.18653/v1/2023.findings-acl.565","is_internal_anchor":true,"ref_index":1,"title":"Understanding Dataset Design Choices for Multi-hop Reasoning","work_id":"38d92463-996e-4540-a8ab-f492c5df8328","year":1904},{"cited_arxiv_id":"2307.03172","doi":"10.1162/tacl","is_internal_anchor":true,"ref_index":2,"title":"Lost in the Middle: How Language Models Use Long Contexts","work_id":"37c05e13-4a24-44f8-a1c4-da1bbe7223aa","year":2025},{"cited_arxiv_id":"2212.10509","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Interleaving Retrieval with Chain-of-Thought Reasoning for Knowledge-Intensive Multi-Step Questions","work_id":"2f30a088-53fc-4701-bbc0-c67bb54bf883","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Write arational plan--- a 1--3 sentence outline of the reasoning chain and key facts needed","work_id":"54b77551-b15e-455d-ba04-fcdb2432ed62","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Each will be answered by a research agent searching the knowledge graph","work_id":"e384dd1b-eb5f-4bcd-9a9d-2f1d124371eb","year":null}],"snapshot_sha256":"12bf867daac9581fad644330cad9938347b0d64bc44a26f82892737663352eed"},"source":{"id":"2605.16598","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T20:44:27.748889Z","id":"1b53e57b-b0d4-4220-87dd-c14587741ab3","model_set":{"reader":"grok-4.3"},"one_line_summary":"GRASP introduces a hierarchical graph-based agentic retrieval method that achieves top accuracy on MuSiQue, 2WikiMultihopQA, and HotpotQA while using 30-50% fewer tokens than strong baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"GRASP achieves top accuracy on multi-hop QA benchmarks while using 40 to 50 percent fewer tokens through hierarchical graph search.","strongest_claim":"GRASP achieves the highest QA accuracy in the open retrieval setting on MuSiQue and 2Wiki while using 40-50 percent fewer tokens than IRCoT+HippoRAG2, and leads on EM and F1 across all three datasets in the LongBench setting while using 30 percent fewer tokens than the next most accurate method.","weakest_assumption":"The paper assumes that the novel three-layer hierarchical graph of entities, propositions, and passages can be built and traversed at index and inference time without introducing prohibitive construction costs or retrieval noise that would erase the reported token savings."}},"verdict_id":"1b53e57b-b0d4-4220-87dd-c14587741ab3"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:76a2adbfde54fc33a12285b77c20f1d8131ee36d944d5df3db653ae0845c4406","target":"record","created_at":"2026-05-20T00:02:31Z","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":"1831d9b3d02a69af3812db20e6f0c03792bd11c7b3ee7a1f535ff41598727ef7","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.MA","submitted_at":"2026-05-15T19:59:35Z","title_canon_sha256":"6a675f605aec9879c6309d02a584fcd235c7ce793ea5105999ba43284abfee11"},"schema_version":"1.0","source":{"id":"2605.16598","kind":"arxiv","version":1}},"canonical_sha256":"1d81c3bddde99e22053c2997ed4c780dd66ad77d2a59c6749a871c9dee74d00d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1d81c3bddde99e22053c2997ed4c780dd66ad77d2a59c6749a871c9dee74d00d","first_computed_at":"2026-05-20T00:02:31.833520Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:02:31.833520Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"XwsdSXB8FqmHpOrg5XONT3QA911GZngWQo4wY5GRF3sggQr3PcAocB/NBvk5oIvpHzSRwnY082kiSoQtxohICw==","signature_status":"signed_v1","signed_at":"2026-05-20T00:02:31.834475Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.16598","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:906c5292d76e8258c585988def40825b6efed503a377ced55d065704e40ba8ed","sha256:76a2adbfde54fc33a12285b77c20f1d8131ee36d944d5df3db653ae0845c4406","sha256:37bbf203b0eaf4625c2bcc52c4a4dc1eba8201ba2ac64cb8749f8f06f5893dbf"],"state_sha256":"de5c1526451f8991eb7a6a07f83ba10dae4f1b59c60943fb23e386aca1cd020c"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"rK3gCR0nWOfi0hisJWh7QwHHTXRlta0A8pfUl9lYZz4f/kHk4X/rNB5u0NYdnDWgrq9x9DFU1MK+UGoZrMkLAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-21T12:30:21.919031Z","bundle_sha256":"a6cda505fb1debba120c3db0d6d8c8a0d32e81b23a6040f248d963ca9c15694b"}}