{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:CMH4WK7DQDNDZU666KI5JQFPNK","short_pith_number":"pith:CMH4WK7D","canonical_record":{"source":{"id":"2605.10503","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T12:59:07Z","cross_cats_sorted":[],"title_canon_sha256":"072f23611c86d5450add866cde9a5085f2f6bd5dccc7c2851b6b91d416beb1ec","abstract_canon_sha256":"a09930773b73552da0c5a470c6cac5a3f8ed2c571ab2abe06fea8c8b53bd8a2b"},"schema_version":"1.0"},"canonical_sha256":"130fcb2be380da3cd3def291d4c0af6ab54a4b29b1e68b2d8a3f92b15e4f2bd8","source":{"kind":"arxiv","id":"2605.10503","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.10503","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"arxiv_version","alias_value":"2605.10503v3","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10503","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"pith_short_12","alias_value":"CMH4WK7DQDND","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"pith_short_16","alias_value":"CMH4WK7DQDNDZU66","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"pith_short_8","alias_value":"CMH4WK7D","created_at":"2026-05-20T00:05:46Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:CMH4WK7DQDNDZU666KI5JQFPNK","target":"record","payload":{"canonical_record":{"source":{"id":"2605.10503","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T12:59:07Z","cross_cats_sorted":[],"title_canon_sha256":"072f23611c86d5450add866cde9a5085f2f6bd5dccc7c2851b6b91d416beb1ec","abstract_canon_sha256":"a09930773b73552da0c5a470c6cac5a3f8ed2c571ab2abe06fea8c8b53bd8a2b"},"schema_version":"1.0"},"canonical_sha256":"130fcb2be380da3cd3def291d4c0af6ab54a4b29b1e68b2d8a3f92b15e4f2bd8","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:05:46.915800Z","signature_b64":"aXylTgafjUC1ENjd4q0mOGeZtZ6re7RNS3jpLh124wS/n6T1FjN5+CboTiTQQCzf7eMVoSK+z5xUoqiMTSKZCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"130fcb2be380da3cd3def291d4c0af6ab54a4b29b1e68b2d8a3f92b15e4f2bd8","last_reissued_at":"2026-05-20T00:05:46.915258Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:05:46.915258Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.10503","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":"2026-05-20T00:05:46Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"I4TgN+V6yzsYqBzgKYQKFoONNgL8MV9tYwaL/WfPqth/s2/MGHZA6mlWE6+cT79wtvBdyaFcKwXk6xU3Jl0JBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T16:35:03.062815Z"},"content_sha256":"5a80f81d660082c1bf00a08ae32b515160af3bd801038c797137893e6d130efa","schema_version":"1.0","event_id":"sha256:5a80f81d660082c1bf00a08ae32b515160af3bd801038c797137893e6d130efa"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:CMH4WK7DQDNDZU666KI5JQFPNK","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SLASH the Sink: Sharpening Structural Attention Inside LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LLMs internally rebuild graph topologies in attention maps, but attention sinks dilute the signal which a simple redistribution can sharpen for better graph performance.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bin Lu, Chenghu Zhou, Meng Jin, Xinbing Wang, Yiming Liu","submitted_at":"2026-05-11T12:59:07Z","abstract_excerpt":"Large Language Models (LLMs) show remarkable semantic understanding but often struggle with structural understanding when processing graph topologies in a serialized format. Existing solutions rely on training external graph-based adapters or fine-tuning, which incur high costs and lost generalizability. In this work, we investigate the internal mechanisms of LLMs and present a critical finding: LLMs spontaneously reconstruct the graph's topology internally, evidenced by a distinct \"sawtooth\" pattern in their attention maps that structurally aligns with the \"token-level adjacency matrix\". Howe"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"LLMs spontaneously reconstruct the graph's topology internally, evidenced by a distinct 'sawtooth' pattern in their attention maps that structurally aligns with the 'token-level adjacency matrix'. This intrinsic structural understanding is diluted by the attention sink, formalized as a representation bottleneck from conflict between anisotropic bias and topology-aware local aggregation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the observed sawtooth pattern is caused by genuine internal topology reconstruction rather than an artifact of tokenization or attention mechanics, and that redistributing attention to amplify it will not degrade performance on non-graph language tasks.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SLASH redistributes attention in LLMs to amplify their spontaneous internal reconstruction of graph topologies, yielding gains on graph and molecular tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LLMs internally rebuild graph topologies in attention maps, but attention sinks dilute the signal which a simple redistribution can sharpen for better graph performance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8d3a58224345aa0f89a03fea9ec499adda8b5abddbea9dafc1910f21c35ffbdf"},"source":{"id":"2605.10503","kind":"arxiv","version":3},"verdict":{"id":"a7e9797a-563c-4c38-a9c0-6aefae7b30d6","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T03:19:49.920445Z","strongest_claim":"LLMs spontaneously reconstruct the graph's topology internally, evidenced by a distinct 'sawtooth' pattern in their attention maps that structurally aligns with the 'token-level adjacency matrix'. This intrinsic structural understanding is diluted by the attention sink, formalized as a representation bottleneck from conflict between anisotropic bias and topology-aware local aggregation.","one_line_summary":"SLASH redistributes attention in LLMs to amplify their spontaneous internal reconstruction of graph topologies, yielding gains on graph and molecular tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the observed sawtooth pattern is caused by genuine internal topology reconstruction rather than an artifact of tokenization or attention mechanics, and that redistributing attention to amplify it will not degrade performance on non-graph language tasks.","pith_extraction_headline":"LLMs internally rebuild graph topologies in attention maps, but attention sinks dilute the signal which a simple redistribution can sharpen for better graph performance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10503/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T14:42:40.497140Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T11:01:17.962650Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:14:10.525018Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"eb021d2b44bea993024f69cce3677bc9a0fc8c3166c48b1b7b95ca37ee68de3d"},"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"},"verdict_id":"a7e9797a-563c-4c38-a9c0-6aefae7b30d6"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:05:46Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"OFkjhX/Vtc19ag1XxAx/rK+vLdU6M2W8uzzlB2ER5PeS9WXHCo0znC7D7mae4sdU6+tiQJAYYL7Cefl/YTUxDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T16:35:03.063353Z"},"content_sha256":"d7ea44191e36fe112a30c820cdb094552d1663d02ac59c27f6aafbbfffe2fd71","schema_version":"1.0","event_id":"sha256:d7ea44191e36fe112a30c820cdb094552d1663d02ac59c27f6aafbbfffe2fd71"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/CMH4WK7DQDNDZU666KI5JQFPNK/bundle.json","state_url":"https://pith.science/pith/CMH4WK7DQDNDZU666KI5JQFPNK/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/CMH4WK7DQDNDZU666KI5JQFPNK/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":"2026-05-28T16:35:03Z","links":{"resolver":"https://pith.science/pith/CMH4WK7DQDNDZU666KI5JQFPNK","bundle":"https://pith.science/pith/CMH4WK7DQDNDZU666KI5JQFPNK/bundle.json","state":"https://pith.science/pith/CMH4WK7DQDNDZU666KI5JQFPNK/state.json","well_known_bundle":"https://pith.science/.well-known/pith/CMH4WK7DQDNDZU666KI5JQFPNK/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:CMH4WK7DQDNDZU666KI5JQFPNK","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":"a09930773b73552da0c5a470c6cac5a3f8ed2c571ab2abe06fea8c8b53bd8a2b","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T12:59:07Z","title_canon_sha256":"072f23611c86d5450add866cde9a5085f2f6bd5dccc7c2851b6b91d416beb1ec"},"schema_version":"1.0","source":{"id":"2605.10503","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.10503","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"arxiv_version","alias_value":"2605.10503v3","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10503","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"pith_short_12","alias_value":"CMH4WK7DQDND","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"pith_short_16","alias_value":"CMH4WK7DQDNDZU66","created_at":"2026-05-20T00:05:46Z"},{"alias_kind":"pith_short_8","alias_value":"CMH4WK7D","created_at":"2026-05-20T00:05:46Z"}],"graph_snapshots":[{"event_id":"sha256:d7ea44191e36fe112a30c820cdb094552d1663d02ac59c27f6aafbbfffe2fd71","target":"graph","created_at":"2026-05-20T00:05:46Z","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":"LLMs spontaneously reconstruct the graph's topology internally, evidenced by a distinct 'sawtooth' pattern in their attention maps that structurally aligns with the 'token-level adjacency matrix'. This intrinsic structural understanding is diluted by the attention sink, formalized as a representation bottleneck from conflict between anisotropic bias and topology-aware local aggregation."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the observed sawtooth pattern is caused by genuine internal topology reconstruction rather than an artifact of tokenization or attention mechanics, and that redistributing attention to amplify it will not degrade performance on non-graph language tasks."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SLASH redistributes attention in LLMs to amplify their spontaneous internal reconstruction of graph topologies, yielding gains on graph and molecular tasks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LLMs internally rebuild graph topologies in attention maps, but attention sinks dilute the signal which a simple redistribution can sharpen for better graph performance."}],"snapshot_sha256":"8d3a58224345aa0f89a03fea9ec499adda8b5abddbea9dafc1910f21c35ffbdf"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T14:42:40.497140Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T11:01:17.962650Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T09:14:10.525018Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.10503/integrity.json","findings":[],"snapshot_sha256":"eb021d2b44bea993024f69cce3677bc9a0fc8c3166c48b1b7b95ca37ee68de3d","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large Language Models (LLMs) show remarkable semantic understanding but often struggle with structural understanding when processing graph topologies in a serialized format. Existing solutions rely on training external graph-based adapters or fine-tuning, which incur high costs and lost generalizability. In this work, we investigate the internal mechanisms of LLMs and present a critical finding: LLMs spontaneously reconstruct the graph's topology internally, evidenced by a distinct \"sawtooth\" pattern in their attention maps that structurally aligns with the \"token-level adjacency matrix\". Howe","authors_text":"Bin Lu, Chenghu Zhou, Meng Jin, Xinbing Wang, Yiming Liu","cross_cats":[],"headline":"LLMs internally rebuild graph topologies in attention maps, but attention sinks dilute the signal which a simple redistribution can sharpen for better graph performance.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T12:59:07Z","title":"SLASH the Sink: Sharpening Structural Attention Inside LLMs"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.10503","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-13T03:19:49.920445Z","id":"a7e9797a-563c-4c38-a9c0-6aefae7b30d6","model_set":{"reader":"grok-4.3"},"one_line_summary":"SLASH redistributes attention in LLMs to amplify their spontaneous internal reconstruction of graph topologies, yielding gains on graph and molecular tasks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"LLMs internally rebuild graph topologies in attention maps, but attention sinks dilute the signal which a simple redistribution can sharpen for better graph performance.","strongest_claim":"LLMs spontaneously reconstruct the graph's topology internally, evidenced by a distinct 'sawtooth' pattern in their attention maps that structurally aligns with the 'token-level adjacency matrix'. This intrinsic structural understanding is diluted by the attention sink, formalized as a representation bottleneck from conflict between anisotropic bias and topology-aware local aggregation.","weakest_assumption":"That the observed sawtooth pattern is caused by genuine internal topology reconstruction rather than an artifact of tokenization or attention mechanics, and that redistributing attention to amplify it will not degrade performance on non-graph language tasks."}},"verdict_id":"a7e9797a-563c-4c38-a9c0-6aefae7b30d6"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:5a80f81d660082c1bf00a08ae32b515160af3bd801038c797137893e6d130efa","target":"record","created_at":"2026-05-20T00:05:46Z","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":"a09930773b73552da0c5a470c6cac5a3f8ed2c571ab2abe06fea8c8b53bd8a2b","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T12:59:07Z","title_canon_sha256":"072f23611c86d5450add866cde9a5085f2f6bd5dccc7c2851b6b91d416beb1ec"},"schema_version":"1.0","source":{"id":"2605.10503","kind":"arxiv","version":3}},"canonical_sha256":"130fcb2be380da3cd3def291d4c0af6ab54a4b29b1e68b2d8a3f92b15e4f2bd8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"130fcb2be380da3cd3def291d4c0af6ab54a4b29b1e68b2d8a3f92b15e4f2bd8","first_computed_at":"2026-05-20T00:05:46.915258Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:05:46.915258Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"aXylTgafjUC1ENjd4q0mOGeZtZ6re7RNS3jpLh124wS/n6T1FjN5+CboTiTQQCzf7eMVoSK+z5xUoqiMTSKZCg==","signature_status":"signed_v1","signed_at":"2026-05-20T00:05:46.915800Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.10503","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:5a80f81d660082c1bf00a08ae32b515160af3bd801038c797137893e6d130efa","sha256:d7ea44191e36fe112a30c820cdb094552d1663d02ac59c27f6aafbbfffe2fd71"],"state_sha256":"516cf063580cad4774e247db69ac6476a1f4919bf77b13c86f2188bfe88b88be"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JZvNZNX5qrQxsFVkqYRxgc3N1a5KWWRfF+5VDQ9GzZcFq8UzcZhaXDEQ3cfRgul+anueqDQCnviHC/Yt0zxhBQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T16:35:03.065807Z","bundle_sha256":"cebe6010c9a5054e3722492ca1e62d9aa3f5a99a442b7a6e100bbb28868045ee"}}