{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:XWAXIAJA7GS2YMYD4LN2SESTS4","short_pith_number":"pith:XWAXIAJA","canonical_record":{"source":{"id":"2605.09778","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-10T21:51:36Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"fc6ff1108ef5dfd97ef4cc2abc0e80563850a5f5500bace934aeb1e07e2e8df8","abstract_canon_sha256":"6f375fd16fa71a902fcd19fce891f04e222f8f189ec6149fbff25aa33fe54b85"},"schema_version":"1.0"},"canonical_sha256":"bd81740120f9a5ac3303e2dba912539701cf7285734c6f0d842d04bb3040b32d","source":{"kind":"arxiv","id":"2605.09778","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09778","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09778v2","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09778","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"pith_short_12","alias_value":"XWAXIAJA7GS2","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"pith_short_16","alias_value":"XWAXIAJA7GS2YMYD","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"pith_short_8","alias_value":"XWAXIAJA","created_at":"2026-07-09T01:20:05Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:XWAXIAJA7GS2YMYD4LN2SESTS4","target":"record","payload":{"canonical_record":{"source":{"id":"2605.09778","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-10T21:51:36Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"fc6ff1108ef5dfd97ef4cc2abc0e80563850a5f5500bace934aeb1e07e2e8df8","abstract_canon_sha256":"6f375fd16fa71a902fcd19fce891f04e222f8f189ec6149fbff25aa33fe54b85"},"schema_version":"1.0"},"canonical_sha256":"bd81740120f9a5ac3303e2dba912539701cf7285734c6f0d842d04bb3040b32d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-09T01:20:05.418344Z","signature_b64":"dRuITxt5x7aJ9XBZQ2QnUCyqWFK0js86RmmU06piS/wRkNQ5os2ZOJi/UJKi/JCYbiFvV7d1UhLHNKT1xbvWDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bd81740120f9a5ac3303e2dba912539701cf7285734c6f0d842d04bb3040b32d","last_reissued_at":"2026-07-09T01:20:05.417868Z","signature_status":"signed_v1","first_computed_at":"2026-07-09T01:20:05.417868Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.09778","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-09T01:20:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"VDq14nlIxM30MMefUC5ced33MlTMfeLDKioXOOql5dicJ/WKM0cMnu4UXQnea8CoWuaSB7PZ6mGjNkpnZQv2Cw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-10T17:09:33.806363Z"},"content_sha256":"667ad7464736e96fb7f2ea32501fb284a9685e1569d185036baaa8589caac5a3","schema_version":"1.0","event_id":"sha256:667ad7464736e96fb7f2ea32501fb284a9685e1569d185036baaa8589caac5a3"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:XWAXIAJA7GS2YMYD4LN2SESTS4","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Nectar: Neural Estimation of Cached-Token Attention via Regression","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"A compact neural network can approximate full attention over long cached contexts, replacing the linear scan with a fixed-cost forward pass while keeping generated text semantically equivalent.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Jo\\~ao Monteiro, Marco Cuturi, Michal Klein, Pierre Ablin","submitted_at":"2026-05-10T21:51:36Z","abstract_excerpt":"Evaluating softmax attention over a fixed long context requires reading every cached key-value pair for each new query token. For a given context (a book, a manual, a legal corpus) the attention output is a deterministic function of the query. We propose Nectar, which fits a compact neural network to this function for queries drawn from a task-relevant distribution. Nectar fits two networks per layer and KV-head: a target network that predicts the attention output and a score network that predicts the log-normalizer. The pair plugs into the standard masked self-attention at inference time, rep"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The approximation error tracks the next-token accuracy gap to full attention, and allocating capacity non-uniformly across layers reduces that gap; text generations of a model equipped with a Nectar module match in semantic content those obtained by giving the same model access to the full cache.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"A compact neural network can accurately approximate the deterministic attention function for queries drawn from a task-relevant distribution, with the approximation quality generalizing to the actual generation distribution at inference.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Nectar fits small per-layer per-head neural networks via regression to predict attention outputs and normalizers, enabling constant-time inference independent of context length while preserving semantic generation quality.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A compact neural network can approximate full attention over long cached contexts, replacing the linear scan with a fixed-cost forward pass while keeping generated text semantically equivalent.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6b606702250ab063cf540e79c563eec5156dc139d2c11530e63abafea74333dc"},"source":{"id":"2605.09778","kind":"arxiv","version":2},"verdict":{"id":"749fa008-dc69-4bfc-9a7d-ccb1452968f0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T02:35:31.779639Z","strongest_claim":"The approximation error tracks the next-token accuracy gap to full attention, and allocating capacity non-uniformly across layers reduces that gap; text generations of a model equipped with a Nectar module match in semantic content those obtained by giving the same model access to the full cache.","one_line_summary":"Nectar fits small per-layer per-head neural networks via regression to predict attention outputs and normalizers, enabling constant-time inference independent of context length while preserving semantic generation quality.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"A compact neural network can accurately approximate the deterministic attention function for queries drawn from a task-relevant distribution, with the approximation quality generalizing to the actual generation distribution at inference.","pith_extraction_headline":"A compact neural network can approximate full attention over long cached contexts, replacing the linear scan with a fixed-cost forward pass while keeping generated text semantically equivalent."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09778/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T07:02:01.408898Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T16:36:31.073748Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:17.903916Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:58:15.513600Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"a3338e0f2784292cff299393da925d70795fdbe5816f17a12abadb541213dc19"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"16435cca009afa9a0d949ceb1ddac2e9e73d97bcdc0cbce8e08436dae3ca06a5"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"749fa008-dc69-4bfc-9a7d-ccb1452968f0"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-09T01:20:05Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"eJABRAioSnRRi2CGcfU5T9S7Jsi90emTzUeZXMG9ZZZwVMJKYuc3hcAq2Gr3/0KB6axS0+Dxkm2iJU7X46V+Aw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-10T17:09:33.807091Z"},"content_sha256":"b4b3bf759edbe8bec4f95643f579993d133080b6b614621d794a7ce2b973a576","schema_version":"1.0","event_id":"sha256:b4b3bf759edbe8bec4f95643f579993d133080b6b614621d794a7ce2b973a576"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/XWAXIAJA7GS2YMYD4LN2SESTS4/bundle.json","state_url":"https://pith.science/pith/XWAXIAJA7GS2YMYD4LN2SESTS4/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/XWAXIAJA7GS2YMYD4LN2SESTS4/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":"2026-08-10T17:09:33Z","links":{"resolver":"https://pith.science/pith/XWAXIAJA7GS2YMYD4LN2SESTS4","bundle":"https://pith.science/pith/XWAXIAJA7GS2YMYD4LN2SESTS4/bundle.json","state":"https://pith.science/pith/XWAXIAJA7GS2YMYD4LN2SESTS4/state.json","well_known_bundle":"https://pith.science/.well-known/pith/XWAXIAJA7GS2YMYD4LN2SESTS4/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:XWAXIAJA7GS2YMYD4LN2SESTS4","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":"6f375fd16fa71a902fcd19fce891f04e222f8f189ec6149fbff25aa33fe54b85","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-10T21:51:36Z","title_canon_sha256":"fc6ff1108ef5dfd97ef4cc2abc0e80563850a5f5500bace934aeb1e07e2e8df8"},"schema_version":"1.0","source":{"id":"2605.09778","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09778","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09778v2","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09778","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"pith_short_12","alias_value":"XWAXIAJA7GS2","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"pith_short_16","alias_value":"XWAXIAJA7GS2YMYD","created_at":"2026-07-09T01:20:05Z"},{"alias_kind":"pith_short_8","alias_value":"XWAXIAJA","created_at":"2026-07-09T01:20:05Z"}],"graph_snapshots":[{"event_id":"sha256:b4b3bf759edbe8bec4f95643f579993d133080b6b614621d794a7ce2b973a576","target":"graph","created_at":"2026-07-09T01:20:05Z","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":"The approximation error tracks the next-token accuracy gap to full attention, and allocating capacity non-uniformly across layers reduces that gap; text generations of a model equipped with a Nectar module match in semantic content those obtained by giving the same model access to the full cache."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"A compact neural network can accurately approximate the deterministic attention function for queries drawn from a task-relevant distribution, with the approximation quality generalizing to the actual generation distribution at inference."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Nectar fits small per-layer per-head neural networks via regression to predict attention outputs and normalizers, enabling constant-time inference independent of context length while preserving semantic generation quality."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A compact neural network can approximate full attention over long cached contexts, replacing the linear scan with a fixed-cost forward pass while keeping generated text semantically equivalent."}],"snapshot_sha256":"6b606702250ab063cf540e79c563eec5156dc139d2c11530e63abafea74333dc"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"16435cca009afa9a0d949ceb1ddac2e9e73d97bcdc0cbce8e08436dae3ca06a5"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T07:02:01.408898Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T16:36:31.073748Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:17.903916Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T09:58:15.513600Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.09778/integrity.json","findings":[],"snapshot_sha256":"a3338e0f2784292cff299393da925d70795fdbe5816f17a12abadb541213dc19","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Evaluating softmax attention over a fixed long context requires reading every cached key-value pair for each new query token. For a given context (a book, a manual, a legal corpus) the attention output is a deterministic function of the query. We propose Nectar, which fits a compact neural network to this function for queries drawn from a task-relevant distribution. Nectar fits two networks per layer and KV-head: a target network that predicts the attention output and a score network that predicts the log-normalizer. The pair plugs into the standard masked self-attention at inference time, rep","authors_text":"Jo\\~ao Monteiro, Marco Cuturi, Michal Klein, Pierre Ablin","cross_cats":["cs.CL"],"headline":"A compact neural network can approximate full attention over long cached contexts, replacing the linear scan with a fixed-cost forward pass while keeping generated text semantically equivalent.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-10T21:51:36Z","title":"Nectar: Neural Estimation of Cached-Token Attention via Regression"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.09778","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-12T02:35:31.779639Z","id":"749fa008-dc69-4bfc-9a7d-ccb1452968f0","model_set":{"reader":"grok-4.3"},"one_line_summary":"Nectar fits small per-layer per-head neural networks via regression to predict attention outputs and normalizers, enabling constant-time inference independent of context length while preserving semantic generation quality.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A compact neural network can approximate full attention over long cached contexts, replacing the linear scan with a fixed-cost forward pass while keeping generated text semantically equivalent.","strongest_claim":"The approximation error tracks the next-token accuracy gap to full attention, and allocating capacity non-uniformly across layers reduces that gap; text generations of a model equipped with a Nectar module match in semantic content those obtained by giving the same model access to the full cache.","weakest_assumption":"A compact neural network can accurately approximate the deterministic attention function for queries drawn from a task-relevant distribution, with the approximation quality generalizing to the actual generation distribution at inference."}},"verdict_id":"749fa008-dc69-4bfc-9a7d-ccb1452968f0"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:667ad7464736e96fb7f2ea32501fb284a9685e1569d185036baaa8589caac5a3","target":"record","created_at":"2026-07-09T01:20:05Z","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":"6f375fd16fa71a902fcd19fce891f04e222f8f189ec6149fbff25aa33fe54b85","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-10T21:51:36Z","title_canon_sha256":"fc6ff1108ef5dfd97ef4cc2abc0e80563850a5f5500bace934aeb1e07e2e8df8"},"schema_version":"1.0","source":{"id":"2605.09778","kind":"arxiv","version":2}},"canonical_sha256":"bd81740120f9a5ac3303e2dba912539701cf7285734c6f0d842d04bb3040b32d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"bd81740120f9a5ac3303e2dba912539701cf7285734c6f0d842d04bb3040b32d","first_computed_at":"2026-07-09T01:20:05.417868Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-09T01:20:05.417868Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"dRuITxt5x7aJ9XBZQ2QnUCyqWFK0js86RmmU06piS/wRkNQ5os2ZOJi/UJKi/JCYbiFvV7d1UhLHNKT1xbvWDw==","signature_status":"signed_v1","signed_at":"2026-07-09T01:20:05.418344Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.09778","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:667ad7464736e96fb7f2ea32501fb284a9685e1569d185036baaa8589caac5a3","sha256:b4b3bf759edbe8bec4f95643f579993d133080b6b614621d794a7ce2b973a576"],"state_sha256":"e1a0873847106842d5dee279420617c0ddff15e9ce889a9127117e8cdc00447a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"8Wfws7qnaTMzT9uSPEXS+MB0Vgx1SITuG3WDcnc9DKrru1PCy/k4KZwgUg+8O+KdVmlWCD2P/qzmG+Gg1oZIBg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-10T17:09:33.812694Z","bundle_sha256":"e1300926a210b197cc017ee0e8b350d4ee3f1a5ff953461856f944bb9e30c5a8"}}