{"as_of":"2026-08-07T22:05:00Z","caps":{"database_statements":6,"inbound":100,"outbound":100},"context_digest":"sha256:67769b1dc323d45441f57c8bddd83f647c44fd17f9366f56cb1c5aee7ad4c5a9","coverage":[{"denominator":20,"lane":"reference_resolution","note":"Typed states for the displayed outbound observations.","records_observed":20,"source":"paper_references, paper_reference_links","source_observed_at":"2026-05-21T01:37:33.760985Z","state":"measured"},{"denominator":20,"lane":"standing_notices","note":"One-hop event checks from named stored sources.","records_observed":20,"source":"scholarly_work_events, retraction_status_cache","source_observed_at":"2026-08-07T06:34:17.273281+00:00","state":"measured"},{"denominator":0,"lane":"inbound_itemization","note":"Pith citing papers itemized under the disclosed page cap.","records_observed":0,"source":"paper_references, paper_reference_links","source_observed_at":null,"state":"measured"},{"denominator":1,"lane":"external_citation_measurements","note":"A source-named dated measurement, never combined with another source.","records_observed":0,"source":"cited_works","source_observed_at":null,"state":"measured"}],"external_citation_measurements":[],"inbound":[],"links":{"evidence":"/evidence","html":"/paper/2604.10027/citation-record","integrity":"/paper/2604.10027/integrity","json":"/paper/2604.10027/citation-record.json","paper":"/paper/2604.10027"},"outbound":[{"citation":{"cited_paper":{"arxiv_id":"2502.13923","last_updated":"2025-02-19T18:00:14Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2025-02-19T18:00:14Z","title":"Qwen2.5-VL Technical Report","version":1},"cited_work":{"arxiv_id":"2502.13923","doi":"10.48550/arxiv.2502.13923","metadata_source":"pith","pith_arxiv_id":"2502.13923","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"Qwen2.5-VL Technical Report","venue":"cs.CV","work_id":"69dffacb-bfe8-442d-be86-48624c60426f","year":2025},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":1,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2502.13923","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:0bd8ba39eb8c9cd9b8be98c1378fbd7305f80f39fc8a9333925bc4da72983751","observation_id":"f1c161c6-de5a-49bc-8b1f-1a36fb3b5d3a","resolution":{"observed_at":"2026-05-21T01:39:22.605100Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-07-12T05:19:13.082554+00:00","source":"crossref_status_cache"},{"observed_at":"2026-07-12T05:19:13.082554+00:00","source":"openalex_status_cache"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-06-05T21:23:00.469572Z","title":"Graph of thoughts: Solving elaborate problems with large language models","venue":null,"work_id":"8d8d2631-891c-45a4-a337-750fb387a09f","year":2026},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":2,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:bdfbdcc52b087de9b5d33dce48da99afaf17c5d9818d9a7c8edc4848534a0147","observation_id":"0c11eee7-bf6d-434b-b005-e5e55f5c01d8","resolution":{"observed_at":"2026-05-21T01:39:23.344392Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2211.12588","last_updated":"2023-10-23T01:27:38Z","snapshot_observed_at":"2026-08-02T13:06:11.850456Z","submitted_at":"2022-11-22T21:06:00Z","title":"Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks","version":4},"cited_work":{"arxiv_id":"2211.12588","doi":"10.48550/arxiv.2211.12588","metadata_source":"pith","pith_arxiv_id":"2211.12588","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks","venue":"cs.CL","work_id":"618aa44c-a6c6-425c-abce-8aa8aa842921","year":2022},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":3,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2211.12588","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:8d099dbb7c0cb65825ab4289d83aadeaf6e6411c4143130d9fbac789cc78b0d4","observation_id":"1dd49a5b-8381-4fad-97c8-96a56df28826","resolution":{"observed_at":"2026-05-21T01:39:22.636753Z","resolver_source":"local_arxiv","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2407.21783","last_updated":"2024-11-23T23:27:33Z","snapshot_observed_at":"2026-07-06T18:55:11.576666Z","submitted_at":"2024-07-31T17:54:27Z","title":"The Llama 3 Herd of Models","version":3},"cited_work":{"arxiv_id":"2407.21783","doi":"10.1016/s0749-0720(15","metadata_source":"pith","pith_arxiv_id":"2407.21783","snapshot_observed_at":"2026-07-11T11:50:26.030339Z","title":"The Llama 3 Herd of Models","venue":"cs.AI","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","year":2024},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":4,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2407.21783","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:47e06b0567ad85930168fc6eb4984b97e33b2909671929ddf9928c93383b40e8","observation_id":"79866a24-79af-42f1-bace-aca02dfdc792","resolution":{"observed_at":"2026-05-21T01:39:22.611754Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":"2505.11739","doi":null,"metadata_source":"arxiv_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-06-05T21:23:00.469572Z","title":"Zerotuning: Unlocking the initial to- ken’s power to enhance large language models without training.arXiv preprint arXiv:2505.11739","venue":null,"work_id":"b8c1adad-d817-45f5-a565-2ac8cd955b86","year":2025},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":5,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:2d15aaf316e2db507ca3ca79d40648d218127596d80e3fe5f440f96469a4f8a5","observation_id":"a026da1f-7c41-4712-a64d-fcf2c42f77b9","resolution":{"observed_at":"2026-05-21T01:39:22.644104Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2404.06395","last_updated":"2024-06-03T08:54:38Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2024-04-09T15:36:50Z","title":"MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies","version":3},"cited_work":{"arxiv_id":"2404.06395","doi":"10.48550/arxiv.2404.06395","metadata_source":"pith","pith_arxiv_id":"2404.06395","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies","venue":"cs.CL","work_id":"f20a4304-bd39-414a-923b-d18322e29258","year":2024},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":6,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2404.06395","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:f870a0af0698f554586b2cfed8057f949e82c1ce7b2a26b75d114618c838969e","observation_id":"7707c984-aa25-41fd-8351-3eca6283c05e","resolution":{"observed_at":"2026-05-21T01:39:22.670454Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-06-05T21:23:00.469572Z","title":"Llm agents for smart city management: Enhancing decision support through multi-agent ai sys- tems.Smart Cities (2624-6511), 8(1)","venue":null,"work_id":"13457acf-fd9d-4b9f-b589-179c768f94cb","year":2026},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:f08dc4bd33196ea15ab134f39964dfaaf1bd8d5b062c2eb68312fa3353d320cd","observation_id":"4720b6ce-2385-4b4a-bcb6-d91f9f25b9b8","resolution":{"observed_at":"2026-05-21T01:39:23.348361Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.04717","last_updated":"2026-04-20T02:55:07Z","snapshot_observed_at":"2026-07-06T21:05:11.303685Z","submitted_at":"2025-04-07T04:00:08Z","title":"Beyond Single-Turn: A Survey on Multi-Turn Interactions with Large Language Models","version":6},"cited_work":{"arxiv_id":"2504.04717","doi":null,"metadata_source":"pith","pith_arxiv_id":"2504.04717","snapshot_observed_at":"2026-07-04T01:19:21.999006Z","title":"Beyond Single-Turn: A Survey on Multi-Turn Interactions with Large Language Models","venue":"cs.CL","work_id":"9b4fa784-bc79-4e0d-bf71-1b63249eb302","year":2025},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":8,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2504.04717","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:7d03248bd304851eb14d2df9f8193c1f6f49c44f2b35447283f884c146a2fc59","observation_id":"8b6e2a00-3baf-486b-8939-2843537ecb44","resolution":{"observed_at":"2026-05-21T01:39:22.650614Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2410.21333","last_updated":"2025-06-13T19:10:02Z","snapshot_observed_at":"2026-08-07T21:27:29.354913Z","submitted_at":"2024-10-27T18:30:41Z","title":"Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse","version":4},"cited_work":{"arxiv_id":"2410.21333","doi":"10.48550/arxiv.2410.21333","metadata_source":"arxiv_reference","pith_arxiv_id":"2410.21333","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"arXiv preprint arXiv:2410.21333 , year=","venue":"arXiv (Cornell University)","work_id":"6b08eaa6-9403-486c-9338-3b48afcce6c9","year":2024},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":9,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2410.21333","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:19781e0cfbde1fc7881213657cd46b1a5fd81eb5a18782629d48819d15aa432a","observation_id":"0de1dc43-2832-4448-ae20-42dfdc9fc68b","resolution":{"observed_at":"2026-05-21T01:39:22.676699Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2310.01427","last_updated":"2023-09-28T05:19:06Z","snapshot_observed_at":"2026-08-03T11:42:20.866424Z","submitted_at":"2023-09-28T05:19:06Z","title":"Attention Sorting Combats Recency Bias In Long Context Language Models","version":1},"cited_work":{"arxiv_id":"2310.01427","doi":"10.48550/arxiv.2310.01427","metadata_source":"arxiv_reference","pith_arxiv_id":"2310.01427","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"Attention sorting combats recency bias in long context language models","venue":"arXiv (Cornell University)","work_id":"0c7cbc17-7ac1-4769-84a0-688c191de159","year":2023},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":10,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2310.01427","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:bd4754da90a18973455672bad9f21d4b85dda975346458f1802c0153157e7662","observation_id":"197824ca-fbb1-4569-b467-120afb0b3932","resolution":{"observed_at":"2026-05-21T01:39:22.657396Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2412.15115","last_updated":"2025-01-03T02:18:21Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2024-12-19T17:56:09Z","title":"Qwen2.5 Technical Report","version":2},"cited_work":{"arxiv_id":"2412.15115","doi":"10.1145/3581783.3612503","metadata_source":"pith","pith_arxiv_id":"2412.15115","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"Qwen2.5 Technical Report","venue":"cs.CL","work_id":"d8432992-4980-4a81-85c7-9fa2c2b87f85","year":2024},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":11,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2412.15115","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:eb7f56a17aa386a2280557a1fd1cce1410b284e07d46a66348cc5066eef41532","observation_id":"05912b6f-9f0f-4f25-96a6-69b943efe0b9","resolution":{"observed_at":"2026-05-21T01:39:22.598067Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2309.05922","last_updated":"2023-09-12T02:34:06Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2023-09-12T02:34:06Z","title":"A Survey of Hallucination in Large Foundation Models","version":1},"cited_work":{"arxiv_id":"2309.05922","doi":null,"metadata_source":"pith","pith_arxiv_id":"2309.05922","snapshot_observed_at":"2026-07-04T19:50:11.151420Z","title":"A Survey of Hallucination in Large Foundation Models","venue":"cs.AI","work_id":"1b84f221-37fa-403a-9bf4-1741910454bf","year":2023},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":12,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2309.05922","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:ab983aa8ecf764aa2b119608046fbc69092c263eb075105afd1c92cb9b84a884","observation_id":"41e4d16e-dd7e-4db1-a225-f658ee2d61cf","resolution":{"observed_at":"2026-05-21T01:39:22.663095Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2508.02546","last_updated":"2025-08-04T15:59:15Z","snapshot_observed_at":"2026-08-06T05:02:16.829699Z","submitted_at":"2025-08-04T15:59:15Z","title":"What are you sinking? A geometric approach on attention sink","version":1},"cited_work":{"arxiv_id":"2508.02546","doi":null,"metadata_source":"arxiv_reference","pith_arxiv_id":"2508.02546","snapshot_observed_at":"2026-07-01T22:36:17.024576Z","title":"What are you sinking? a geometric approach on attention sink","venue":null,"work_id":"e02e0047-d770-4802-8bc7-7bcd7ab066b7","year":2015},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":13,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2508.02546","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:77641ee17ee041057c3007602a967072fb0c9f74f716a5260097d28f3d780b87","observation_id":"a199941f-9665-4718-9f86-0167199ca53e","resolution":{"observed_at":"2026-05-21T01:39:22.630672Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2503.19786","last_updated":"2025-03-25T15:52:34Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2025-03-25T15:52:34Z","title":"Gemma 3 Technical Report","version":1},"cited_work":{"arxiv_id":"2503.19786","doi":"10.1007/978-3-540-48085-3_36","metadata_source":"pith","pith_arxiv_id":"2503.19786","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"Gemma 3 Technical Report","venue":"cs.CL","work_id":"f93e08bf-9e96-409b-8ac6-b8385fd17fd7","year":2025},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":14,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2503.19786","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:fb3541e6b0655b7260695b236b170b1abea48335faad2bf3539eb802021d4a74","observation_id":"0d4dc6d5-4f5e-4e8e-9a9a-8a1abfd04d3a","resolution":{"observed_at":"2026-05-21T01:39:22.584536Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2308.10248","last_updated":"2024-10-10T13:20:13Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2023-08-20T12:21:05Z","title":"Steering Language Models With Activation Engineering","version":5},"cited_work":{"arxiv_id":"2308.10248","doi":"10.18653/v1/2024.findings-acl.611","metadata_source":"pith","pith_arxiv_id":"2308.10248","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"Steering Language Models With Activation Engineering","venue":"cs.CL","work_id":"d525fe06-5560-4e97-86fc-7a0e551f5b17","year":2023},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":15,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2308.10248","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:e69a10d7921de79acc51b4c987f8843b1de2c64191ea862d35a189aff423419c","observation_id":"41c94adc-4fe6-46cf-bff0-fd677d3c1725","resolution":{"observed_at":"2026-05-21T01:39:22.617731Z","resolver_source":"local_arxiv","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2403.05313","last_updated":"2024-03-08T13:42:19Z","snapshot_observed_at":"2026-07-06T17:41:34.856417Z","submitted_at":"2024-03-08T13:42:19Z","title":"RAT: Retrieval Augmented Thoughts Elicit Context-Aware Reasoning in Long-Horizon Generation","version":1},"cited_work":{"arxiv_id":"2403.05313","doi":"10.48550/arxiv.2403.05313","metadata_source":"arxiv_reference","pith_arxiv_id":"2403.05313","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"RAT: retrieval augmented thoughts elicit context-aware reasoning in long-horizon generation","venue":"arXiv (Cornell University)","work_id":"c2f49ce5-dd6b-4167-8f80-b91186d47917","year":2024},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":16,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"cited_paper":"/paper/2403.05313","citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:32e90bd63b6059617440eeecad9ab73e63dd026ab8131753cbf16126a8745e22","observation_id":"f377d4da-48ab-499b-bdc8-b560e90ac653","resolution":{"observed_at":"2026-05-21T01:39:22.624227Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-06-05T21:23:00.469572Z","title":"Chain-of-note: Enhancing robustness in retrieval-augmented language models","venue":null,"work_id":"0b08cb1c-2b58-46bd-ae7a-1ecbe77dcaad","year":2026},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":17,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:555339854a96654517d630138b375dde7b6d49d3d712f1736104d1c6d065ba4d","observation_id":"d2a9c266-8b47-432a-b333-ad88de2c34aa","resolution":{"observed_at":"2026-05-21T01:39:23.336498Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":"2502.01969","doi":null,"metadata_source":"arxiv_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-07-02T02:56:29.259403Z","title":"Mitigating object hallucinations in large vision-language models via attention calibration.arXiv preprint arXiv:2502.01969","venue":null,"work_id":"628d8d1f-b8ab-4656-ad68-23937c836b1d","year":2025},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":18,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:ac7bcbe1e9ca47c355c8f78aa9d665fbd775c3937bfa754c233e65be580a5ed9","observation_id":"6e8cb558-0dd3-4bf0-9b84-887acdb27540","resolution":{"observed_at":"2026-05-21T01:39:22.591730Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-06-05T21:23:00.469572Z","title":"\"\" h_ori: hidden states of the original sequence (Lori ×D h)) h_info: hidden states of the external information (L inf o ×D h)) cfg: configuration for injection rules","venue":null,"work_id":"c03af759-1933-4d2d-8fb8-ec9f3b275bce","year":2026},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":19,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:64f3db77c5fb304a78274771e0936d638f10b0dd05d1499977f1ec0d0ce1fcb5","observation_id":"fa4b5282-8ac4-4755-b849-48774024a623","resolution":{"observed_at":"2026-05-21T01:39:23.340484Z","resolver_source":"raw_fallback","status":"verified_fuzzy"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":"raw_reference","pith_arxiv_id":null,"snapshot_observed_at":"2026-06-05T21:23:00.469572Z","title":"Drift Test","venue":null,"work_id":"21f1c989-4505-42e3-832d-35e00603c4ea","year":2026},"citing_paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","version":2},"reference_index":20,"source":"pdf_text","source_observed_at":"2026-05-21T01:37:33.760985Z"},"links":{"citing_paper":"/paper/2604.10027"},"observation_digest":"sha256:60fc64241f973d6f4b37a7179ad09cb5956a47f6d7280773632f8e260e9827cd","observation_id":"2f71aaba-a0f5-4b94-83b0-db2adf401d37","resolution":{"observed_at":"2026-05-21T01:39:23.351710Z","resolver_source":"raw_fallback","status":"malformed_identifier"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"state":"measured"}}],"paper":{"arxiv_id":"2604.10027","last_updated":"2026-05-16T22:29:39Z","latest_version":2,"primary_category":"cs.CV","snapshot_observed_at":"2026-08-02T03:35:34.661841Z","submitted_at":"2026-04-11T04:49:11Z","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models"},"reference_resolution":{"displayed":20,"state_counts":{"malformed_identifier":1,"metadata_mismatch":2,"parse_uncertain":0,"unresolved":0,"verified_exact":13,"verified_fuzzy":4},"total_outbound_references":20},"refusal":"A citation records a reference. It does not transfer a finding from one paper to another.","schema":"pith.paper-citation-record.v1","standing_sources":[{"observed_at":"2026-08-07T06:34:17.273281+00:00","source":"crossref"},{"observed_at":"2026-08-07T06:34:11.927384+00:00","source":"retraction_watch"}],"thesis":"As of 7 August 2026, this Paper Citation Record lists 20 of 20 outbound references and 0 inbound Pith citation observations for arXiv:2604.10027."}