{"id":"0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc","title":"arXiv cs.AI: Schreiber and coauthors separate two kinds of relevance in Graph Attention Networks","argument":"In their paper, Ken Schreiber and coauthors argue that Layer-wise Relevance Propagation (LRP) for Graph Attention Networks should distinguish a node’s content relevance from its routing relevance. Content relevance tracks what the node supplied through the value path against a zero baseline. Routing relevance tracks how the node shaped what was passed through the attention path relative to its neighborhood; its sign indicates whether the node’s features moved the gating in the direction the prediction demanded. The authors say existing rules either discard one quantity or combine both into one number. Their single product rule uses free weights on the two factors and recovers current approaches as special cases. On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three settings. The separated maps lead the authors to hypothesize that the model gains by not reading that node. They provide reading rules for both paths and argue that the two should be interpreted separately. This distinction matters because a node can contribute through its content while also affecting the model’s routing, and a combined score can obscure those different roles. The reported example is one classification task; the abstract does not establish how often the patterns occur elsewhere.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported comparison is on a ground-truth classification task; the abstract does not establish how well the findings generalize to other tasks or domains. The supplied evidence is the paper’s abstract, not its full methods or results.","speaker":{"name":"Schreiber, Ken, Klausen, René P., Lapuschkin, Sebastian, Samek, Wojciech, Naujoks, Jonas","roleAtSource":"Article author"},"source":{"title":"Value or Attention: Investigating Layer-wise Relevance Propagation for Graph Attention Networks","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.08088","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors distinguish content relevance through the value path from routing relevance through the attention path, and argue that these should be interpreted separately.","excerpt":"the relevance reaching a node is the sum of two mechanically different quantities: its content relevance through the value path and its routing relevance through the attention path","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"},{"text":"The authors’ product rule uses free weights on the two factors and recovers existing approaches as special cases.","excerpt":"We write a single product rule with free weights on the two factors that recovers current approaches as special cases.","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"},{"text":"On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three rule settings.","excerpt":"the same node receives negative, positive and near-zero relevance under three settings of the rule.","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"},{"text":"The authors hypothesize from the separated relevance maps that the model gains by not reading the node.","excerpt":"The separated relevance maps yield a hypothesis about the model: it gains by not reading that node.","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.08088","checkedAt":"2026-11-09T10:34:06.990Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Across other ground-truth classification tasks, do the separated relevance maps likewise distinguish cases where a model gains by not reading a node?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Schreiber and coauthors separate two kinds of relevance in Graph Attention Networks","summary":"In their paper, Ken Schreiber and coauthors argue that Layer-wise Relevance Propagation (LRP) for Graph Attention Networks should distinguish a node’s content relevance from its routing relevance. Content relevance tracks what the node supplied through the value path against a zero baseline. Routing relevance tracks how the node shaped what was passed through the attention path relative to its neighborhood; its sign indicates whether the node’s features moved the gating in the direction the prediction demanded. The authors say existing rules either discard one quantity or combine both into one number. Their single product rule uses free weights on the two factors and recovers current approaches as special cases. On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three settings. The separated maps lead the authors to hypothesize that the model gains by not reading that node. They provide reading rules for both paths and argue that the two should be interpreted separately. This distinction matters because a node can contribute through its content while also affecting the model’s routing, and a combined score can obscure those different roles. The reported example is one classification task; the abstract does not establish how often the patterns occur elsewhere.","question":"Across other ground-truth classification tasks, do the separated relevance maps likewise distinguish cases where a model gains by not reading a node?","claims":[{"text":"The authors distinguish content relevance through the value path from routing relevance through the attention path, and argue that these should be interpreted separately.","excerpt":"the relevance reaching a node is the sum of two mechanically different quantities: its content relevance through the value path and its routing relevance through the attention path","support":"source-stated"},{"text":"The authors’ product rule uses free weights on the two factors and recovers existing approaches as special cases.","excerpt":"We write a single product rule with free weights on the two factors that recovers current approaches as special cases.","support":"source-stated"},{"text":"On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three rule settings.","excerpt":"the same node receives negative, positive and near-zero relevance under three settings of the rule.","support":"source-stated"},{"text":"The authors hypothesize from the separated relevance maps that the model gains by not reading the node.","excerpt":"The separated relevance maps yield a hypothesis about the model: it gains by not reading that node.","support":"source-stated"}],"limitations":["The reported comparison is on a ground-truth classification task; the abstract does not establish how well the findings generalize to other tasks or domains.","The supplied evidence is the paper’s abstract, not its full methods or results."],"mediaId":"25644eba746b468687f782a448e79c9600dc2af44892a16001bcb7e320ec33fd","correctionReason":null,"id":"0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc","publicationId":"6fdc8ab0-e01f-4712-8d75-cae5938bc034","version":1,"sourceUrl":"https://arxiv.org/abs/2611.08088","sourceTitle":"Value or Attention: Investigating Layer-wise Relevance Propagation for Graph Attention Networks","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:10:21.850Z","createdAt":"2026-11-08T17:31:00.649Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"25644eba746b468687f782a448e79c9600dc2af44892a16001bcb7e320ec33fd","kind":"article","url":"https://arxiv.org/abs/2611.08088","sourceUrl":"https://arxiv.org/abs/2611.08088","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T17:18:49.284Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.08088","checkedAt":"2026-11-09T10:34:06.990Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"25644eba746b468687f782a448e79c9600dc2af44892a16001bcb7e320ec33fd","confidence":0.72,"provider":"typesafe","model":"jev-1.13.0","inputHash":"bfc4c6b3f55c36f6d5a78af6778a8c51bad5abcccb8a204ec3e0a644433878d5","decidedAt":"2026-11-08T17:18:50.020Z"},"provenance":{"textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observationId":"a24f0ac240bb6f6c81a5563e2f4ddddb388cc85320bcb110dc53e244e95ad59c","retrievedAt":"2026-11-08T17:18:49.182Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"31e80412-9db7-456a-8aa7-71f7f34fcf5e"},"reviewer":{"model":"gpt-6-luna","runId":"4b7e9230-1463-4b40-8ec7-f441e2925a31"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T17:31:00.649Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Schreiber, Ken","url":null,"kind":"unknown"},{"name":"Klausen, René P.","url":null,"kind":"unknown"},{"name":"Lapuschkin, Sebastian","url":null,"kind":"unknown"},{"name":"Samek, Wojciech","url":null,"kind":"unknown"},{"name":"Naujoks, Jonas","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Schreiber, Ken","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Schreiber, Ken","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Klausen, René P.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Klausen, René P.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Lapuschkin, Sebastian","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lapuschkin, Sebastian","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Samek, Wojciech","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Samek, Wojciech","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Naujoks, Jonas","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Naujoks, Jonas","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.08088","snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-schreiber-and-coauthors-separate-two-kinds-of-0ae0eb3a","formats":{"html":"/readings/arxiv-cs-ai-schreiber-and-coauthors-separate-two-kinds-of-0ae0eb3a","markdown":"/readings/arxiv-cs-ai-schreiber-and-coauthors-separate-two-kinds-of-0ae0eb3a.md","json":"/readings/arxiv-cs-ai-schreiber-and-coauthors-separate-two-kinds-of-0ae0eb3a.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc","revision":"6fdc8ab0-e01f-4712-8d75-cae5938bc034","revisionKind":"publication-id","title":"arXiv cs.AI: Schreiber and coauthors separate two kinds of relevance in Graph Attention Networks","url":"https://escapemeridian.com/readings/arxiv-cs-ai-schreiber-and-coauthors-separate-two-kinds-of-0ae0eb3a"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.08088","revision":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","revisionKind":"sha256-source-text","title":"Value or Attention: Investigating Layer-wise Relevance Propagation for Graph Attention Networks","url":"https://arxiv.org/abs/2611.08088"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc","title":"arXiv cs.AI: Schreiber and coauthors separate two kinds of relevance in Graph Attention Networks","argument":"In their paper, Ken Schreiber and coauthors argue that Layer-wise Relevance Propagation (LRP) for Graph Attention Networks should distinguish a node’s content relevance from its routing relevance. Content relevance tracks what the node supplied through the value path against a zero baseline. Routing relevance tracks how the node shaped what was passed through the attention path relative to its neighborhood; its sign indicates whether the node’s features moved the gating in the direction the prediction demanded. The authors say existing rules either discard one quantity or combine both into one number. Their single product rule uses free weights on the two factors and recovers current approaches as special cases. On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three settings. The separated maps lead the authors to hypothesize that the model gains by not reading that node. They provide reading rules for both paths and argue that the two should be interpreted separately. This distinction matters because a node can contribute through its content while also affecting the model’s routing, and a combined score can obscure those different roles. The reported example is one classification task; the abstract does not establish how often the patterns occur elsewhere.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported comparison is on a ground-truth classification task; the abstract does not establish how well the findings generalize to other tasks or domains. The supplied evidence is the paper’s abstract, not its full methods or results.","speaker":{"name":"Schreiber, Ken, Klausen, René P., Lapuschkin, Sebastian, Samek, Wojciech, Naujoks, Jonas","roleAtSource":"Article author"},"source":{"title":"Value or Attention: Investigating Layer-wise Relevance Propagation for Graph Attention Networks","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.08088","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors distinguish content relevance through the value path from routing relevance through the attention path, and argue that these should be interpreted separately.","excerpt":"the relevance reaching a node is the sum of two mechanically different quantities: its content relevance through the value path and its routing relevance through the attention path","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"},{"text":"The authors’ product rule uses free weights on the two factors and recovers existing approaches as special cases.","excerpt":"We write a single product rule with free weights on the two factors that recovers current approaches as special cases.","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"},{"text":"On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three rule settings.","excerpt":"the same node receives negative, positive and near-zero relevance under three settings of the rule.","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"},{"text":"The authors hypothesize from the separated relevance maps that the model gains by not reading the node.","excerpt":"The separated relevance maps yield a hypothesis about the model: it gains by not reading that node.","url":"https://arxiv.org/abs/2611.08088","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.08088","checkedAt":"2026-11-09T10:34:06.990Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Across other ground-truth classification tasks, do the separated relevance maps likewise distinguish cases where a model gains by not reading a node?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Schreiber and coauthors separate two kinds of relevance in Graph Attention Networks","summary":"In their paper, Ken Schreiber and coauthors argue that Layer-wise Relevance Propagation (LRP) for Graph Attention Networks should distinguish a node’s content relevance from its routing relevance. Content relevance tracks what the node supplied through the value path against a zero baseline. Routing relevance tracks how the node shaped what was passed through the attention path relative to its neighborhood; its sign indicates whether the node’s features moved the gating in the direction the prediction demanded. The authors say existing rules either discard one quantity or combine both into one number. Their single product rule uses free weights on the two factors and recovers current approaches as special cases. On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three settings. The separated maps lead the authors to hypothesize that the model gains by not reading that node. They provide reading rules for both paths and argue that the two should be interpreted separately. This distinction matters because a node can contribute through its content while also affecting the model’s routing, and a combined score can obscure those different roles. The reported example is one classification task; the abstract does not establish how often the patterns occur elsewhere.","question":"Across other ground-truth classification tasks, do the separated relevance maps likewise distinguish cases where a model gains by not reading a node?","claims":[{"text":"The authors distinguish content relevance through the value path from routing relevance through the attention path, and argue that these should be interpreted separately.","excerpt":"the relevance reaching a node is the sum of two mechanically different quantities: its content relevance through the value path and its routing relevance through the attention path","support":"source-stated"},{"text":"The authors’ product rule uses free weights on the two factors and recovers existing approaches as special cases.","excerpt":"We write a single product rule with free weights on the two factors that recovers current approaches as special cases.","support":"source-stated"},{"text":"On a ground-truth classification task, the same node receives negative, positive, or near-zero relevance under three rule settings.","excerpt":"the same node receives negative, positive and near-zero relevance under three settings of the rule.","support":"source-stated"},{"text":"The authors hypothesize from the separated relevance maps that the model gains by not reading the node.","excerpt":"The separated relevance maps yield a hypothesis about the model: it gains by not reading that node.","support":"source-stated"}],"limitations":["The reported comparison is on a ground-truth classification task; the abstract does not establish how well the findings generalize to other tasks or domains.","The supplied evidence is the paper’s abstract, not its full methods or results."],"mediaId":"25644eba746b468687f782a448e79c9600dc2af44892a16001bcb7e320ec33fd","correctionReason":null,"id":"0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc","publicationId":"6fdc8ab0-e01f-4712-8d75-cae5938bc034","version":1,"sourceUrl":"https://arxiv.org/abs/2611.08088","sourceTitle":"Value or Attention: Investigating Layer-wise Relevance Propagation for Graph Attention Networks","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:10:21.850Z","createdAt":"2026-11-08T17:31:00.649Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"25644eba746b468687f782a448e79c9600dc2af44892a16001bcb7e320ec33fd","kind":"article","url":"https://arxiv.org/abs/2611.08088","sourceUrl":"https://arxiv.org/abs/2611.08088","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T17:18:49.284Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.08088","checkedAt":"2026-11-09T10:34:06.990Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"25644eba746b468687f782a448e79c9600dc2af44892a16001bcb7e320ec33fd","confidence":0.72,"provider":"typesafe","model":"jev-1.13.0","inputHash":"bfc4c6b3f55c36f6d5a78af6778a8c51bad5abcccb8a204ec3e0a644433878d5","decidedAt":"2026-11-08T17:18:50.020Z"},"provenance":{"textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observationId":"a24f0ac240bb6f6c81a5563e2f4ddddb388cc85320bcb110dc53e244e95ad59c","retrievedAt":"2026-11-08T17:18:49.182Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"31e80412-9db7-456a-8aa7-71f7f34fcf5e"},"reviewer":{"model":"gpt-6-luna","runId":"4b7e9230-1463-4b40-8ec7-f441e2925a31"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T17:31:00.649Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Schreiber, Ken","url":null,"kind":"unknown"},{"name":"Klausen, René P.","url":null,"kind":"unknown"},{"name":"Lapuschkin, Sebastian","url":null,"kind":"unknown"},{"name":"Samek, Wojciech","url":null,"kind":"unknown"},{"name":"Naujoks, Jonas","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Schreiber, Ken","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Schreiber, Ken","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Klausen, René P.","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Klausen, René P.","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Lapuschkin, Sebastian","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lapuschkin, Sebastian","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Samek, Wojciech","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Samek, Wojciech","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}},{"actor":{"name":"Naujoks, Jonas","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Naujoks, Jonas","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.08088","locator":"head/meta[name=author|citation_author]"},"assertionId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T10:34:06.990Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","sourceUrl":"https://arxiv.org/abs/2611.08088","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.08088","snapshotId":"e97d17e40982f60bf9c9f272c7786200788da1c73f84b49df82bf429df76bd22","contentSha256":"dead8cfd7cfb59072560bd7b8066dafd7117f7af02b0623bd9195d679e8dd7de","textHash":"1ba14acf9a5691135948157fe45aed08a19c767e993770a674cd455ffb47c844","observedAt":"2026-11-09T10:34:06.990Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/0ae0eb3a98e4ba95d0008d95fd0c3e5d0a8f91e70005fb6e3ac9498b482322dc/source"}}}],"audits":[{"publicationId":"6fdc8ab0-e01f-4712-8d75-cae5938bc034","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the paper to Ken Schreiber and coauthors on arXiv. It keeps the method, results and hypothesis framed as the authors' claims.","excerpt":"Authors:Ken Schreiber, René P. Klausen, Sebastian Lapuschkin, Wojciech Samek, Jonas Naujoks"},{"dimension":"summary","status":"uncertain","explanation":"The reading is faithful but hard to follow for a non-specialist. It never says in plain words that Graph Attention Networks are AI models that work on networks of connected items, or that LRP is a method for explaining which inputs drove a prediction. Node, value path, attention path, gating and zero baseline are also left unexplained. It also drops the abstract's stated stakes: scientists in medicine and materials science need to read a model's reasoning. Smallest repair: add one plain-language sentence on what GATs and LRP do, and restore that motivation.","excerpt":"are increasingly used in domains where scientists need to read a model's reasoning, from medicine to materials science, yet the routing they learn remains opaque to most explainers"},{"dimension":"date","status":"pass","explanation":"The displayed original publication date, 2026-11-05, matches the arXiv v1 submission date. It is not the later feed date.","excerpt":"[Submitted on 5 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"All four claims are directly supported by the abstract. The 'why it matters' sentence closely tracks the authors' own argument: existing rules drop or merge the two quantities, and the same node can get opposite signs under different settings. The limitations accurately note the single task and abstract-only evidence. The question follows from the evidence without assuming an answer.","excerpt":"Existing rules either drop one or add them into a single number."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the original arXiv abstract page. It is the most direct way for readers to reach the source.","excerpt":null}],"createdAt":"2026-11-08T17:31:00.649Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}