{"id":"241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291","title":"arXiv cs.AI: THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","argument":"The authors of THG-Omni argue that answer-level supervision alone does not teach video models to interpret structured evidence, and that predictions often fail to show which participants, events, or intervals support an answer. Their framework links participant tracks with temporally grounded annotations. It checks candidate semantic claims against the source video and available audio before using them to construct training questions; track geometry supplies spatial and temporal supervision. At inference, THG-Omni uses a predicted graph constructed from the evaluation video. The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception. On NExT-QA, they report 89.54% accuracy with the matched graph, compared with 85.10% without graph input. Further validation reports 78.22% accuracy with compact graph input, 78.02% with the full graph, and 77.58% without graph input. The authors also say a diagnostic shows that the predicted graph preserves evidence useful for answering questions even without the source video and audio. The abstract does not quantify the IntentBench improvements or provide enough experimental detail to assess how broadly the results apply.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract does not quantify the reported IntentBench improvements or give enough experimental detail to assess how broadly the results apply. The NExT-QA accuracy comparison reports results with a matched graph and without graph input; the abstract does not report that comparison using a predicted graph.","speaker":{"name":"Dang, Jisheng, Wei, Kai, Lin, Ronghao, Wang, Bimei, Peng, Hong, Zheng, Wei-Shi, Hu, Bin, Tian, Qi, Chua, Tat-Seng","roleAtSource":"Article author"},"source":{"title":"THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07836","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that answer-level supervision alone does not teach models to interpret structured video evidence, and that predictions often do not identify their supporting participants, events, or intervals.","excerpt":"answer-level supervision alone does not teach a model to interpret structured video evidence","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"THG-Omni combines participant tracks with temporally grounded annotations and checks candidate semantic claims against video and available audio before using them to construct training questions.","excerpt":"Candidate semantic claims are checked against the source video and available audio before they are used to construct training questions","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","excerpt":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"On NExT-QA, the authors report 89.54% accuracy with the matched graph, compared with 85.10% without graph input.","excerpt":"On NExT-QA, THG-Omni achieves 89.54\\% accuracy with the matched graph, compared with 85.10\\% without graph input.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"In further validation, compact graph input scored 78.22% accuracy, compared with 78.02% using the full graph and 77.58% without graph input.","excerpt":"compact graph input achieves 78.22\\% accuracy, compared with 78.02\\% using the full graph and 77.58\\% without graph input.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception.","excerpt":"Experiments on IntentBench show consistent improvements over the unadapted backbone across social reasoning, emotion understanding, and deception perception.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07836","checkedAt":"2026-11-09T07:56:07.576Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How does NExT-QA performance with the predicted graph compare with the reported matched-graph result?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","summary":"The authors of THG-Omni argue that answer-level supervision alone does not teach video models to interpret structured evidence, and that predictions often fail to show which participants, events, or intervals support an answer. Their framework links participant tracks with temporally grounded annotations. It checks candidate semantic claims against the source video and available audio before using them to construct training questions; track geometry supplies spatial and temporal supervision. At inference, THG-Omni uses a predicted graph constructed from the evaluation video. The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception. On NExT-QA, they report 89.54% accuracy with the matched graph, compared with 85.10% without graph input. Further validation reports 78.22% accuracy with compact graph input, 78.02% with the full graph, and 77.58% without graph input. The authors also say a diagnostic shows that the predicted graph preserves evidence useful for answering questions even without the source video and audio. The abstract does not quantify the IntentBench improvements or provide enough experimental detail to assess how broadly the results apply.","question":"How does NExT-QA performance with the predicted graph compare with the reported matched-graph result?","claims":[{"text":"The authors argue that answer-level supervision alone does not teach models to interpret structured video evidence, and that predictions often do not identify their supporting participants, events, or intervals.","excerpt":"answer-level supervision alone does not teach a model to interpret structured video evidence","support":"source-stated"},{"text":"THG-Omni combines participant tracks with temporally grounded annotations and checks candidate semantic claims against video and available audio before using them to construct training questions.","excerpt":"Candidate semantic claims are checked against the source video and available audio before they are used to construct training questions","support":"source-stated"},{"text":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","excerpt":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","support":"source-stated"},{"text":"On NExT-QA, the authors report 89.54% accuracy with the matched graph, compared with 85.10% without graph input.","excerpt":"On NExT-QA, THG-Omni achieves 89.54\\% accuracy with the matched graph, compared with 85.10\\% without graph input.","support":"source-stated"},{"text":"In further validation, compact graph input scored 78.22% accuracy, compared with 78.02% using the full graph and 77.58% without graph input.","excerpt":"compact graph input achieves 78.22\\% accuracy, compared with 78.02\\% using the full graph and 77.58\\% without graph input.","support":"source-stated"},{"text":"The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception.","excerpt":"Experiments on IntentBench show consistent improvements over the unadapted backbone across social reasoning, emotion understanding, and deception perception.","support":"source-stated"}],"limitations":["The abstract does not quantify the reported IntentBench improvements or give enough experimental detail to assess how broadly the results apply.","The NExT-QA accuracy comparison reports results with a matched graph and without graph input; the abstract does not report that comparison using a predicted graph."],"mediaId":"41cf5dbe1c2ee8bca226f1306e8f8d78cc77655bd12f54bf045e9042c632b829","correctionReason":null,"id":"241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291","publicationId":"0ee548b0-b492-4ee9-8a0c-9be202364b4d","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07836","sourceTitle":"THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:12:31.691Z","createdAt":"2026-11-08T23:24:04.768Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"41cf5dbe1c2ee8bca226f1306e8f8d78cc77655bd12f54bf045e9042c632b829","kind":"article","url":"https://arxiv.org/abs/2611.07836","sourceUrl":"https://arxiv.org/abs/2611.07836","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T23:22:54.875Z","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/2611.07836","checkedAt":"2026-11-09T07:56:07.576Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"41cf5dbe1c2ee8bca226f1306e8f8d78cc77655bd12f54bf045e9042c632b829","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"fa95369af316a6c7d1fb403cccbb129997a72d372a16783fae7fba08fad9ac21","decidedAt":"2026-11-08T23:22:55.283Z"},"provenance":{"textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observationId":"dc80577a27dddd3f53b25e71ec27976da5d442a6d06ac575dc9d5beb7870752a","retrievedAt":"2026-11-08T23:22:54.817Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"a7dfba0e-0af7-4be4-8db9-27939337160a"},"reviewer":{"model":"gpt-6-luna","runId":"23790541-cf5d-486b-8de6-e1a27de61ad4"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T23:24:04.768Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Dang, Jisheng","url":null,"kind":"unknown"},{"name":"Wei, Kai","url":null,"kind":"unknown"},{"name":"Lin, Ronghao","url":null,"kind":"unknown"},{"name":"Wang, Bimei","url":null,"kind":"unknown"},{"name":"Peng, Hong","url":null,"kind":"unknown"},{"name":"Zheng, Wei-Shi","url":null,"kind":"unknown"},{"name":"Hu, Bin","url":null,"kind":"unknown"},{"name":"Tian, Qi","url":null,"kind":"unknown"},{"name":"Chua, Tat-Seng","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Dang, Jisheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Dang, Jisheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Wei, Kai","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wei, Kai","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Lin, Ronghao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lin, Ronghao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Wang, Bimei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Bimei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Peng, Hong","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Peng, Hong","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Zheng, Wei-Shi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zheng, Wei-Shi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Hu, Bin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Hu, Bin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Tian, Qi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Tian, Qi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Chua, Tat-Seng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chua, Tat-Seng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:8","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}}],"participationCoverage":{"total":9,"returned":9,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07836","snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z","basis":"publisher-page-metadata","sourceLinks":49,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-thg-omni-traceable-human-centric-graph-supervision-241a0daa","formats":{"html":"/readings/arxiv-cs-ai-thg-omni-traceable-human-centric-graph-supervision-241a0daa","markdown":"/readings/arxiv-cs-ai-thg-omni-traceable-human-centric-graph-supervision-241a0daa.md","json":"/readings/arxiv-cs-ai-thg-omni-traceable-human-centric-graph-supervision-241a0daa.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":"241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291","revision":"0ee548b0-b492-4ee9-8a0c-9be202364b4d","revisionKind":"publication-id","title":"arXiv cs.AI: THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","url":"https://escapemeridian.com/readings/arxiv-cs-ai-thg-omni-traceable-human-centric-graph-supervision-241a0daa"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.07836","revision":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","revisionKind":"sha256-source-text","title":"THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","url":"https://arxiv.org/abs/2611.07836"},"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":"241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291","title":"arXiv cs.AI: THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","argument":"The authors of THG-Omni argue that answer-level supervision alone does not teach video models to interpret structured evidence, and that predictions often fail to show which participants, events, or intervals support an answer. Their framework links participant tracks with temporally grounded annotations. It checks candidate semantic claims against the source video and available audio before using them to construct training questions; track geometry supplies spatial and temporal supervision. At inference, THG-Omni uses a predicted graph constructed from the evaluation video. The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception. On NExT-QA, they report 89.54% accuracy with the matched graph, compared with 85.10% without graph input. Further validation reports 78.22% accuracy with compact graph input, 78.02% with the full graph, and 77.58% without graph input. The authors also say a diagnostic shows that the predicted graph preserves evidence useful for answering questions even without the source video and audio. The abstract does not quantify the IntentBench improvements or provide enough experimental detail to assess how broadly the results apply.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract does not quantify the reported IntentBench improvements or give enough experimental detail to assess how broadly the results apply. The NExT-QA accuracy comparison reports results with a matched graph and without graph input; the abstract does not report that comparison using a predicted graph.","speaker":{"name":"Dang, Jisheng, Wei, Kai, Lin, Ronghao, Wang, Bimei, Peng, Hong, Zheng, Wei-Shi, Hu, Bin, Tian, Qi, Chua, Tat-Seng","roleAtSource":"Article author"},"source":{"title":"THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07836","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors argue that answer-level supervision alone does not teach models to interpret structured video evidence, and that predictions often do not identify their supporting participants, events, or intervals.","excerpt":"answer-level supervision alone does not teach a model to interpret structured video evidence","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"THG-Omni combines participant tracks with temporally grounded annotations and checks candidate semantic claims against video and available audio before using them to construct training questions.","excerpt":"Candidate semantic claims are checked against the source video and available audio before they are used to construct training questions","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","excerpt":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"On NExT-QA, the authors report 89.54% accuracy with the matched graph, compared with 85.10% without graph input.","excerpt":"On NExT-QA, THG-Omni achieves 89.54\\% accuracy with the matched graph, compared with 85.10\\% without graph input.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"In further validation, compact graph input scored 78.22% accuracy, compared with 78.02% using the full graph and 77.58% without graph input.","excerpt":"compact graph input achieves 78.22\\% accuracy, compared with 78.02\\% using the full graph and 77.58\\% without graph input.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"},{"text":"The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception.","excerpt":"Experiments on IntentBench show consistent improvements over the unadapted backbone across social reasoning, emotion understanding, and deception perception.","url":"https://arxiv.org/abs/2611.07836","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07836","checkedAt":"2026-11-09T07:56:07.576Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How does NExT-QA performance with the predicted graph compare with the reported matched-graph result?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","summary":"The authors of THG-Omni argue that answer-level supervision alone does not teach video models to interpret structured evidence, and that predictions often fail to show which participants, events, or intervals support an answer. Their framework links participant tracks with temporally grounded annotations. It checks candidate semantic claims against the source video and available audio before using them to construct training questions; track geometry supplies spatial and temporal supervision. At inference, THG-Omni uses a predicted graph constructed from the evaluation video. The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception. On NExT-QA, they report 89.54% accuracy with the matched graph, compared with 85.10% without graph input. Further validation reports 78.22% accuracy with compact graph input, 78.02% with the full graph, and 77.58% without graph input. The authors also say a diagnostic shows that the predicted graph preserves evidence useful for answering questions even without the source video and audio. The abstract does not quantify the IntentBench improvements or provide enough experimental detail to assess how broadly the results apply.","question":"How does NExT-QA performance with the predicted graph compare with the reported matched-graph result?","claims":[{"text":"The authors argue that answer-level supervision alone does not teach models to interpret structured video evidence, and that predictions often do not identify their supporting participants, events, or intervals.","excerpt":"answer-level supervision alone does not teach a model to interpret structured video evidence","support":"source-stated"},{"text":"THG-Omni combines participant tracks with temporally grounded annotations and checks candidate semantic claims against video and available audio before using them to construct training questions.","excerpt":"Candidate semantic claims are checked against the source video and available audio before they are used to construct training questions","support":"source-stated"},{"text":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","excerpt":"At inference, THG-Omni uses a predicted graph constructed from the evaluation video.","support":"source-stated"},{"text":"On NExT-QA, the authors report 89.54% accuracy with the matched graph, compared with 85.10% without graph input.","excerpt":"On NExT-QA, THG-Omni achieves 89.54\\% accuracy with the matched graph, compared with 85.10\\% without graph input.","support":"source-stated"},{"text":"In further validation, compact graph input scored 78.22% accuracy, compared with 78.02% using the full graph and 77.58% without graph input.","excerpt":"compact graph input achieves 78.22\\% accuracy, compared with 78.02\\% using the full graph and 77.58\\% without graph input.","support":"source-stated"},{"text":"The authors report consistent improvements over the unadapted backbone on IntentBench across social reasoning, emotion understanding, and deception perception.","excerpt":"Experiments on IntentBench show consistent improvements over the unadapted backbone across social reasoning, emotion understanding, and deception perception.","support":"source-stated"}],"limitations":["The abstract does not quantify the reported IntentBench improvements or give enough experimental detail to assess how broadly the results apply.","The NExT-QA accuracy comparison reports results with a matched graph and without graph input; the abstract does not report that comparison using a predicted graph."],"mediaId":"41cf5dbe1c2ee8bca226f1306e8f8d78cc77655bd12f54bf045e9042c632b829","correctionReason":null,"id":"241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291","publicationId":"0ee548b0-b492-4ee9-8a0c-9be202364b4d","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07836","sourceTitle":"THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:12:31.691Z","createdAt":"2026-11-08T23:24:04.768Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"41cf5dbe1c2ee8bca226f1306e8f8d78cc77655bd12f54bf045e9042c632b829","kind":"article","url":"https://arxiv.org/abs/2611.07836","sourceUrl":"https://arxiv.org/abs/2611.07836","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T23:22:54.875Z","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/2611.07836","checkedAt":"2026-11-09T07:56:07.576Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"41cf5dbe1c2ee8bca226f1306e8f8d78cc77655bd12f54bf045e9042c632b829","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"fa95369af316a6c7d1fb403cccbb129997a72d372a16783fae7fba08fad9ac21","decidedAt":"2026-11-08T23:22:55.283Z"},"provenance":{"textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observationId":"dc80577a27dddd3f53b25e71ec27976da5d442a6d06ac575dc9d5beb7870752a","retrievedAt":"2026-11-08T23:22:54.817Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"a7dfba0e-0af7-4be4-8db9-27939337160a"},"reviewer":{"model":"gpt-6-luna","runId":"23790541-cf5d-486b-8de6-e1a27de61ad4"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T23:24:04.768Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Dang, Jisheng","url":null,"kind":"unknown"},{"name":"Wei, Kai","url":null,"kind":"unknown"},{"name":"Lin, Ronghao","url":null,"kind":"unknown"},{"name":"Wang, Bimei","url":null,"kind":"unknown"},{"name":"Peng, Hong","url":null,"kind":"unknown"},{"name":"Zheng, Wei-Shi","url":null,"kind":"unknown"},{"name":"Hu, Bin","url":null,"kind":"unknown"},{"name":"Tian, Qi","url":null,"kind":"unknown"},{"name":"Chua, Tat-Seng","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Dang, Jisheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Dang, Jisheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Wei, Kai","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wei, Kai","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Lin, Ronghao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lin, Ronghao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Wang, Bimei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Bimei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Peng, Hong","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Peng, Hong","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Zheng, Wei-Shi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zheng, Wei-Shi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Hu, Bin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Hu, Bin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Tian, Qi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Tian, Qi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}},{"actor":{"name":"Chua, Tat-Seng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chua, Tat-Seng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07836","locator":"head/meta[name=author|citation_author]"},"assertionId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070:participation:8","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:56:07.576Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","sourceUrl":"https://arxiv.org/abs/2611.07836","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z"}}],"participationCoverage":{"total":9,"returned":9,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07836","snapshotId":"0855651b30506de76ae595b30300a6af426245ea30654ea79a7b9b6fde7e4070","contentSha256":"f62fa60d784b3e199e5a55d63e565b3775955f106da1c8206d4f9eea8d760525","textHash":"69c5fe235af8def4504a7843f6dd72d0338559abb8b67c370d06b33c3b073527","observedAt":"2026-11-09T07:56:07.576Z","basis":"publisher-page-metadata","sourceLinks":49,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/241a0daa832045144b0f2c0bf957dbe7f31a41d50f43d93cf7b5f11bf3349291/source"}}}],"audits":[{"publicationId":"0ee548b0-b492-4ee9-8a0c-9be202364b4d","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading consistently attributes the motivation, method and results to the paper's authors (\"The authors argue\", \"they report\"). The title uses the arXiv cs.AI feed label, which matches the cs.AI cross-listing. No claim is presented as established fact or as Escape Meridian's view.","excerpt":"Subjects:\nComputer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful but hard for non-specialists to follow. Central terms are never explained in plain language: what the human-centric graph is (a structured map of participants, their attributes, actions, affect, interactions and timing), omni-modal, answer-level supervision, unadapted backbone, inference, matched graph, IntentBench and NExT-QA. It also drops the named tools (Segment Anything Model 3 tracks, Qwen3-VL annotations). It does not say why traceability matters, namely answers grounded in identifiable participants and events. Smallest repair: add a plain-language sentence defining the graph and its purpose, and add tooltips for the benchmark and model terms.","excerpt":"teach an omni-modal reasoning model to interpret participant attributes, actions, affect, interactions, and their timing from a structured graph"},{"dimension":"date","status":"pass","explanation":"evidence.originalPublished is 2026-11-05, which matches the arXiv v1 submission date. It is not the later feed date of 2026-11-06.","excerpt":"[Submitted on 5 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Every claim and figure (89.54% vs 85.10%; 78.22%, 78.02%, 77.58%) matches the abstract verbatim and is framed as an author report. The limitations are accurate: the IntentBench gains are not quantified, and the NExT-QA comparison uses the matched graph, not the predicted graph used at inference. The question follows directly from that gap and is not leading. The reading does not invent any benefits or costs.","excerpt":"On NExT-QA, THG-Omni achieves 89.54\\% accuracy with the matched graph, compared with 85.10\\% without graph input."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct way to reach the source. No better source media was found.","excerpt":null}],"createdAt":"2026-11-08T23:24:04.768Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}