{"id":"e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b","title":"arXiv cs.AI: InnerLogic-Surg, an open surgical vision model built through continual pretraining","argument":"The authors of InnerLogic-Surg present an open family of surgical vision encoders made by continually pretraining an image encoder on surgery-specific images. Their training data came from internet-sourced surgical videos spanning more than fifteen procedure types. The recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and adds a learned feature-upsampling module for high-resolution dense prediction at low token cost. In a benchmark spanning workflow, instrument and anatomy recognition, segmentation, detection and monocular depth, the largest model achieved the best linear-probe performance in all 20 tasks under the same architecture. The authors also report that the representations generalize across procedures, anatomical regions and task types, and call the encoders a promising, scalable alternative to procedure-specific representation learning. These are benchmark and transfer results; how they translate to performance in surgical practice remains unresolved.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported benchmark uses linear-probe performance; the abstract does not establish how that performance translates to clinical use. The abstract describes the video collection as internet-sourced and curated, but does not specify the curation criteria or provide further detail about the dataset.","speaker":{"name":"Yuan, Kun, Ding, Hao, Shen, Yiqing, Podobnik, Gasper, Killeen, Benjamin, Porras, Tito, Krieger, Axel, Unberath, Mathias","roleAtSource":"Article author"},"source":{"title":"InnerLogic-Surg: An Open Surgical Foundation Model via Continual Pretraining","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07883","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors present InnerLogic-Surg as an open family of surgical vision encoders produced through continual pretraining on surgery-specific images.","excerpt":"Here we present InnerLogic-Surg, an open family of surgical vision encoders","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"The training data were curated from 11,965 internet-sourced surgical videos spanning about 2,900 hours and more than fifteen procedure types.","excerpt":"11,965 internet-sourced surgical videos spanning ~2,900 hours and more than fifteen procedure types.","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"The training recipe uses DINOv3's pretraining stage and a learned feature-upsampling module, which the authors say enables high-resolution dense prediction at low token cost.","excerpt":"Our training recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and additionally applies a learned feature-upsampling module for high-resolution dense prediction at low token cost.","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"On a benchmark of 20 surgical tasks, the largest model achieved the best linear-probe performance in every task under the same architecture.","excerpt":"the largest InnerLogic-Surg model achieves the best linear-probe performance in all 20 surgical tasks.","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"The authors report that surgical continual pretraining produces representations that transfer across procedures, anatomical regions and task types.","excerpt":"surgical continual pretraining yields transferable representations that generalize across procedures, anatomical regions, and task types","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07883","checkedAt":"2026-11-08T20:52:07.370Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well do the reported benchmark and transfer results predict performance in surgical practice?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: InnerLogic-Surg, an open surgical vision model built through continual pretraining","summary":"The authors of InnerLogic-Surg present an open family of surgical vision encoders made by continually pretraining an image encoder on surgery-specific images. Their training data came from internet-sourced surgical videos spanning more than fifteen procedure types. The recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and adds a learned feature-upsampling module for high-resolution dense prediction at low token cost. In a benchmark spanning workflow, instrument and anatomy recognition, segmentation, detection and monocular depth, the largest model achieved the best linear-probe performance in all 20 tasks under the same architecture. The authors also report that the representations generalize across procedures, anatomical regions and task types, and call the encoders a promising, scalable alternative to procedure-specific representation learning. These are benchmark and transfer results; how they translate to performance in surgical practice remains unresolved.","question":"How well do the reported benchmark and transfer results predict performance in surgical practice?","claims":[{"text":"The authors present InnerLogic-Surg as an open family of surgical vision encoders produced through continual pretraining on surgery-specific images.","excerpt":"Here we present InnerLogic-Surg, an open family of surgical vision encoders","support":"source-stated"},{"text":"The training data were curated from 11,965 internet-sourced surgical videos spanning about 2,900 hours and more than fifteen procedure types.","excerpt":"11,965 internet-sourced surgical videos spanning ~2,900 hours and more than fifteen procedure types.","support":"source-stated"},{"text":"The training recipe uses DINOv3's pretraining stage and a learned feature-upsampling module, which the authors say enables high-resolution dense prediction at low token cost.","excerpt":"Our training recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and additionally applies a learned feature-upsampling module for high-resolution dense prediction at low token cost.","support":"source-stated"},{"text":"On a benchmark of 20 surgical tasks, the largest model achieved the best linear-probe performance in every task under the same architecture.","excerpt":"the largest InnerLogic-Surg model achieves the best linear-probe performance in all 20 surgical tasks.","support":"source-stated"},{"text":"The authors report that surgical continual pretraining produces representations that transfer across procedures, anatomical regions and task types.","excerpt":"surgical continual pretraining yields transferable representations that generalize across procedures, anatomical regions, and task types","support":"source-stated"}],"limitations":["The reported benchmark uses linear-probe performance; the abstract does not establish how that performance translates to clinical use.","The abstract describes the video collection as internet-sourced and curated, but does not specify the curation criteria or provide further detail about the dataset."],"mediaId":"c9083af93a6a1835bbb5569da0100231957ab95a343eba492b65bde010be1796","correctionReason":null,"id":"e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b","publicationId":"2262755a-e64e-4e69-8da6-0b1995deee9d","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07883","sourceTitle":"InnerLogic-Surg: An Open Surgical Foundation Model via Continual Pretraining","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:12:37.848Z","createdAt":"2026-11-08T19:49:31.477Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"c9083af93a6a1835bbb5569da0100231957ab95a343eba492b65bde010be1796","kind":"article","url":"https://arxiv.org/abs/2611.07883","sourceUrl":"https://arxiv.org/abs/2611.07883","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T19:48:41.690Z","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/2611.07883","checkedAt":"2026-11-08T20:52:07.370Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"c9083af93a6a1835bbb5569da0100231957ab95a343eba492b65bde010be1796","confidence":0.95,"provider":"typesafe","model":"jev-1.13.0","inputHash":"62fc1dfe1c56414783ba7a382c4b944c371d6788a4e1cfc32ef8c8c28db8c162","decidedAt":"2026-11-08T19:48:43.366Z"},"provenance":{"textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observationId":"eacad20bfc6342ba08aa476fd3307223666d16dab3b42feeba5abc4cd823b799","retrievedAt":"2026-11-08T19:48:41.365Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"3f772efd-a870-487a-8224-8ad97dfa051f"},"reviewer":{"model":"gpt-6-luna","runId":"4bfa6e8f-1ce0-436a-858e-55b24de5c3db"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T19:49:31.477Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Yuan, Kun","url":null,"kind":"unknown"},{"name":"Ding, Hao","url":null,"kind":"unknown"},{"name":"Shen, Yiqing","url":null,"kind":"unknown"},{"name":"Podobnik, Gasper","url":null,"kind":"unknown"},{"name":"Killeen, Benjamin","url":null,"kind":"unknown"},{"name":"Porras, Tito","url":null,"kind":"unknown"},{"name":"Krieger, Axel","url":null,"kind":"unknown"},{"name":"Unberath, Mathias","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Yuan, Kun","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yuan, Kun","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Ding, Hao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Ding, Hao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Shen, Yiqing","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shen, Yiqing","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Podobnik, Gasper","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Podobnik, Gasper","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Killeen, Benjamin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Killeen, Benjamin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Porras, Tito","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Porras, Tito","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Krieger, Axel","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Krieger, Axel","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Unberath, Mathias","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Unberath, Mathias","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07883","snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z","basis":"publisher-page-metadata","sourceLinks":47,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-innerlogic-surg-an-open-surgical-vision-model-built-e0b3aa9d","formats":{"html":"/readings/arxiv-cs-ai-innerlogic-surg-an-open-surgical-vision-model-built-e0b3aa9d","markdown":"/readings/arxiv-cs-ai-innerlogic-surg-an-open-surgical-vision-model-built-e0b3aa9d.md","json":"/readings/arxiv-cs-ai-innerlogic-surg-an-open-surgical-vision-model-built-e0b3aa9d.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":[{"id":"term-pre-training","name":"Pre-training","explanation":"A model's initial training on a large dataset, before further adaptation for particular uses.","referenceUrl":"https://developers.google.com/machine-learning/glossary#pre-training"},{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"}],"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":"e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b","revision":"2262755a-e64e-4e69-8da6-0b1995deee9d","revisionKind":"publication-id","title":"arXiv cs.AI: InnerLogic-Surg, an open surgical vision model built through continual pretraining","url":"https://escapemeridian.com/readings/arxiv-cs-ai-innerlogic-surg-an-open-surgical-vision-model-built-e0b3aa9d"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.07883","revision":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","revisionKind":"sha256-source-text","title":"InnerLogic-Surg: An Open Surgical Foundation Model via Continual Pretraining","url":"https://arxiv.org/abs/2611.07883"},"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":"e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b","title":"arXiv cs.AI: InnerLogic-Surg, an open surgical vision model built through continual pretraining","argument":"The authors of InnerLogic-Surg present an open family of surgical vision encoders made by continually pretraining an image encoder on surgery-specific images. Their training data came from internet-sourced surgical videos spanning more than fifteen procedure types. The recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and adds a learned feature-upsampling module for high-resolution dense prediction at low token cost. In a benchmark spanning workflow, instrument and anatomy recognition, segmentation, detection and monocular depth, the largest model achieved the best linear-probe performance in all 20 tasks under the same architecture. The authors also report that the representations generalize across procedures, anatomical regions and task types, and call the encoders a promising, scalable alternative to procedure-specific representation learning. These are benchmark and transfer results; how they translate to performance in surgical practice remains unresolved.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported benchmark uses linear-probe performance; the abstract does not establish how that performance translates to clinical use. The abstract describes the video collection as internet-sourced and curated, but does not specify the curation criteria or provide further detail about the dataset.","speaker":{"name":"Yuan, Kun, Ding, Hao, Shen, Yiqing, Podobnik, Gasper, Killeen, Benjamin, Porras, Tito, Krieger, Axel, Unberath, Mathias","roleAtSource":"Article author"},"source":{"title":"InnerLogic-Surg: An Open Surgical Foundation Model via Continual Pretraining","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07883","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors present InnerLogic-Surg as an open family of surgical vision encoders produced through continual pretraining on surgery-specific images.","excerpt":"Here we present InnerLogic-Surg, an open family of surgical vision encoders","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"The training data were curated from 11,965 internet-sourced surgical videos spanning about 2,900 hours and more than fifteen procedure types.","excerpt":"11,965 internet-sourced surgical videos spanning ~2,900 hours and more than fifteen procedure types.","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"The training recipe uses DINOv3's pretraining stage and a learned feature-upsampling module, which the authors say enables high-resolution dense prediction at low token cost.","excerpt":"Our training recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and additionally applies a learned feature-upsampling module for high-resolution dense prediction at low token cost.","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"On a benchmark of 20 surgical tasks, the largest model achieved the best linear-probe performance in every task under the same architecture.","excerpt":"the largest InnerLogic-Surg model achieves the best linear-probe performance in all 20 surgical tasks.","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"},{"text":"The authors report that surgical continual pretraining produces representations that transfer across procedures, anatomical regions and task types.","excerpt":"surgical continual pretraining yields transferable representations that generalize across procedures, anatomical regions, and task types","url":"https://arxiv.org/abs/2611.07883","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07883","checkedAt":"2026-11-08T20:52:07.370Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well do the reported benchmark and transfer results predict performance in surgical practice?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: InnerLogic-Surg, an open surgical vision model built through continual pretraining","summary":"The authors of InnerLogic-Surg present an open family of surgical vision encoders made by continually pretraining an image encoder on surgery-specific images. Their training data came from internet-sourced surgical videos spanning more than fifteen procedure types. The recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and adds a learned feature-upsampling module for high-resolution dense prediction at low token cost. In a benchmark spanning workflow, instrument and anatomy recognition, segmentation, detection and monocular depth, the largest model achieved the best linear-probe performance in all 20 tasks under the same architecture. The authors also report that the representations generalize across procedures, anatomical regions and task types, and call the encoders a promising, scalable alternative to procedure-specific representation learning. These are benchmark and transfer results; how they translate to performance in surgical practice remains unresolved.","question":"How well do the reported benchmark and transfer results predict performance in surgical practice?","claims":[{"text":"The authors present InnerLogic-Surg as an open family of surgical vision encoders produced through continual pretraining on surgery-specific images.","excerpt":"Here we present InnerLogic-Surg, an open family of surgical vision encoders","support":"source-stated"},{"text":"The training data were curated from 11,965 internet-sourced surgical videos spanning about 2,900 hours and more than fifteen procedure types.","excerpt":"11,965 internet-sourced surgical videos spanning ~2,900 hours and more than fifteen procedure types.","support":"source-stated"},{"text":"The training recipe uses DINOv3's pretraining stage and a learned feature-upsampling module, which the authors say enables high-resolution dense prediction at low token cost.","excerpt":"Our training recipe inherits DINOv3's pretraining stage to preserve representation ability after surgical adaptation, and additionally applies a learned feature-upsampling module for high-resolution dense prediction at low token cost.","support":"source-stated"},{"text":"On a benchmark of 20 surgical tasks, the largest model achieved the best linear-probe performance in every task under the same architecture.","excerpt":"the largest InnerLogic-Surg model achieves the best linear-probe performance in all 20 surgical tasks.","support":"source-stated"},{"text":"The authors report that surgical continual pretraining produces representations that transfer across procedures, anatomical regions and task types.","excerpt":"surgical continual pretraining yields transferable representations that generalize across procedures, anatomical regions, and task types","support":"source-stated"}],"limitations":["The reported benchmark uses linear-probe performance; the abstract does not establish how that performance translates to clinical use.","The abstract describes the video collection as internet-sourced and curated, but does not specify the curation criteria or provide further detail about the dataset."],"mediaId":"c9083af93a6a1835bbb5569da0100231957ab95a343eba492b65bde010be1796","correctionReason":null,"id":"e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b","publicationId":"2262755a-e64e-4e69-8da6-0b1995deee9d","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07883","sourceTitle":"InnerLogic-Surg: An Open Surgical Foundation Model via Continual Pretraining","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:12:37.848Z","createdAt":"2026-11-08T19:49:31.477Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"c9083af93a6a1835bbb5569da0100231957ab95a343eba492b65bde010be1796","kind":"article","url":"https://arxiv.org/abs/2611.07883","sourceUrl":"https://arxiv.org/abs/2611.07883","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T19:48:41.690Z","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/2611.07883","checkedAt":"2026-11-08T20:52:07.370Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"c9083af93a6a1835bbb5569da0100231957ab95a343eba492b65bde010be1796","confidence":0.95,"provider":"typesafe","model":"jev-1.13.0","inputHash":"62fc1dfe1c56414783ba7a382c4b944c371d6788a4e1cfc32ef8c8c28db8c162","decidedAt":"2026-11-08T19:48:43.366Z"},"provenance":{"textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observationId":"eacad20bfc6342ba08aa476fd3307223666d16dab3b42feeba5abc4cd823b799","retrievedAt":"2026-11-08T19:48:41.365Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"3f772efd-a870-487a-8224-8ad97dfa051f"},"reviewer":{"model":"gpt-6-luna","runId":"4bfa6e8f-1ce0-436a-858e-55b24de5c3db"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T19:49:31.477Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Yuan, Kun","url":null,"kind":"unknown"},{"name":"Ding, Hao","url":null,"kind":"unknown"},{"name":"Shen, Yiqing","url":null,"kind":"unknown"},{"name":"Podobnik, Gasper","url":null,"kind":"unknown"},{"name":"Killeen, Benjamin","url":null,"kind":"unknown"},{"name":"Porras, Tito","url":null,"kind":"unknown"},{"name":"Krieger, Axel","url":null,"kind":"unknown"},{"name":"Unberath, Mathias","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Yuan, Kun","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yuan, Kun","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Ding, Hao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Ding, Hao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Shen, Yiqing","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shen, Yiqing","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Podobnik, Gasper","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Podobnik, Gasper","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Killeen, Benjamin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Killeen, Benjamin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Porras, Tito","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Porras, Tito","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Krieger, Axel","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Krieger, Axel","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}},{"actor":{"name":"Unberath, Mathias","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Unberath, Mathias","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07883","locator":"head/meta[name=author|citation_author]"},"assertionId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-08T20:52:07.370Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","sourceUrl":"https://arxiv.org/abs/2611.07883","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07883","snapshotId":"9097ee311f832163ba8e45a4e127bc3859c1c3efccf956df20a1426c8777bbdf","contentSha256":"39fc0135ab58e77ef6f3d09834eb1bdac8b219616d6d082a6915bf4d17144b2e","textHash":"3a995a0ebf2341f9c3c11336fd4a6271b576ff25a23f48f6135d64cf0cd2e23b","observedAt":"2026-11-08T20:52:07.370Z","basis":"publisher-page-metadata","sourceLinks":47,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/e0b3aa9d52cbda2b4a8afc3d3b89fba882d14a608444d97116a72fef1f4b486b/source"}}}],"audits":[{"publicationId":"2262755a-e64e-4e69-8da6-0b1995deee9d","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the work and its claims to the paper's authors on arXiv. The paper's primary category is cs.CV, cross-listed in cs.AI, so the 'arXiv cs.AI' label matches the publisher metadata. Performance and generalization statements are framed as the authors' reports.","excerpt":"Authors:Kun Yuan, Hao Ding, Yiqing Shen, Gasper Podobnik, Benjamin Killeen, Tito Porras, Axel Krieger, Mathias Unberath"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but central terms go unexplained for a non-specialist: vision encoders, continual pretraining, DINOv3, linear-probe, dense prediction, token cost and monocular depth. A lay reader cannot tell what the headline result measures. It also omits the authors' framing of why this matters: most surgical systems rely on encoders pretrained on natural images, or on specialists trained on small datasets. Smallest repair: add that one-sentence motivation, and give plain-language glosses or tooltip entries for vision encoder, continual pretraining and linear-probe.","excerpt":"most surgical systems still rely on encoders pretrained on natural images or on specialist models trained from scratch on comparatively small datasets"},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-11-05 matches the arXiv v1 submission date. The feed date of 2026-11-06 is correctly not used.","excerpt":"[v1]\nThu, 5 Nov 2026 12:03:01 UTC (20,250 KB)"},{"dimension":"support","status":"pass","explanation":"Each claim matches its excerpt: the open encoder family, 11,965 videos and about 2,900 hours, the DINOv3 recipe with feature upsampling, best linear-probe results on all 20 tasks under the same architecture, and transfer across procedures. The caveat about clinical translation is labeled as unresolved rather than asserted. The question follows from the evidence without assuming an answer.","excerpt":"under the same architecture, the largest InnerLogic-Surg model achieves the best linear-probe performance in all 20 surgical tasks"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way to reach the source.","excerpt":null}],"createdAt":"2026-11-08T19:49:31.477Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}