{"id":"d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e","title":"RetinaGuard tests when extra evidence helps retinal screening decisions","argument":"The authors of RetinaGuard argue that automated retinal screening should accept decisions without human review only when their risk is controlled. Their framework gathers evidence about each fundus photograph through quality, diagnostic, uncertainty, verification and explanation agents. A learned failure predictor estimates whether a decision is unsafe; auxiliary tools are invoked only when their expected value of information exceeds their cost, and conformal risk control certifies the acceptance threshold. On the RFMiD test split, the authors compare RetinaGuard with confidence and calibrated-uncertainty referral using two RETFound models. With the weaker model, verifier disagreement added significant information: at matched referral budgets and equal certified safety, RetinaGuard captured more unsafe decisions than lowest-confidence referral. At a nominal 5% unsafe level, it also accepted more cases. With the stronger model, the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points. The authors’ central qualification is that agentic evidence gathering helps according to how much the primary model’s uncertainty already reveals about its errors. They also report that gating cut computation by about 30%, and that a quality agent reduced unsafe acceptances under most image corruptions.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the paper abstract, which reports results on the RFMiD test split and describes comparisons using two RETFound primary models; it does not establish performance beyond that evaluation.","speaker":{"name":"Sulman, Sohaib, Saeed, Muhammad Atif, Jamil, Akhtar, Hameed, Alaa Ali, Aydin, Roland","roleAtSource":"Article author"},"source":{"title":"RetinaGuard: Risk-Controlled Agentic Triage for Multi-Disease Retinal Screening","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.10689","publishedAt":"2026-11-07","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors present RetinaGuard as a triage framework that gathers multiple kinds of evidence and controls which automated decisions are accepted without review.","excerpt":"quality, diagnostic, uncertainty, verification and explanation agents gather evidence for each fundus photograph","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"The framework uses a learned failure predictor, invokes auxiliary tools according to their expected value relative to cost, and sets its acceptance threshold with conformal risk control.","excerpt":"a learned failure predictor estimates the probability that the decision is unsafe, auxiliary tools are invoked only when their expected value of information exceeds their cost, and the acceptance threshold is certified by conformal risk control.","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"On the RFMiD test split, the weaker RETFound model showed an advantage for RetinaGuard over lowest-confidence referral at matched referral budgets and equal certified safety.","excerpt":"With the weaker model (macro-AUROC 0.896), verifier disagreement added significant information, RetinaGuard captured 119 of 280 unsafe decisions at a 20% budget against 88 for lowest-confidence referral","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"With the stronger RETFound model, verifier disagreement added no information beyond uncertainty, and RetinaGuard's advantage was smaller.","excerpt":"With the stronger model (0.925) the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points.","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"The authors say the benefit of agentic evidence gathering depends on how much the primary model's uncertainty already reveals about its errors.","excerpt":"The benefit of agentic evidence gathering thus depends on how much the primary model's own uncertainty already reveals about its errors.","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"The reported evaluation on the RFMiD test split does not establish performance beyond that evaluation.","excerpt":"On the RFMiD test split, RetinaGuard was compared with confidence and calibrated-uncertainty referral","url":"https://arxiv.org/abs/2611.10689","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.10689","checkedAt":"2026-11-12T08:56:10.527Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would RetinaGuard perform beyond the RFMiD test split, where the authors report these results?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"RetinaGuard tests when extra evidence helps retinal screening decisions","summary":"The authors of RetinaGuard argue that automated retinal screening should accept decisions without human review only when their risk is controlled. Their framework gathers evidence about each fundus photograph through quality, diagnostic, uncertainty, verification and explanation agents. A learned failure predictor estimates whether a decision is unsafe; auxiliary tools are invoked only when their expected value of information exceeds their cost, and conformal risk control certifies the acceptance threshold. On the RFMiD test split, the authors compare RetinaGuard with confidence and calibrated-uncertainty referral using two RETFound models. With the weaker model, verifier disagreement added significant information: at matched referral budgets and equal certified safety, RetinaGuard captured more unsafe decisions than lowest-confidence referral. At a nominal 5% unsafe level, it also accepted more cases. With the stronger model, the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points. The authors’ central qualification is that agentic evidence gathering helps according to how much the primary model’s uncertainty already reveals about its errors. They also report that gating cut computation by about 30%, and that a quality agent reduced unsafe acceptances under most image corruptions.","question":"How would RetinaGuard perform beyond the RFMiD test split, where the authors report these results?","claims":[{"text":"The authors present RetinaGuard as a triage framework that gathers multiple kinds of evidence and controls which automated decisions are accepted without review.","excerpt":"quality, diagnostic, uncertainty, verification and explanation agents gather evidence for each fundus photograph","support":"source-stated"},{"text":"The framework uses a learned failure predictor, invokes auxiliary tools according to their expected value relative to cost, and sets its acceptance threshold with conformal risk control.","excerpt":"a learned failure predictor estimates the probability that the decision is unsafe, auxiliary tools are invoked only when their expected value of information exceeds their cost, and the acceptance threshold is certified by conformal risk control.","support":"source-stated"},{"text":"On the RFMiD test split, the weaker RETFound model showed an advantage for RetinaGuard over lowest-confidence referral at matched referral budgets and equal certified safety.","excerpt":"With the weaker model (macro-AUROC 0.896), verifier disagreement added significant information, RetinaGuard captured 119 of 280 unsafe decisions at a 20% budget against 88 for lowest-confidence referral","support":"source-stated"},{"text":"With the stronger RETFound model, verifier disagreement added no information beyond uncertainty, and RetinaGuard's advantage was smaller.","excerpt":"With the stronger model (0.925) the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points.","support":"source-stated"},{"text":"The authors say the benefit of agentic evidence gathering depends on how much the primary model's uncertainty already reveals about its errors.","excerpt":"The benefit of agentic evidence gathering thus depends on how much the primary model's own uncertainty already reveals about its errors.","support":"source-stated"},{"text":"The reported evaluation on the RFMiD test split does not establish performance beyond that evaluation.","excerpt":"On the RFMiD test split, RetinaGuard was compared with confidence and calibrated-uncertainty referral","support":"interpretation"}],"limitations":["The supplied evidence is the paper abstract, which reports results on the RFMiD test split and describes comparisons using two RETFound primary models; it does not establish performance beyond that evaluation."],"mediaId":"dbb7fd695abc8fd4d52f64e521c5fbb2169b049ae05f24bd76fa742a2f22601b","correctionReason":null,"id":"d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e","publicationId":"f933e933-f8e3-4183-8b75-ab396827b93f","version":1,"sourceUrl":"https://arxiv.org/abs/2611.10689","sourceTitle":"RetinaGuard: Risk-Controlled Agentic Triage for Multi-Disease Retinal Screening","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-07","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:39:58.190Z","createdAt":"2026-11-12T06:49:14.861Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"dbb7fd695abc8fd4d52f64e521c5fbb2169b049ae05f24bd76fa742a2f22601b","kind":"article","url":"https://arxiv.org/abs/2611.10689","sourceUrl":"https://arxiv.org/abs/2611.10689","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T06:48:05.346Z","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/2611.10689","checkedAt":"2026-11-12T08:56:10.527Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"dbb7fd695abc8fd4d52f64e521c5fbb2169b049ae05f24bd76fa742a2f22601b","confidence":0.63,"provider":"typesafe","model":"jev-1.13.0","inputHash":"c5cb5f4500a699d7a6352dd41789169356ad461bc658f2d63c0268fb61f2a769","decidedAt":"2026-11-12T06:48:06.869Z"},"provenance":{"textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observationId":"c9c7ffdb52b0a229589758b28ac9c70088e951719bf54d28042386a150a15382","retrievedAt":"2026-11-12T06:48:05.238Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"262c628c-0e53-41eb-8e4a-f95e9ce123e4"},"reviewer":{"model":"gpt-6-luna","runId":"3125a2df-46a9-4773-84c9-2b5e3f06978d"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T06:49:14.861Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Sulman, Sohaib","url":null,"kind":"unknown"},{"name":"Saeed, Muhammad Atif","url":null,"kind":"unknown"},{"name":"Jamil, Akhtar","url":null,"kind":"unknown"},{"name":"Hameed, Alaa Ali","url":null,"kind":"unknown"},{"name":"Aydin, Roland","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Sulman, Sohaib","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sulman, Sohaib","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Saeed, Muhammad Atif","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Saeed, Muhammad Atif","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Jamil, Akhtar","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jamil, Akhtar","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Hameed, Alaa Ali","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Hameed, Alaa Ali","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Aydin, Roland","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Aydin, Roland","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.10689","snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e/source"}},"canonicalUrl":"https://escapemeridian.com/readings/retinaguard-tests-when-extra-evidence-helps-retinal-screening-d648daae","formats":{"html":"/readings/retinaguard-tests-when-extra-evidence-helps-retinal-screening-d648daae","markdown":"/readings/retinaguard-tests-when-extra-evidence-helps-retinal-screening-d648daae.md","json":"/readings/retinaguard-tests-when-extra-evidence-helps-retinal-screening-d648daae.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-agentic","name":"Agentic AI","explanation":"AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer.","referenceUrl":"https://developers.google.com/machine-learning/glossary#agentic"}],"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":"d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e","revision":"f933e933-f8e3-4183-8b75-ab396827b93f","revisionKind":"publication-id","title":"RetinaGuard tests when extra evidence helps retinal screening decisions","url":"https://escapemeridian.com/readings/retinaguard-tests-when-extra-evidence-helps-retinal-screening-d648daae"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.10689","revision":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","revisionKind":"sha256-source-text","title":"RetinaGuard: Risk-Controlled Agentic Triage for Multi-Disease Retinal Screening","url":"https://arxiv.org/abs/2611.10689"},"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":"d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e","title":"RetinaGuard tests when extra evidence helps retinal screening decisions","argument":"The authors of RetinaGuard argue that automated retinal screening should accept decisions without human review only when their risk is controlled. Their framework gathers evidence about each fundus photograph through quality, diagnostic, uncertainty, verification and explanation agents. A learned failure predictor estimates whether a decision is unsafe; auxiliary tools are invoked only when their expected value of information exceeds their cost, and conformal risk control certifies the acceptance threshold. On the RFMiD test split, the authors compare RetinaGuard with confidence and calibrated-uncertainty referral using two RETFound models. With the weaker model, verifier disagreement added significant information: at matched referral budgets and equal certified safety, RetinaGuard captured more unsafe decisions than lowest-confidence referral. At a nominal 5% unsafe level, it also accepted more cases. With the stronger model, the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points. The authors’ central qualification is that agentic evidence gathering helps according to how much the primary model’s uncertainty already reveals about its errors. They also report that gating cut computation by about 30%, and that a quality agent reduced unsafe acceptances under most image corruptions.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the paper abstract, which reports results on the RFMiD test split and describes comparisons using two RETFound primary models; it does not establish performance beyond that evaluation.","speaker":{"name":"Sulman, Sohaib, Saeed, Muhammad Atif, Jamil, Akhtar, Hameed, Alaa Ali, Aydin, Roland","roleAtSource":"Article author"},"source":{"title":"RetinaGuard: Risk-Controlled Agentic Triage for Multi-Disease Retinal Screening","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.10689","publishedAt":"2026-11-07","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors present RetinaGuard as a triage framework that gathers multiple kinds of evidence and controls which automated decisions are accepted without review.","excerpt":"quality, diagnostic, uncertainty, verification and explanation agents gather evidence for each fundus photograph","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"The framework uses a learned failure predictor, invokes auxiliary tools according to their expected value relative to cost, and sets its acceptance threshold with conformal risk control.","excerpt":"a learned failure predictor estimates the probability that the decision is unsafe, auxiliary tools are invoked only when their expected value of information exceeds their cost, and the acceptance threshold is certified by conformal risk control.","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"On the RFMiD test split, the weaker RETFound model showed an advantage for RetinaGuard over lowest-confidence referral at matched referral budgets and equal certified safety.","excerpt":"With the weaker model (macro-AUROC 0.896), verifier disagreement added significant information, RetinaGuard captured 119 of 280 unsafe decisions at a 20% budget against 88 for lowest-confidence referral","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"With the stronger RETFound model, verifier disagreement added no information beyond uncertainty, and RetinaGuard's advantage was smaller.","excerpt":"With the stronger model (0.925) the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points.","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"The authors say the benefit of agentic evidence gathering depends on how much the primary model's uncertainty already reveals about its errors.","excerpt":"The benefit of agentic evidence gathering thus depends on how much the primary model's own uncertainty already reveals about its errors.","url":"https://arxiv.org/abs/2611.10689","relationship":"Source-stated"},{"text":"The reported evaluation on the RFMiD test split does not establish performance beyond that evaluation.","excerpt":"On the RFMiD test split, RetinaGuard was compared with confidence and calibrated-uncertainty referral","url":"https://arxiv.org/abs/2611.10689","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.10689","checkedAt":"2026-11-12T08:56:10.527Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would RetinaGuard perform beyond the RFMiD test split, where the authors report these results?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"RetinaGuard tests when extra evidence helps retinal screening decisions","summary":"The authors of RetinaGuard argue that automated retinal screening should accept decisions without human review only when their risk is controlled. Their framework gathers evidence about each fundus photograph through quality, diagnostic, uncertainty, verification and explanation agents. A learned failure predictor estimates whether a decision is unsafe; auxiliary tools are invoked only when their expected value of information exceeds their cost, and conformal risk control certifies the acceptance threshold. On the RFMiD test split, the authors compare RetinaGuard with confidence and calibrated-uncertainty referral using two RETFound models. With the weaker model, verifier disagreement added significant information: at matched referral budgets and equal certified safety, RetinaGuard captured more unsafe decisions than lowest-confidence referral. At a nominal 5% unsafe level, it also accepted more cases. With the stronger model, the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points. The authors’ central qualification is that agentic evidence gathering helps according to how much the primary model’s uncertainty already reveals about its errors. They also report that gating cut computation by about 30%, and that a quality agent reduced unsafe acceptances under most image corruptions.","question":"How would RetinaGuard perform beyond the RFMiD test split, where the authors report these results?","claims":[{"text":"The authors present RetinaGuard as a triage framework that gathers multiple kinds of evidence and controls which automated decisions are accepted without review.","excerpt":"quality, diagnostic, uncertainty, verification and explanation agents gather evidence for each fundus photograph","support":"source-stated"},{"text":"The framework uses a learned failure predictor, invokes auxiliary tools according to their expected value relative to cost, and sets its acceptance threshold with conformal risk control.","excerpt":"a learned failure predictor estimates the probability that the decision is unsafe, auxiliary tools are invoked only when their expected value of information exceeds their cost, and the acceptance threshold is certified by conformal risk control.","support":"source-stated"},{"text":"On the RFMiD test split, the weaker RETFound model showed an advantage for RetinaGuard over lowest-confidence referral at matched referral budgets and equal certified safety.","excerpt":"With the weaker model (macro-AUROC 0.896), verifier disagreement added significant information, RetinaGuard captured 119 of 280 unsafe decisions at a 20% budget against 88 for lowest-confidence referral","support":"source-stated"},{"text":"With the stronger RETFound model, verifier disagreement added no information beyond uncertainty, and RetinaGuard's advantage was smaller.","excerpt":"With the stronger model (0.925) the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points.","support":"source-stated"},{"text":"The authors say the benefit of agentic evidence gathering depends on how much the primary model's uncertainty already reveals about its errors.","excerpt":"The benefit of agentic evidence gathering thus depends on how much the primary model's own uncertainty already reveals about its errors.","support":"source-stated"},{"text":"The reported evaluation on the RFMiD test split does not establish performance beyond that evaluation.","excerpt":"On the RFMiD test split, RetinaGuard was compared with confidence and calibrated-uncertainty referral","support":"interpretation"}],"limitations":["The supplied evidence is the paper abstract, which reports results on the RFMiD test split and describes comparisons using two RETFound primary models; it does not establish performance beyond that evaluation."],"mediaId":"dbb7fd695abc8fd4d52f64e521c5fbb2169b049ae05f24bd76fa742a2f22601b","correctionReason":null,"id":"d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e","publicationId":"f933e933-f8e3-4183-8b75-ab396827b93f","version":1,"sourceUrl":"https://arxiv.org/abs/2611.10689","sourceTitle":"RetinaGuard: Risk-Controlled Agentic Triage for Multi-Disease Retinal Screening","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-07","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:39:58.190Z","createdAt":"2026-11-12T06:49:14.861Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"dbb7fd695abc8fd4d52f64e521c5fbb2169b049ae05f24bd76fa742a2f22601b","kind":"article","url":"https://arxiv.org/abs/2611.10689","sourceUrl":"https://arxiv.org/abs/2611.10689","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T06:48:05.346Z","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/2611.10689","checkedAt":"2026-11-12T08:56:10.527Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"dbb7fd695abc8fd4d52f64e521c5fbb2169b049ae05f24bd76fa742a2f22601b","confidence":0.63,"provider":"typesafe","model":"jev-1.13.0","inputHash":"c5cb5f4500a699d7a6352dd41789169356ad461bc658f2d63c0268fb61f2a769","decidedAt":"2026-11-12T06:48:06.869Z"},"provenance":{"textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observationId":"c9c7ffdb52b0a229589758b28ac9c70088e951719bf54d28042386a150a15382","retrievedAt":"2026-11-12T06:48:05.238Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"262c628c-0e53-41eb-8e4a-f95e9ce123e4"},"reviewer":{"model":"gpt-6-luna","runId":"3125a2df-46a9-4773-84c9-2b5e3f06978d"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T06:49:14.861Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Sulman, Sohaib","url":null,"kind":"unknown"},{"name":"Saeed, Muhammad Atif","url":null,"kind":"unknown"},{"name":"Jamil, Akhtar","url":null,"kind":"unknown"},{"name":"Hameed, Alaa Ali","url":null,"kind":"unknown"},{"name":"Aydin, Roland","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Sulman, Sohaib","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sulman, Sohaib","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Saeed, Muhammad Atif","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Saeed, Muhammad Atif","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Jamil, Akhtar","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jamil, Akhtar","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Hameed, Alaa Ali","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Hameed, Alaa Ali","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}},{"actor":{"name":"Aydin, Roland","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Aydin, Roland","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.10689","locator":"head/meta[name=author|citation_author]"},"assertionId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T08:56:10.527Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","sourceUrl":"https://arxiv.org/abs/2611.10689","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.10689","snapshotId":"be2fd16b1ef0204c7dc72aefc4af97889515e2b072c86e7d78778587b802167d","contentSha256":"9958b791ba6e2c867173bf61d502bda80f6c45239c0071064b2de9a0785e701d","textHash":"2cddf348a30101e469c1355ef0c088483c66415d7499ddc84728bfe035daec69","observedAt":"2026-11-12T08:56:10.527Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/d648daae3434a5e3dcfb63d2ca97da15cf294da074383312e34d18e50740e36e/source"}}}],"audits":[{"publicationId":"f933e933-f8e3-4183-8b75-ab396827b93f","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes RetinaGuard, its design and its results to the paper's authors, and keeps the central conclusion framed as their claim. This matches the arXiv abstract page.","excerpt":"This paper presents RetinaGuard, an agentic triage framework"},{"dimension":"summary","status":"uncertain","explanation":"The reading tracks the abstract faithfully and preserves its key qualification: the benefit shrank with the stronger model. But a non-specialist cannot easily follow it. Central terms are left unexplained in the text: fundus photograph (a photo of the back of the eye), RFMiD (a retinal image dataset), RETFound (the primary models), referral budget, conformal risk control and nominal 5% unsafe level. Smallest repair: add plain-language glosses or tooltips for these terms. Optionally restore the abstract's numbers (119 vs 88 of 280; 37.0% vs 10.3%) and its note that coverage stayed near 37.7% in both settings.","excerpt":"in which quality, diagnostic, uncertainty, verification and explanation agents gather evidence for each fundus photograph"},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-07 matches the arXiv v1 submission date. The later feed date was correctly not used.","excerpt":"[Submitted on 7 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is grounded in the abstract. The one interpretive claim and the limitation correctly note that results come only from the RFMiD test split. The question is a genuine open issue and not a leading claim. No benefits, costs or stakeholders are invented.","excerpt":"With the stronger model (0.925) the verifier added nothing beyond uncertainty and the advantage shrank to a few percentage points."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page, which is the original source.","excerpt":null}],"createdAt":"2026-11-12T06:49:14.861Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}