{"id":"8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f","title":"Raviez, Wang and Delaney measure how far back an EHR model looks for colorectal cancer signals","argument":"In their arXiv paper, Achille Raviez, Lan Wang and Brendan Delaney argue that clinical prediction models should report time-context: the calendar lookback period they use, distinct from how much text they can process. They test a model trained on primary-care records from North West London, comparing 15 lookback windows in the same cohort of cases and matched controls. Its AUROC, a measure of discrimination, rose from 0.551 at 2 years to 0.590 at 18 years, then plateaued; about 80% of the gain came within 10 years. An XGBoost baseline reached a similar peak. The authors also report a simulation suggesting that rotary position embeddings (RoPE) over elapsed time can alias beyond about 20 years at the default base frequency, and say this motivates retuning θ. Their practical argument is that the calendar span of records is a distinct design choice, not captured by token-context alone, and should be reported. The results describe this model, task and evaluated cohort; they do not establish how the pattern would hold in other patient populations or in clinical use.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports an evaluation on a frozen cohort of cases and matched controls registered for at least 30 years; the reading does not establish how results generalize to other patient populations. The abstract reports model evaluation and a simulation, not evidence of clinical deployment or improved patient outcomes.","speaker":{"name":"Raviez, Achille, Wang, Lan, Delaney, Brendan","roleAtSource":"Article author"},"source":{"title":"How Deep is the Signal? Time-Context up to 30 Years for Colorectal Cancer Prediction in a Primary-Care EHR Model","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07265","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors define time-context as the calendar lookback window before a prediction point and distinguish it from token-context.","excerpt":"We introduce time-context, the calendar lookback window before the prediction point","url":"https://arxiv.org/abs/2611.07265","relationship":"Source-stated"},{"text":"In the evaluated cohort, discrimination rose through an 18-year lookback and then plateaued, with about 80% of the gain reached within 10 years.","excerpt":"Discrimination rose from an AUROC of 0.551 at 2 years to 0.590 at 18 years, then plateaued, with about 80% of the gain reached within 10 years","url":"https://arxiv.org/abs/2611.07265","relationship":"Source-stated"},{"text":"The authors say their simulation suggests rotary position embeddings over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned θ.","excerpt":"A simulation further suggests that rotary position embeddings (RoPE) over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned $\\theta$.","url":"https://arxiv.org/abs/2611.07265","relationship":"Source-stated"},{"text":"Because the evaluation used patients from one primary-care region, whether the reported pattern holds in other populations remains an open question.","excerpt":"on 1.4B coded events from 3.6M primary-care patients in North West London","url":"https://arxiv.org/abs/2611.07265","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07265","checkedAt":"2026-11-09T13:12:06.812Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would the same time-context pattern appear when the model is evaluated on patients outside the North West London cohort?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Raviez, Wang and Delaney measure how far back an EHR model looks for colorectal cancer signals","summary":"In their arXiv paper, Achille Raviez, Lan Wang and Brendan Delaney argue that clinical prediction models should report time-context: the calendar lookback period they use, distinct from how much text they can process. They test a model trained on primary-care records from North West London, comparing 15 lookback windows in the same cohort of cases and matched controls. Its AUROC, a measure of discrimination, rose from 0.551 at 2 years to 0.590 at 18 years, then plateaued; about 80% of the gain came within 10 years. An XGBoost baseline reached a similar peak. The authors also report a simulation suggesting that rotary position embeddings (RoPE) over elapsed time can alias beyond about 20 years at the default base frequency, and say this motivates retuning θ. Their practical argument is that the calendar span of records is a distinct design choice, not captured by token-context alone, and should be reported. The results describe this model, task and evaluated cohort; they do not establish how the pattern would hold in other patient populations or in clinical use.","question":"Would the same time-context pattern appear when the model is evaluated on patients outside the North West London cohort?","claims":[{"text":"The authors define time-context as the calendar lookback window before a prediction point and distinguish it from token-context.","excerpt":"We introduce time-context, the calendar lookback window before the prediction point","support":"source-stated"},{"text":"In the evaluated cohort, discrimination rose through an 18-year lookback and then plateaued, with about 80% of the gain reached within 10 years.","excerpt":"Discrimination rose from an AUROC of 0.551 at 2 years to 0.590 at 18 years, then plateaued, with about 80% of the gain reached within 10 years","support":"source-stated"},{"text":"The authors say their simulation suggests rotary position embeddings over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned θ.","excerpt":"A simulation further suggests that rotary position embeddings (RoPE) over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned $\\theta$.","support":"source-stated"},{"text":"Because the evaluation used patients from one primary-care region, whether the reported pattern holds in other populations remains an open question.","excerpt":"on 1.4B coded events from 3.6M primary-care patients in North West London","support":"interpretation"}],"limitations":["The abstract reports an evaluation on a frozen cohort of cases and matched controls registered for at least 30 years; the reading does not establish how results generalize to other patient populations.","The abstract reports model evaluation and a simulation, not evidence of clinical deployment or improved patient outcomes."],"mediaId":"9c15a1fafc98c83114accea70cbd8c77ac0827028d882ce6d067be385c133292","correctionReason":null,"id":"8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f","publicationId":"bedf6e32-6121-4e72-8f3b-695c53662053","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07265","sourceTitle":"How Deep is the Signal? Time-Context up to 30 Years for Colorectal Cancer Prediction in a Primary-Care EHR Model","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:12:36.271Z","createdAt":"2026-11-08T20:09:43.621Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"9c15a1fafc98c83114accea70cbd8c77ac0827028d882ce6d067be385c133292","kind":"article","url":"https://arxiv.org/abs/2611.07265","sourceUrl":"https://arxiv.org/abs/2611.07265","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T20:08:42.514Z","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/2611.07265","checkedAt":"2026-11-09T13:12:06.812Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"9c15a1fafc98c83114accea70cbd8c77ac0827028d882ce6d067be385c133292","confidence":0.77,"provider":"typesafe","model":"jev-1.13.0","inputHash":"fb55730bf11048e4f42689fc5607957e07ca73a8e4902ba829be9f3e31a14e3a","decidedAt":"2026-11-08T20:08:43.981Z"},"provenance":{"textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observationId":"e4fa53256a2baaecdc89ac742b3b599e8b3ee746c5f23209a9964bd64e47ef6a","retrievedAt":"2026-11-08T20:08:42.318Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"ed128d89-5584-42a0-8924-7ee8765d891d"},"reviewer":{"model":"gpt-6-luna","runId":"d8efb2ce-3f72-4da0-8d30-b5ecff8e7217"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T20:09:43.621Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Raviez, Achille","url":null,"kind":"unknown"},{"name":"Wang, Lan","url":null,"kind":"unknown"},{"name":"Delaney, Brendan","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Raviez, Achille","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Raviez, Achille","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07265","locator":"head/meta[name=author|citation_author]"},"assertionId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T13:12:06.812Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","sourceUrl":"https://arxiv.org/abs/2611.07265","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z"}},{"actor":{"name":"Wang, Lan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Lan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07265","locator":"head/meta[name=author|citation_author]"},"assertionId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T13:12:06.812Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","sourceUrl":"https://arxiv.org/abs/2611.07265","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z"}},{"actor":{"name":"Delaney, Brendan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Delaney, Brendan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07265","locator":"head/meta[name=author|citation_author]"},"assertionId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T13:12:06.812Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","sourceUrl":"https://arxiv.org/abs/2611.07265","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07265","snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f/source"}},"canonicalUrl":"https://escapemeridian.com/readings/raviez-wang-and-delaney-measure-how-far-back-an-ehr-model-looks-8fbae831","formats":{"html":"/readings/raviez-wang-and-delaney-measure-how-far-back-an-ehr-model-looks-8fbae831","markdown":"/readings/raviez-wang-and-delaney-measure-how-far-back-an-ehr-model-looks-8fbae831.md","json":"/readings/raviez-wang-and-delaney-measure-how-far-back-an-ehr-model-looks-8fbae831.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-embedding-vector","name":"Embedding vector","explanation":"A list of numbers that represents a piece of text, an image or another item so software can compare meanings.","referenceUrl":"https://developers.google.com/machine-learning/glossary#embedding-vector"}],"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"},{"id":"time","name":"TIME","kind":"organization","context":"TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.","referenceUrls":["https://time.com/"],"path":"/entities/time","json":"/entities/time.json","markdown":"/entities/time.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f","revision":"bedf6e32-6121-4e72-8f3b-695c53662053","revisionKind":"publication-id","title":"Raviez, Wang and Delaney measure how far back an EHR model looks for colorectal cancer signals","url":"https://escapemeridian.com/readings/raviez-wang-and-delaney-measure-how-far-back-an-ehr-model-looks-8fbae831"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.07265","revision":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","revisionKind":"sha256-source-text","title":"How Deep is the Signal? Time-Context up to 30 Years for Colorectal Cancer Prediction in a Primary-Care EHR Model","url":"https://arxiv.org/abs/2611.07265"},"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":"8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f","title":"Raviez, Wang and Delaney measure how far back an EHR model looks for colorectal cancer signals","argument":"In their arXiv paper, Achille Raviez, Lan Wang and Brendan Delaney argue that clinical prediction models should report time-context: the calendar lookback period they use, distinct from how much text they can process. They test a model trained on primary-care records from North West London, comparing 15 lookback windows in the same cohort of cases and matched controls. Its AUROC, a measure of discrimination, rose from 0.551 at 2 years to 0.590 at 18 years, then plateaued; about 80% of the gain came within 10 years. An XGBoost baseline reached a similar peak. The authors also report a simulation suggesting that rotary position embeddings (RoPE) over elapsed time can alias beyond about 20 years at the default base frequency, and say this motivates retuning θ. Their practical argument is that the calendar span of records is a distinct design choice, not captured by token-context alone, and should be reported. The results describe this model, task and evaluated cohort; they do not establish how the pattern would hold in other patient populations or in clinical use.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports an evaluation on a frozen cohort of cases and matched controls registered for at least 30 years; the reading does not establish how results generalize to other patient populations. The abstract reports model evaluation and a simulation, not evidence of clinical deployment or improved patient outcomes.","speaker":{"name":"Raviez, Achille, Wang, Lan, Delaney, Brendan","roleAtSource":"Article author"},"source":{"title":"How Deep is the Signal? Time-Context up to 30 Years for Colorectal Cancer Prediction in a Primary-Care EHR Model","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07265","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors define time-context as the calendar lookback window before a prediction point and distinguish it from token-context.","excerpt":"We introduce time-context, the calendar lookback window before the prediction point","url":"https://arxiv.org/abs/2611.07265","relationship":"Source-stated"},{"text":"In the evaluated cohort, discrimination rose through an 18-year lookback and then plateaued, with about 80% of the gain reached within 10 years.","excerpt":"Discrimination rose from an AUROC of 0.551 at 2 years to 0.590 at 18 years, then plateaued, with about 80% of the gain reached within 10 years","url":"https://arxiv.org/abs/2611.07265","relationship":"Source-stated"},{"text":"The authors say their simulation suggests rotary position embeddings over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned θ.","excerpt":"A simulation further suggests that rotary position embeddings (RoPE) over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned $\\theta$.","url":"https://arxiv.org/abs/2611.07265","relationship":"Source-stated"},{"text":"Because the evaluation used patients from one primary-care region, whether the reported pattern holds in other populations remains an open question.","excerpt":"on 1.4B coded events from 3.6M primary-care patients in North West London","url":"https://arxiv.org/abs/2611.07265","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07265","checkedAt":"2026-11-09T13:12:06.812Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would the same time-context pattern appear when the model is evaluated on patients outside the North West London cohort?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Raviez, Wang and Delaney measure how far back an EHR model looks for colorectal cancer signals","summary":"In their arXiv paper, Achille Raviez, Lan Wang and Brendan Delaney argue that clinical prediction models should report time-context: the calendar lookback period they use, distinct from how much text they can process. They test a model trained on primary-care records from North West London, comparing 15 lookback windows in the same cohort of cases and matched controls. Its AUROC, a measure of discrimination, rose from 0.551 at 2 years to 0.590 at 18 years, then plateaued; about 80% of the gain came within 10 years. An XGBoost baseline reached a similar peak. The authors also report a simulation suggesting that rotary position embeddings (RoPE) over elapsed time can alias beyond about 20 years at the default base frequency, and say this motivates retuning θ. Their practical argument is that the calendar span of records is a distinct design choice, not captured by token-context alone, and should be reported. The results describe this model, task and evaluated cohort; they do not establish how the pattern would hold in other patient populations or in clinical use.","question":"Would the same time-context pattern appear when the model is evaluated on patients outside the North West London cohort?","claims":[{"text":"The authors define time-context as the calendar lookback window before a prediction point and distinguish it from token-context.","excerpt":"We introduce time-context, the calendar lookback window before the prediction point","support":"source-stated"},{"text":"In the evaluated cohort, discrimination rose through an 18-year lookback and then plateaued, with about 80% of the gain reached within 10 years.","excerpt":"Discrimination rose from an AUROC of 0.551 at 2 years to 0.590 at 18 years, then plateaued, with about 80% of the gain reached within 10 years","support":"source-stated"},{"text":"The authors say their simulation suggests rotary position embeddings over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned θ.","excerpt":"A simulation further suggests that rotary position embeddings (RoPE) over elapsed time alias beyond about 20 years at the default base frequency, motivating a retuned $\\theta$.","support":"source-stated"},{"text":"Because the evaluation used patients from one primary-care region, whether the reported pattern holds in other populations remains an open question.","excerpt":"on 1.4B coded events from 3.6M primary-care patients in North West London","support":"interpretation"}],"limitations":["The abstract reports an evaluation on a frozen cohort of cases and matched controls registered for at least 30 years; the reading does not establish how results generalize to other patient populations.","The abstract reports model evaluation and a simulation, not evidence of clinical deployment or improved patient outcomes."],"mediaId":"9c15a1fafc98c83114accea70cbd8c77ac0827028d882ce6d067be385c133292","correctionReason":null,"id":"8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f","publicationId":"bedf6e32-6121-4e72-8f3b-695c53662053","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07265","sourceTitle":"How Deep is the Signal? Time-Context up to 30 Years for Colorectal Cancer Prediction in a Primary-Care EHR Model","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:12:36.271Z","createdAt":"2026-11-08T20:09:43.621Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"9c15a1fafc98c83114accea70cbd8c77ac0827028d882ce6d067be385c133292","kind":"article","url":"https://arxiv.org/abs/2611.07265","sourceUrl":"https://arxiv.org/abs/2611.07265","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-08T20:08:42.514Z","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/2611.07265","checkedAt":"2026-11-09T13:12:06.812Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"9c15a1fafc98c83114accea70cbd8c77ac0827028d882ce6d067be385c133292","confidence":0.77,"provider":"typesafe","model":"jev-1.13.0","inputHash":"fb55730bf11048e4f42689fc5607957e07ca73a8e4902ba829be9f3e31a14e3a","decidedAt":"2026-11-08T20:08:43.981Z"},"provenance":{"textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observationId":"e4fa53256a2baaecdc89ac742b3b599e8b3ee746c5f23209a9964bd64e47ef6a","retrievedAt":"2026-11-08T20:08:42.318Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"ed128d89-5584-42a0-8924-7ee8765d891d"},"reviewer":{"model":"gpt-6-luna","runId":"d8efb2ce-3f72-4da0-8d30-b5ecff8e7217"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-08T20:09:43.621Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Raviez, Achille","url":null,"kind":"unknown"},{"name":"Wang, Lan","url":null,"kind":"unknown"},{"name":"Delaney, Brendan","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Raviez, Achille","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Raviez, Achille","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07265","locator":"head/meta[name=author|citation_author]"},"assertionId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T13:12:06.812Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","sourceUrl":"https://arxiv.org/abs/2611.07265","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z"}},{"actor":{"name":"Wang, Lan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Lan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07265","locator":"head/meta[name=author|citation_author]"},"assertionId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T13:12:06.812Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","sourceUrl":"https://arxiv.org/abs/2611.07265","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z"}},{"actor":{"name":"Delaney, Brendan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Delaney, Brendan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07265","locator":"head/meta[name=author|citation_author]"},"assertionId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T13:12:06.812Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","sourceUrl":"https://arxiv.org/abs/2611.07265","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z"}}],"participationCoverage":{"total":3,"returned":3,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07265","snapshotId":"5a3aef6013004525c0c70c1e52e98982dd40a4ea3e6b4bccbfdc1030e1e6515a","contentSha256":"c52640551a0d4127147a3778c150471a0e096d7da0e940dc2fc05c1b08ac0158","textHash":"880540211b18a12ebf98d8e52439306eb51d95113d4396839e26bc5fb659928e","observedAt":"2026-11-09T13:12:06.812Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/8fbae831588155ed1b151c5e594921eece31db16b01defe804bfc8eb739f186f/source"}}}],"audits":[{"publicationId":"bedf6e32-6121-4e72-8f3b-695c53662053","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly credits the arXiv paper and its argument to Achille Raviez, Lan Wang and Brendan Delaney. The findings and the recommendation to report time-context stay attributed to the authors.","excerpt":"Authors:Achille Raviez, Lan Wang, Brendan Delaney"},{"dimension":"summary","status":"uncertain","explanation":"The reading is mostly faithful, but it defines token-context as \"how much text they can process.\" These EHR models read coded medical events, not text. That paraphrase misstates the contrast at the heart of the paper. Smallest repair: \"how many coded events (tokens) they can process.\" EHR is never expanded, and the summary body never says the prediction target is incident colorectal cancer or mentions the 12-month pre-diagnostic buffer. Terms like XGBoost, alias and θ are kept but unexplained; they need term-library tooltips.","excerpt":"EHR foundation models read medical events within a fixed context window, and long context has so far been measured only in tokens."},{"dimension":"date","status":"pass","explanation":"originalPublished is 2026-11-05, which matches the v1 submission date. The feed date of 2026-11-06 is correctly not used.","excerpt":"[Submitted on 5 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"The source-stated claims match the abstract: the AUROC values, the plateau, the 80% gain within 10 years, the XGBoost peak of 0.589, and the RoPE simulation. The generalization question is labeled as interpretation and kept conditional. The limitations correctly note that the work shows no clinical deployment or patient-outcome evidence.","excerpt":"an XGBoost baseline converged to the same level (peak AUROC 0.589)"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified arXiv abstract page for this exact paper. It is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-08T20:09:43.621Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}