{"id":"8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b","title":"Liu and Yang: Computation, not weight access, enabled these network audits","argument":"Zhiyu Liu and Xinyu Yang report that large language models could audit simulated clinical prediction networks more effectively when given a calculation tool than when asked to inspect network weights in a prompt. Their experiments used tanh networks trained on simulated patient cohorts, including networks with planted relationships, no relationship, or subgroup labels that under-recorded deaths.\n\nWithout tools, GPT-6 Astra recovered planted relationships in smaller networks but became unreliable as networks grew; under a default instruction it also declined most audit and repair tasks as unidentifiable. An explicit instruction removed those refusals. With a calculation-only tool, models could use either white-box access to weights or black-box access to predictions. The authors report that black-box access performed as well as white-box access, and that a well-designed script matched comparator models on the audit and the best LLM repair.\n\nThe authors’ result is about these simulation experiments, not evidence that language models can reliably audit clinical prediction systems in use. It highlights a distinction between supplying a model with information and enabling it to calculate with that information. How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The experiments used simulated cohorts of patients with subarachnoid haemorrhage and simulated networks, so the findings do not establish performance on clinical prediction systems in use. The abstract reports single-prompt tests and specific models and network conditions; it does not establish how results generalize beyond them.","speaker":{"name":"Liu, Zhiyu, Yang, Xinyu","roleAtSource":"Article author"},"source":{"title":"Computation, not weight access: large language models as auditors of clinical prediction networks","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07847","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"Liu and Yang tested language-model audits using simulated patient cohorts and networks with planted relationships, no relationship, or under-recorded subgroup deaths.","excerpt":"Tanh networks of 105 to 38,401 weights were trained on simulated cohorts of patients with subarachnoid haemorrhage, each with a planted relationship or none, and one network was trained on labels that under-recorded deaths in a subgroup.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"Without tools, GPT-6 Astra became unreliable on larger networks and often refused tasks under the default instruction; an explicit system instruction removed those refusals.","excerpt":"Without tools, Astra recovered planted relationships in 353-weight networks but became unreliable at 3,457 weights, and under the default instruction it declined 11 of 12 audit and repair tasks as unidentifiable; the explicit instruction removed these refusals.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"With a calculation-only tool, the models audited networks using either weight access or prediction-only access, with the authors reporting comparable performance.","excerpt":"Black-box access performed as well as white-box access, and other GPT-6 models read as accurately.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"The authors report that a well-designed script matched comparator models on the audit and the best LLM repair.","excerpt":"A well-designed script matched the comparator models on the audit and the best LLM repair.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"The results concern the reported simulation experiments; they do not establish audit reliability on clinical prediction systems in use.","excerpt":"internally pre-specified simulation experiments. Tanh networks of 105 to 38,401 weights were trained on simulated cohorts","url":"https://arxiv.org/abs/2611.07847","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07847","checkedAt":"2026-11-09T00:56:10.127Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Liu and Yang: Computation, not weight access, enabled these network audits","summary":"Zhiyu Liu and Xinyu Yang report that large language models could audit simulated clinical prediction networks more effectively when given a calculation tool than when asked to inspect network weights in a prompt. Their experiments used tanh networks trained on simulated patient cohorts, including networks with planted relationships, no relationship, or subgroup labels that under-recorded deaths.\n\nWithout tools, GPT-6 Astra recovered planted relationships in smaller networks but became unreliable as networks grew; under a default instruction it also declined most audit and repair tasks as unidentifiable. An explicit instruction removed those refusals. With a calculation-only tool, models could use either white-box access to weights or black-box access to predictions. The authors report that black-box access performed as well as white-box access, and that a well-designed script matched comparator models on the audit and the best LLM repair.\n\nThe authors’ result is about these simulation experiments, not evidence that language models can reliably audit clinical prediction systems in use. It highlights a distinction between supplying a model with information and enabling it to calculate with that information. How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?","question":"How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?","claims":[{"text":"Liu and Yang tested language-model audits using simulated patient cohorts and networks with planted relationships, no relationship, or under-recorded subgroup deaths.","excerpt":"Tanh networks of 105 to 38,401 weights were trained on simulated cohorts of patients with subarachnoid haemorrhage, each with a planted relationship or none, and one network was trained on labels that under-recorded deaths in a subgroup.","support":"source-stated"},{"text":"Without tools, GPT-6 Astra became unreliable on larger networks and often refused tasks under the default instruction; an explicit system instruction removed those refusals.","excerpt":"Without tools, Astra recovered planted relationships in 353-weight networks but became unreliable at 3,457 weights, and under the default instruction it declined 11 of 12 audit and repair tasks as unidentifiable; the explicit instruction removed these refusals.","support":"source-stated"},{"text":"With a calculation-only tool, the models audited networks using either weight access or prediction-only access, with the authors reporting comparable performance.","excerpt":"Black-box access performed as well as white-box access, and other GPT-6 models read as accurately.","support":"source-stated"},{"text":"The authors report that a well-designed script matched comparator models on the audit and the best LLM repair.","excerpt":"A well-designed script matched the comparator models on the audit and the best LLM repair.","support":"source-stated"},{"text":"The results concern the reported simulation experiments; they do not establish audit reliability on clinical prediction systems in use.","excerpt":"internally pre-specified simulation experiments. Tanh networks of 105 to 38,401 weights were trained on simulated cohorts","support":"interpretation"}],"limitations":["The experiments used simulated cohorts of patients with subarachnoid haemorrhage and simulated networks, so the findings do not establish performance on clinical prediction systems in use.","The abstract reports single-prompt tests and specific models and network conditions; it does not establish how results generalize beyond them."],"mediaId":"64d5edb3b57e8b40fdf40a5692aaa6299d1e5803ab781f8fa3dada5189249672","correctionReason":null,"id":"8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b","publicationId":"e13211aa-4d2d-4db5-8650-2146a4375b77","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07847","sourceTitle":"Computation, not weight access: large language models as auditors of clinical prediction networks","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:10:56.749Z","createdAt":"2026-11-09T00:24:11.749Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"64d5edb3b57e8b40fdf40a5692aaa6299d1e5803ab781f8fa3dada5189249672","kind":"article","url":"https://arxiv.org/abs/2611.07847","sourceUrl":"https://arxiv.org/abs/2611.07847","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T00:23:27.000Z","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/2611.07847","checkedAt":"2026-11-09T00:56:10.127Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"64d5edb3b57e8b40fdf40a5692aaa6299d1e5803ab781f8fa3dada5189249672","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"92d46e65a2ea9e719cf944f93ab620e5b61f5bd5c9e751827471b45b605eeef9","decidedAt":"2026-11-09T00:23:27.826Z"},"provenance":{"textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observationId":"c53e5fc26bf3fd2606b1fa3a92cce8c19f1adbe5881b83550f234efef6dcd375","retrievedAt":"2026-11-09T00:23:26.876Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"e3ad91e4-7b11-4b3c-83d2-b183eebc3f3f"},"reviewer":{"model":"gpt-6-luna","runId":"b84a1f9c-a69a-4501-8814-edc534dec904"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T00:24:11.749Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Liu, Zhiyu","url":null,"kind":"unknown"},{"name":"Yang, Xinyu","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Liu, Zhiyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Zhiyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07847","locator":"head/meta[name=author|citation_author]"},"assertionId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T00:56:10.127Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb","sourceUrl":"https://arxiv.org/abs/2611.07847","contentSha256":"c8e57fe8d3e31611d7b682820472d43e52aaa49c24ec40af515eee3de57cd8d6","textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observedAt":"2026-11-09T00:56:10.127Z"}},{"actor":{"name":"Yang, Xinyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yang, Xinyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07847","locator":"head/meta[name=author|citation_author]"},"assertionId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T00:56:10.127Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb","sourceUrl":"https://arxiv.org/abs/2611.07847","contentSha256":"c8e57fe8d3e31611d7b682820472d43e52aaa49c24ec40af515eee3de57cd8d6","textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observedAt":"2026-11-09T00:56:10.127Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07847","snapshotId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb","contentSha256":"c8e57fe8d3e31611d7b682820472d43e52aaa49c24ec40af515eee3de57cd8d6","textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observedAt":"2026-11-09T00:56:10.127Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b/source"}},"canonicalUrl":"https://escapemeridian.com/readings/liu-and-yang-computation-not-weight-access-enabled-these-8a0121e5","formats":{"html":"/readings/liu-and-yang-computation-not-weight-access-enabled-these-8a0121e5","markdown":"/readings/liu-and-yang-computation-not-weight-access-enabled-these-8a0121e5.md","json":"/readings/liu-and-yang-computation-not-weight-access-enabled-these-8a0121e5.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-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"}],"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":"8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b","revision":"e13211aa-4d2d-4db5-8650-2146a4375b77","revisionKind":"publication-id","title":"Liu and Yang: Computation, not weight access, enabled these network audits","url":"https://escapemeridian.com/readings/liu-and-yang-computation-not-weight-access-enabled-these-8a0121e5"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.07847","revision":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","revisionKind":"sha256-source-text","title":"Computation, not weight access: large language models as auditors of clinical prediction networks","url":"https://arxiv.org/abs/2611.07847"},"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":"8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b","title":"Liu and Yang: Computation, not weight access, enabled these network audits","argument":"Zhiyu Liu and Xinyu Yang report that large language models could audit simulated clinical prediction networks more effectively when given a calculation tool than when asked to inspect network weights in a prompt. Their experiments used tanh networks trained on simulated patient cohorts, including networks with planted relationships, no relationship, or subgroup labels that under-recorded deaths.\n\nWithout tools, GPT-6 Astra recovered planted relationships in smaller networks but became unreliable as networks grew; under a default instruction it also declined most audit and repair tasks as unidentifiable. An explicit instruction removed those refusals. With a calculation-only tool, models could use either white-box access to weights or black-box access to predictions. The authors report that black-box access performed as well as white-box access, and that a well-designed script matched comparator models on the audit and the best LLM repair.\n\nThe authors’ result is about these simulation experiments, not evidence that language models can reliably audit clinical prediction systems in use. It highlights a distinction between supplying a model with information and enabling it to calculate with that information. How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The experiments used simulated cohorts of patients with subarachnoid haemorrhage and simulated networks, so the findings do not establish performance on clinical prediction systems in use. The abstract reports single-prompt tests and specific models and network conditions; it does not establish how results generalize beyond them.","speaker":{"name":"Liu, Zhiyu, Yang, Xinyu","roleAtSource":"Article author"},"source":{"title":"Computation, not weight access: large language models as auditors of clinical prediction networks","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07847","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"Liu and Yang tested language-model audits using simulated patient cohorts and networks with planted relationships, no relationship, or under-recorded subgroup deaths.","excerpt":"Tanh networks of 105 to 38,401 weights were trained on simulated cohorts of patients with subarachnoid haemorrhage, each with a planted relationship or none, and one network was trained on labels that under-recorded deaths in a subgroup.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"Without tools, GPT-6 Astra became unreliable on larger networks and often refused tasks under the default instruction; an explicit system instruction removed those refusals.","excerpt":"Without tools, Astra recovered planted relationships in 353-weight networks but became unreliable at 3,457 weights, and under the default instruction it declined 11 of 12 audit and repair tasks as unidentifiable; the explicit instruction removed these refusals.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"With a calculation-only tool, the models audited networks using either weight access or prediction-only access, with the authors reporting comparable performance.","excerpt":"Black-box access performed as well as white-box access, and other GPT-6 models read as accurately.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"The authors report that a well-designed script matched comparator models on the audit and the best LLM repair.","excerpt":"A well-designed script matched the comparator models on the audit and the best LLM repair.","url":"https://arxiv.org/abs/2611.07847","relationship":"Source-stated"},{"text":"The results concern the reported simulation experiments; they do not establish audit reliability on clinical prediction systems in use.","excerpt":"internally pre-specified simulation experiments. Tanh networks of 105 to 38,401 weights were trained on simulated cohorts","url":"https://arxiv.org/abs/2611.07847","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07847","checkedAt":"2026-11-09T00:56:10.127Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Liu and Yang: Computation, not weight access, enabled these network audits","summary":"Zhiyu Liu and Xinyu Yang report that large language models could audit simulated clinical prediction networks more effectively when given a calculation tool than when asked to inspect network weights in a prompt. Their experiments used tanh networks trained on simulated patient cohorts, including networks with planted relationships, no relationship, or subgroup labels that under-recorded deaths.\n\nWithout tools, GPT-6 Astra recovered planted relationships in smaller networks but became unreliable as networks grew; under a default instruction it also declined most audit and repair tasks as unidentifiable. An explicit instruction removed those refusals. With a calculation-only tool, models could use either white-box access to weights or black-box access to predictions. The authors report that black-box access performed as well as white-box access, and that a well-designed script matched comparator models on the audit and the best LLM repair.\n\nThe authors’ result is about these simulation experiments, not evidence that language models can reliably audit clinical prediction systems in use. It highlights a distinction between supplying a model with information and enabling it to calculate with that information. How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?","question":"How well would the reported audit results hold on clinical prediction networks beyond the simulated cohorts and network conditions tested here?","claims":[{"text":"Liu and Yang tested language-model audits using simulated patient cohorts and networks with planted relationships, no relationship, or under-recorded subgroup deaths.","excerpt":"Tanh networks of 105 to 38,401 weights were trained on simulated cohorts of patients with subarachnoid haemorrhage, each with a planted relationship or none, and one network was trained on labels that under-recorded deaths in a subgroup.","support":"source-stated"},{"text":"Without tools, GPT-6 Astra became unreliable on larger networks and often refused tasks under the default instruction; an explicit system instruction removed those refusals.","excerpt":"Without tools, Astra recovered planted relationships in 353-weight networks but became unreliable at 3,457 weights, and under the default instruction it declined 11 of 12 audit and repair tasks as unidentifiable; the explicit instruction removed these refusals.","support":"source-stated"},{"text":"With a calculation-only tool, the models audited networks using either weight access or prediction-only access, with the authors reporting comparable performance.","excerpt":"Black-box access performed as well as white-box access, and other GPT-6 models read as accurately.","support":"source-stated"},{"text":"The authors report that a well-designed script matched comparator models on the audit and the best LLM repair.","excerpt":"A well-designed script matched the comparator models on the audit and the best LLM repair.","support":"source-stated"},{"text":"The results concern the reported simulation experiments; they do not establish audit reliability on clinical prediction systems in use.","excerpt":"internally pre-specified simulation experiments. Tanh networks of 105 to 38,401 weights were trained on simulated cohorts","support":"interpretation"}],"limitations":["The experiments used simulated cohorts of patients with subarachnoid haemorrhage and simulated networks, so the findings do not establish performance on clinical prediction systems in use.","The abstract reports single-prompt tests and specific models and network conditions; it does not establish how results generalize beyond them."],"mediaId":"64d5edb3b57e8b40fdf40a5692aaa6299d1e5803ab781f8fa3dada5189249672","correctionReason":null,"id":"8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b","publicationId":"e13211aa-4d2d-4db5-8650-2146a4375b77","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07847","sourceTitle":"Computation, not weight access: large language models as auditors of clinical prediction networks","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:10:56.749Z","createdAt":"2026-11-09T00:24:11.749Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"64d5edb3b57e8b40fdf40a5692aaa6299d1e5803ab781f8fa3dada5189249672","kind":"article","url":"https://arxiv.org/abs/2611.07847","sourceUrl":"https://arxiv.org/abs/2611.07847","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T00:23:27.000Z","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/2611.07847","checkedAt":"2026-11-09T00:56:10.127Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"64d5edb3b57e8b40fdf40a5692aaa6299d1e5803ab781f8fa3dada5189249672","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"92d46e65a2ea9e719cf944f93ab620e5b61f5bd5c9e751827471b45b605eeef9","decidedAt":"2026-11-09T00:23:27.826Z"},"provenance":{"textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observationId":"c53e5fc26bf3fd2606b1fa3a92cce8c19f1adbe5881b83550f234efef6dcd375","retrievedAt":"2026-11-09T00:23:26.876Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"e3ad91e4-7b11-4b3c-83d2-b183eebc3f3f"},"reviewer":{"model":"gpt-6-luna","runId":"b84a1f9c-a69a-4501-8814-edc534dec904"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T00:24:11.749Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Liu, Zhiyu","url":null,"kind":"unknown"},{"name":"Yang, Xinyu","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Liu, Zhiyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Zhiyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07847","locator":"head/meta[name=author|citation_author]"},"assertionId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T00:56:10.127Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb","sourceUrl":"https://arxiv.org/abs/2611.07847","contentSha256":"c8e57fe8d3e31611d7b682820472d43e52aaa49c24ec40af515eee3de57cd8d6","textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observedAt":"2026-11-09T00:56:10.127Z"}},{"actor":{"name":"Yang, Xinyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yang, Xinyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07847","locator":"head/meta[name=author|citation_author]"},"assertionId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T00:56:10.127Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb","sourceUrl":"https://arxiv.org/abs/2611.07847","contentSha256":"c8e57fe8d3e31611d7b682820472d43e52aaa49c24ec40af515eee3de57cd8d6","textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observedAt":"2026-11-09T00:56:10.127Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07847","snapshotId":"ab6ee50aa6a2c1d9df5f83c18f7726c836402246965dd8dde0eecc7ae3943eeb","contentSha256":"c8e57fe8d3e31611d7b682820472d43e52aaa49c24ec40af515eee3de57cd8d6","textHash":"389ad0964e6a307898749ab4cbc5b8c78b1a8758e9beee8912bbf695d3fa3439","observedAt":"2026-11-09T00:56:10.127Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/8a0121e57121b7dd3869bef7ec17f6aa72d325f16c0c3622be61175d11fddc2b/source"}}}],"audits":[{"publicationId":"e13211aa-4d2d-4db5-8650-2146a4375b77","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the work and its conclusions to Zhiyu Liu and Xinyu Yang, the listed authors. It keeps reported results as the authors' findings rather than established fact.","excerpt":"Authors:Zhiyu Liu, Xinyu Yang"},{"dimension":"summary","status":"uncertain","explanation":"The reading leaves out the headline tool-assisted results: all 29 runs up to 38,401 weights recovered the planted relationship, there were no false alarms in 15 null runs, and the biased subgroup was found in 6 of 6 runs. It also drops the authors' conclusion that elicitation mattered alongside computation. Central terms such as weights, tanh networks and planted relationship are not explained in plain language. Smallest repair: add the with-tool results, keep 'computation and elicitation', and briefly gloss these terms.","excerpt":"For auditing these networks, what mattered was computation and elicitation, not access to the weights."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-05 matches the v1 submission date on arXiv.","excerpt":"[Submitted on 5 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is grounded in the abstract. The limitation to simulated settings is a careful, labeled interpretation that follows from the stated simulation design. The question does not assume an answer.","excerpt":"We separated these conditions in internally pre-specified simulation experiments."},{"dimension":"media","status":"pass","explanation":"The selected media is the verified arXiv abstract page, which is the direct original source.","excerpt":null}],"createdAt":"2026-11-09T00:24:11.749Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}