{"id":"94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164","title":"arXiv cs.AI: Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","argument":"Manu Agrawal and Shrey Nagpal argue that tool-using agents can be operationally blind: their host may know whether a tool is healthy while the model choosing an action does not. Their proposed Agent Harness Awareness design gives the model a small, fresh, verified snapshot before consequential actions, while the host retains measurement, enforcement, evaluation and final dispatch authority.\n\nToolRoute holds the task and candidate semantics constant while varying operational information. In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a neutral control and raised single-dispatch completion from 69.4% to 95.8%. A separate 27-decision live-execution replication reported lower mean regret. It completed the action in 9/9 model-by-fault cases, compared with 5/9 for the neutral control.\n\nThe authors argue these studies establish the tools-and-network instance: selected host telemetry can improve pre-dispatch operational choice. The results are bounded to these studies. Runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection remain open empirical extensions.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported evidence concerns a tools-and-network instance tested in a 216-decision ablation and a separate 27-decision end-to-end replication; the source identifies other areas as open empirical extensions. The abstract reports selected outcome measures, but does not establish how the results generalize beyond the described studies.","speaker":{"name":"Agrawal, Manu, Nagpal, Shrey","roleAtSource":"Article author"},"source":{"title":"Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.05391","publishedAt":"2026-09-13","dateBasis":"Date basis not provided."},"evidence":[{"text":"Agrawal and Nagpal describe operational blindness as a gap between the host's operational information and what the model choosing the next action knows.","excerpt":"We call this gap operational blindness.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"The proposed design gives the model a small, fresh, verified state snapshot before consequential actions, while the host retains control over measurement, enforcement, evaluation and final dispatch.","excerpt":"the host supplies a small, fresh, verified state snapshot before a consequential action while retaining measurement, enforcement, evaluation, and final dispatch authority.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a same-shape neutral control and raised single-dispatch completion from 69.4% to 95.8%.","excerpt":"In a 216-decision ablation across GPT-5.6 Sol, Claude Sonnet 5, and Gemini 3.7 Flash, verified tool-health state reduced mean observable policy regret by 96.9%","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"In a separate 27-decision live-execution replication, the verified-state condition completed the live action in 9/9 model-by-fault cases, versus 5/9 for the neutral control.","excerpt":"completed the live action in 9/9 model-by-fault cases versus 5/9 for the neutral control.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"The authors say the studies establish the tools-and-network instance of Agent Harness Awareness, while listing several areas as open empirical extensions.","excerpt":"Runtime state, hardware pressure, economics, retry-aware recovery, and heterogeneous selection remain open empirical extensions of the same architecture.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.05391","checkedAt":"2026-11-09T07:40:08.663Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would Agent Harness Awareness perform for the authors’ listed open extensions, including runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","summary":"Manu Agrawal and Shrey Nagpal argue that tool-using agents can be operationally blind: their host may know whether a tool is healthy while the model choosing an action does not. Their proposed Agent Harness Awareness design gives the model a small, fresh, verified snapshot before consequential actions, while the host retains measurement, enforcement, evaluation and final dispatch authority.\n\nToolRoute holds the task and candidate semantics constant while varying operational information. In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a neutral control and raised single-dispatch completion from 69.4% to 95.8%. A separate 27-decision live-execution replication reported lower mean regret. It completed the action in 9/9 model-by-fault cases, compared with 5/9 for the neutral control.\n\nThe authors argue these studies establish the tools-and-network instance: selected host telemetry can improve pre-dispatch operational choice. The results are bounded to these studies. Runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection remain open empirical extensions.","question":"How would Agent Harness Awareness perform for the authors’ listed open extensions, including runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection?","claims":[{"text":"Agrawal and Nagpal describe operational blindness as a gap between the host's operational information and what the model choosing the next action knows.","excerpt":"We call this gap operational blindness.","support":"source-stated"},{"text":"The proposed design gives the model a small, fresh, verified state snapshot before consequential actions, while the host retains control over measurement, enforcement, evaluation and final dispatch.","excerpt":"the host supplies a small, fresh, verified state snapshot before a consequential action while retaining measurement, enforcement, evaluation, and final dispatch authority.","support":"source-stated"},{"text":"In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a same-shape neutral control and raised single-dispatch completion from 69.4% to 95.8%.","excerpt":"In a 216-decision ablation across GPT-5.6 Sol, Claude Sonnet 5, and Gemini 3.7 Flash, verified tool-health state reduced mean observable policy regret by 96.9%","support":"source-stated"},{"text":"In a separate 27-decision live-execution replication, the verified-state condition completed the live action in 9/9 model-by-fault cases, versus 5/9 for the neutral control.","excerpt":"completed the live action in 9/9 model-by-fault cases versus 5/9 for the neutral control.","support":"source-stated"},{"text":"The authors say the studies establish the tools-and-network instance of Agent Harness Awareness, while listing several areas as open empirical extensions.","excerpt":"Runtime state, hardware pressure, economics, retry-aware recovery, and heterogeneous selection remain open empirical extensions of the same architecture.","support":"source-stated"}],"limitations":["The reported evidence concerns a tools-and-network instance tested in a 216-decision ablation and a separate 27-decision end-to-end replication; the source identifies other areas as open empirical extensions.","The abstract reports selected outcome measures, but does not establish how the results generalize beyond the described studies."],"mediaId":"fb43f0f2d5d92507744d153888d054cbe855a756c32081fc099ce5bafa6d67b6","correctionReason":null,"id":"94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164","publicationId":"53294bcd-6bd2-49e5-89e1-db9d4fed024d","version":1,"sourceUrl":"https://arxiv.org/abs/2611.05391","sourceTitle":"Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-09-13","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:00:31.251Z","createdAt":"2026-11-09T07:08:10.529Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"fb43f0f2d5d92507744d153888d054cbe855a756c32081fc099ce5bafa6d67b6","kind":"article","url":"https://arxiv.org/abs/2611.05391","sourceUrl":"https://arxiv.org/abs/2611.05391","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T07:03:46.613Z","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/2611.05391","checkedAt":"2026-11-09T07:40:08.663Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"fb43f0f2d5d92507744d153888d054cbe855a756c32081fc099ce5bafa6d67b6","confidence":0.91,"provider":"typesafe","model":"jev-1.13.0","inputHash":"f6f23824bc951b93c165a66016ab7f9704ca4907da35c4ce5adb0c04311af8e6","decidedAt":"2026-11-09T07:03:47.340Z"},"provenance":{"textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observationId":"23f17f3afa8abb703e2eb2d063e27ba3ffd9122041ce7c1f62c13f27677e20f8","retrievedAt":"2026-11-09T07:03:46.473Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"d2234bcf-71e0-4a96-8c9f-7ae178abb55b"},"reviewer":{"model":"gpt-6-luna","runId":"bada03fd-1ff8-4582-8a67-5a9858b54ea5"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T07:08:10.529Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Agrawal, Manu","url":null,"kind":"unknown"},{"name":"Nagpal, Shrey","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Agrawal, Manu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Agrawal, Manu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05391","locator":"head/meta[name=author|citation_author]"},"assertionId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:40:08.663Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d","sourceUrl":"https://arxiv.org/abs/2611.05391","contentSha256":"4897518d3b3634c9fa7766a85c30d22916e765147257e257c6fb7480543254d6","textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observedAt":"2026-11-09T07:40:08.663Z"}},{"actor":{"name":"Nagpal, Shrey","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Nagpal, Shrey","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05391","locator":"head/meta[name=author|citation_author]"},"assertionId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:40:08.663Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d","sourceUrl":"https://arxiv.org/abs/2611.05391","contentSha256":"4897518d3b3634c9fa7766a85c30d22916e765147257e257c6fb7480543254d6","textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observedAt":"2026-11-09T07:40:08.663Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.05391","snapshotId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d","contentSha256":"4897518d3b3634c9fa7766a85c30d22916e765147257e257c6fb7480543254d6","textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observedAt":"2026-11-09T07:40:08.663Z","basis":"publisher-page-metadata","sourceLinks":41,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-agent-harness-awareness-turning-verified-telemetry-94cb62f5","formats":{"html":"/readings/arxiv-cs-ai-agent-harness-awareness-turning-verified-telemetry-94cb62f5","markdown":"/readings/arxiv-cs-ai-agent-harness-awareness-turning-verified-telemetry-94cb62f5.md","json":"/readings/arxiv-cs-ai-agent-harness-awareness-turning-verified-telemetry-94cb62f5.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-agent-harness","name":"Agent harness","explanation":"An agent harness is the system that lets an AI model act as an agent: it processes inputs, orchestrates tool calls and returns results. An evaluation harness runs evals end to end, including grading outputs.","referenceUrl":"https://www.anthropic.com/research/demystifying-evals-for-ai-agents"}],"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":"94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164","revision":"53294bcd-6bd2-49e5-89e1-db9d4fed024d","revisionKind":"publication-id","title":"arXiv cs.AI: Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","url":"https://escapemeridian.com/readings/arxiv-cs-ai-agent-harness-awareness-turning-verified-telemetry-94cb62f5"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.05391","revision":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","revisionKind":"sha256-source-text","title":"Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","url":"https://arxiv.org/abs/2611.05391"},"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":"94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164","title":"arXiv cs.AI: Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","argument":"Manu Agrawal and Shrey Nagpal argue that tool-using agents can be operationally blind: their host may know whether a tool is healthy while the model choosing an action does not. Their proposed Agent Harness Awareness design gives the model a small, fresh, verified snapshot before consequential actions, while the host retains measurement, enforcement, evaluation and final dispatch authority.\n\nToolRoute holds the task and candidate semantics constant while varying operational information. In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a neutral control and raised single-dispatch completion from 69.4% to 95.8%. A separate 27-decision live-execution replication reported lower mean regret. It completed the action in 9/9 model-by-fault cases, compared with 5/9 for the neutral control.\n\nThe authors argue these studies establish the tools-and-network instance: selected host telemetry can improve pre-dispatch operational choice. The results are bounded to these studies. Runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection remain open empirical extensions.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported evidence concerns a tools-and-network instance tested in a 216-decision ablation and a separate 27-decision end-to-end replication; the source identifies other areas as open empirical extensions. The abstract reports selected outcome measures, but does not establish how the results generalize beyond the described studies.","speaker":{"name":"Agrawal, Manu, Nagpal, Shrey","roleAtSource":"Article author"},"source":{"title":"Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.05391","publishedAt":"2026-09-13","dateBasis":"Date basis not provided."},"evidence":[{"text":"Agrawal and Nagpal describe operational blindness as a gap between the host's operational information and what the model choosing the next action knows.","excerpt":"We call this gap operational blindness.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"The proposed design gives the model a small, fresh, verified state snapshot before consequential actions, while the host retains control over measurement, enforcement, evaluation and final dispatch.","excerpt":"the host supplies a small, fresh, verified state snapshot before a consequential action while retaining measurement, enforcement, evaluation, and final dispatch authority.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a same-shape neutral control and raised single-dispatch completion from 69.4% to 95.8%.","excerpt":"In a 216-decision ablation across GPT-5.6 Sol, Claude Sonnet 5, and Gemini 3.7 Flash, verified tool-health state reduced mean observable policy regret by 96.9%","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"In a separate 27-decision live-execution replication, the verified-state condition completed the live action in 9/9 model-by-fault cases, versus 5/9 for the neutral control.","excerpt":"completed the live action in 9/9 model-by-fault cases versus 5/9 for the neutral control.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"},{"text":"The authors say the studies establish the tools-and-network instance of Agent Harness Awareness, while listing several areas as open empirical extensions.","excerpt":"Runtime state, hardware pressure, economics, retry-aware recovery, and heterogeneous selection remain open empirical extensions of the same architecture.","url":"https://arxiv.org/abs/2611.05391","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.05391","checkedAt":"2026-11-09T07:40:08.663Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would Agent Harness Awareness perform for the authors’ listed open extensions, including runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","summary":"Manu Agrawal and Shrey Nagpal argue that tool-using agents can be operationally blind: their host may know whether a tool is healthy while the model choosing an action does not. Their proposed Agent Harness Awareness design gives the model a small, fresh, verified snapshot before consequential actions, while the host retains measurement, enforcement, evaluation and final dispatch authority.\n\nToolRoute holds the task and candidate semantics constant while varying operational information. In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a neutral control and raised single-dispatch completion from 69.4% to 95.8%. A separate 27-decision live-execution replication reported lower mean regret. It completed the action in 9/9 model-by-fault cases, compared with 5/9 for the neutral control.\n\nThe authors argue these studies establish the tools-and-network instance: selected host telemetry can improve pre-dispatch operational choice. The results are bounded to these studies. Runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection remain open empirical extensions.","question":"How would Agent Harness Awareness perform for the authors’ listed open extensions, including runtime state, hardware pressure, economics, retry-aware recovery and heterogeneous selection?","claims":[{"text":"Agrawal and Nagpal describe operational blindness as a gap between the host's operational information and what the model choosing the next action knows.","excerpt":"We call this gap operational blindness.","support":"source-stated"},{"text":"The proposed design gives the model a small, fresh, verified state snapshot before consequential actions, while the host retains control over measurement, enforcement, evaluation and final dispatch.","excerpt":"the host supplies a small, fresh, verified state snapshot before a consequential action while retaining measurement, enforcement, evaluation, and final dispatch authority.","support":"source-stated"},{"text":"In a 216-decision ablation, verified tool-health state reduced mean observable policy regret by 96.9% against a same-shape neutral control and raised single-dispatch completion from 69.4% to 95.8%.","excerpt":"In a 216-decision ablation across GPT-5.6 Sol, Claude Sonnet 5, and Gemini 3.7 Flash, verified tool-health state reduced mean observable policy regret by 96.9%","support":"source-stated"},{"text":"In a separate 27-decision live-execution replication, the verified-state condition completed the live action in 9/9 model-by-fault cases, versus 5/9 for the neutral control.","excerpt":"completed the live action in 9/9 model-by-fault cases versus 5/9 for the neutral control.","support":"source-stated"},{"text":"The authors say the studies establish the tools-and-network instance of Agent Harness Awareness, while listing several areas as open empirical extensions.","excerpt":"Runtime state, hardware pressure, economics, retry-aware recovery, and heterogeneous selection remain open empirical extensions of the same architecture.","support":"source-stated"}],"limitations":["The reported evidence concerns a tools-and-network instance tested in a 216-decision ablation and a separate 27-decision end-to-end replication; the source identifies other areas as open empirical extensions.","The abstract reports selected outcome measures, but does not establish how the results generalize beyond the described studies."],"mediaId":"fb43f0f2d5d92507744d153888d054cbe855a756c32081fc099ce5bafa6d67b6","correctionReason":null,"id":"94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164","publicationId":"53294bcd-6bd2-49e5-89e1-db9d4fed024d","version":1,"sourceUrl":"https://arxiv.org/abs/2611.05391","sourceTitle":"Agent Harness Awareness: Turning Verified Telemetry into Operational Control for Tool-Using Agents","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-09-13","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:00:31.251Z","createdAt":"2026-11-09T07:08:10.529Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"fb43f0f2d5d92507744d153888d054cbe855a756c32081fc099ce5bafa6d67b6","kind":"article","url":"https://arxiv.org/abs/2611.05391","sourceUrl":"https://arxiv.org/abs/2611.05391","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T07:03:46.613Z","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/2611.05391","checkedAt":"2026-11-09T07:40:08.663Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"fb43f0f2d5d92507744d153888d054cbe855a756c32081fc099ce5bafa6d67b6","confidence":0.91,"provider":"typesafe","model":"jev-1.13.0","inputHash":"f6f23824bc951b93c165a66016ab7f9704ca4907da35c4ce5adb0c04311af8e6","decidedAt":"2026-11-09T07:03:47.340Z"},"provenance":{"textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observationId":"23f17f3afa8abb703e2eb2d063e27ba3ffd9122041ce7c1f62c13f27677e20f8","retrievedAt":"2026-11-09T07:03:46.473Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"d2234bcf-71e0-4a96-8c9f-7ae178abb55b"},"reviewer":{"model":"gpt-6-luna","runId":"bada03fd-1ff8-4582-8a67-5a9858b54ea5"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T07:08:10.529Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Agrawal, Manu","url":null,"kind":"unknown"},{"name":"Nagpal, Shrey","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Agrawal, Manu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Agrawal, Manu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05391","locator":"head/meta[name=author|citation_author]"},"assertionId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:40:08.663Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d","sourceUrl":"https://arxiv.org/abs/2611.05391","contentSha256":"4897518d3b3634c9fa7766a85c30d22916e765147257e257c6fb7480543254d6","textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observedAt":"2026-11-09T07:40:08.663Z"}},{"actor":{"name":"Nagpal, Shrey","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Nagpal, Shrey","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.05391","locator":"head/meta[name=author|citation_author]"},"assertionId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T07:40:08.663Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d","sourceUrl":"https://arxiv.org/abs/2611.05391","contentSha256":"4897518d3b3634c9fa7766a85c30d22916e765147257e257c6fb7480543254d6","textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observedAt":"2026-11-09T07:40:08.663Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.05391","snapshotId":"a81cb2fecf4c2101159618376f1a2e2ab88e6d4855b9c5ea7f62b2c06bfe424d","contentSha256":"4897518d3b3634c9fa7766a85c30d22916e765147257e257c6fb7480543254d6","textHash":"e8e37559c14d0e0b0a2b2448877879268efc26ee585406b10bfade14dace2cbb","observedAt":"2026-11-09T07:40:08.663Z","basis":"publisher-page-metadata","sourceLinks":41,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/94cb62f5628c80f2404eca2aa6beeab434d9cd44dad8db88dd0a66cc7c58c164/source"}}}],"audits":[{"publicationId":"53294bcd-6bd2-49e5-89e1-db9d4fed024d","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft correctly names Manu Agrawal and Shrey Nagpal as the authors and presents the findings as their claims. Results and the 'establish' statement remain attributed to the authors.","excerpt":"Authors:Manu Agrawal, Shrey Nagpal"},{"dimension":"summary","status":"uncertain","explanation":"Central terms are left unexplained for a non-specialist: harness, telemetry, policy regret, single-dispatch completion, ablation, same-shape neutral control, and model-by-fault cases. Smallest repair: add term-library tooltips or brief plain-language glosses. The draft also flattens the authors' force in two places. It writes 'can improve' where the source says 'materially improves'. It reduces the live replication's 8,956 ms to 0 ms result to 'lower mean regret'. Restore both wordings.","excerpt":"selected host telemetry materially improves pre-dispatch operational choice"},{"dimension":"date","status":"uncertain","explanation":"The originalPublished date of 2026-09-13 matches the stated v1 submission date. However, the identifier arXiv:2611.05391 follows a year-month pattern that implies a November 2026 submission. That conflicts with a September submission and leaves the first-publication date ambiguous. The feed date of 2026-11-05 also aligns with the identifier.","excerpt":"[v1]\nSun, 13 Sep 2026 15:01:55 UTC (94 KB)"},{"dimension":"support","status":"pass","explanation":"The claims are verbatim-supported by the abstract: the 96.9% regret reduction, the 69.4% to 95.8% completion rate, 9/9 versus 5/9, and the listed open extensions. The limitations and the question come from the authors' own stated open extensions. No benefits or costs are invented.","excerpt":"completed the live action in 9/9 model-by-fault cases versus 5/9 for the neutral control"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper. It is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-09T07:08:10.529Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}