{"id":"ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95","title":"arXiv cs.AI: LoMASA, a Local Multi-Agent Sales Assistant for B2B Quoting","argument":"Yu Liu and Xingbang Du present LoMASA, a training-free multi-agent system designed to help frontline sales staff respond to changing B2B quotes. It runs on one desktop computer: a 35B mixture-of-experts model makes decisions using confidence read from answer-option probabilities, while a 27B dense model writes customer replies and escalation briefs when consulted. Code computes every price, margin and option, and hard pricing rules cannot be overridden. In a test split of 500 simulated episodes built from public supply-chain data, salespeople had to decide 19.5% of decisions; 38.0% were handled automatically, and hard rules escalated 42.5% to a manager. For automatically handled decisions, the authors report accuracy matching a rule baseline and fewer commitments whose margin later fell below the floor, at the price of additional manager escalations. They also report that option probabilities were well calibrated without training, unlike confidence stated in words by the same model. The 27B model rarely changed a decision, a case where the design did not help. These results concern simulated episodes; how the system performs in live sales work remains an open question.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results come from a test split of simulated episodes built from public supply-chain data; they do not establish performance in live sales work.","speaker":{"name":"Liu, Yu, Du, Xingbang","roleAtSource":"Article author"},"source":{"title":"LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.11234","publishedAt":"2026-11-08","dateBasis":"Date basis not provided."},"evidence":[{"text":"Liu and Du present LoMASA as a training-free system for B2B quoting that runs on one desktop computer.","excerpt":"We present LoMASA (Local Multi-Agent Sales Assistant), a training-free multi-agent system that runs entirely on one desktop computer.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"The system uses a 35B mixture-of-experts model for decisions and a 27B dense model for customer replies and escalation briefs; code computes prices, margins and options, and hard pricing rules cannot be overridden.","excerpt":"Code computes every price, margin and option, and hard pricing rules cannot be overridden.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"In a test split of 500 simulated episodes, salespeople made 19.5% of decisions, 38.0% were settled automatically, and 42.5% were escalated by hard rules.","excerpt":"On a test split of 500 simulated episodes with 1512 decisions built from public supply-chain data, the salesperson has to decide only 19.5% of decisions: 38.0% are settled automatically, with a median latency of 2.67 seconds, and 42.5% are escalated to a manager by the hard rules.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"On automatically settled decisions, LoMASA matched the rule baseline's accuracy and made fewer commitments whose margin later fell below the floor; the authors describe additional manager escalations as the price.","excerpt":"On the automatically settled decisions LoMASA is as accurate as a rule baseline (0.894 vs.\\ 0.894) and makes fewer commitments whose margin later falls below the floor (2.61% vs 4.53%); the price is additional escalations to the manager.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"The authors report that option probabilities were well calibrated without training and that the 27B model rarely changed a decision.","excerpt":"Option probabilities are well calibrated without any training, unlike confidence stated in words by the same model. We also report where the design does not help: the 27B model rarely changes a decision.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.11234","checkedAt":"2026-11-12T06:48:09.143Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would LoMASA perform on real sales decisions beyond the simulated episodes used in this test?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: LoMASA, a Local Multi-Agent Sales Assistant for B2B Quoting","summary":"Yu Liu and Xingbang Du present LoMASA, a training-free multi-agent system designed to help frontline sales staff respond to changing B2B quotes. It runs on one desktop computer: a 35B mixture-of-experts model makes decisions using confidence read from answer-option probabilities, while a 27B dense model writes customer replies and escalation briefs when consulted. Code computes every price, margin and option, and hard pricing rules cannot be overridden. In a test split of 500 simulated episodes built from public supply-chain data, salespeople had to decide 19.5% of decisions; 38.0% were handled automatically, and hard rules escalated 42.5% to a manager. For automatically handled decisions, the authors report accuracy matching a rule baseline and fewer commitments whose margin later fell below the floor, at the price of additional manager escalations. They also report that option probabilities were well calibrated without training, unlike confidence stated in words by the same model. The 27B model rarely changed a decision, a case where the design did not help. These results concern simulated episodes; how the system performs in live sales work remains an open question.","question":"How would LoMASA perform on real sales decisions beyond the simulated episodes used in this test?","claims":[{"text":"Liu and Du present LoMASA as a training-free system for B2B quoting that runs on one desktop computer.","excerpt":"We present LoMASA (Local Multi-Agent Sales Assistant), a training-free multi-agent system that runs entirely on one desktop computer.","support":"source-stated"},{"text":"The system uses a 35B mixture-of-experts model for decisions and a 27B dense model for customer replies and escalation briefs; code computes prices, margins and options, and hard pricing rules cannot be overridden.","excerpt":"Code computes every price, margin and option, and hard pricing rules cannot be overridden.","support":"source-stated"},{"text":"In a test split of 500 simulated episodes, salespeople made 19.5% of decisions, 38.0% were settled automatically, and 42.5% were escalated by hard rules.","excerpt":"On a test split of 500 simulated episodes with 1512 decisions built from public supply-chain data, the salesperson has to decide only 19.5% of decisions: 38.0% are settled automatically, with a median latency of 2.67 seconds, and 42.5% are escalated to a manager by the hard rules.","support":"source-stated"},{"text":"On automatically settled decisions, LoMASA matched the rule baseline's accuracy and made fewer commitments whose margin later fell below the floor; the authors describe additional manager escalations as the price.","excerpt":"On the automatically settled decisions LoMASA is as accurate as a rule baseline (0.894 vs.\\ 0.894) and makes fewer commitments whose margin later falls below the floor (2.61% vs 4.53%); the price is additional escalations to the manager.","support":"source-stated"},{"text":"The authors report that option probabilities were well calibrated without training and that the 27B model rarely changed a decision.","excerpt":"Option probabilities are well calibrated without any training, unlike confidence stated in words by the same model. We also report where the design does not help: the 27B model rarely changes a decision.","support":"source-stated"}],"limitations":["The reported results come from a test split of simulated episodes built from public supply-chain data; they do not establish performance in live sales work."],"mediaId":"65a86c3d47f5a1f3fd972a3d506401603cf822492b2abde39b764ae0384dd2e3","correctionReason":null,"id":"ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95","publicationId":"ab3e7758-5b4c-48c6-885a-c93962fe0be5","version":1,"sourceUrl":"https://arxiv.org/abs/2611.11234","sourceTitle":"LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-08","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:34:22.042Z","createdAt":"2026-11-12T06:32:18.760Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"65a86c3d47f5a1f3fd972a3d506401603cf822492b2abde39b764ae0384dd2e3","kind":"article","url":"https://arxiv.org/abs/2611.11234","sourceUrl":"https://arxiv.org/abs/2611.11234","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T06:31:20.896Z","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/2611.11234","checkedAt":"2026-11-12T06:48:09.143Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"65a86c3d47f5a1f3fd972a3d506401603cf822492b2abde39b764ae0384dd2e3","confidence":0.9,"provider":"typesafe","model":"jev-1.13.0","inputHash":"7e264bc3b7b2f8c19e5fd8c722a2e29abcf37ccb3765d2fbbd2c4d055ff81e2f","decidedAt":"2026-11-12T06:31:21.816Z"},"provenance":{"textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observationId":"1706be9b94628110e176b26c43c25860461cbc1a4e27d40fc643876ff930aa4e","retrievedAt":"2026-11-12T06:31:20.792Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"95f8eecc-4881-4eb1-8180-9f7013b14c63"},"reviewer":{"model":"gpt-6-luna","runId":"c0b852d3-e7b0-4c98-8e44-065c14f1240e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T06:32:18.760Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Liu, Yu","url":null,"kind":"unknown"},{"name":"Du, Xingbang","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Liu, Yu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Yu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.11234","locator":"head/meta[name=author|citation_author]"},"assertionId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T06:48:09.143Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f","sourceUrl":"https://arxiv.org/abs/2611.11234","contentSha256":"2b2301c08c47db89606aa26b23c161c73c17a53b0751bce83631f3e2ed4d2193","textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observedAt":"2026-11-12T06:48:09.143Z"}},{"actor":{"name":"Du, Xingbang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Du, Xingbang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.11234","locator":"head/meta[name=author|citation_author]"},"assertionId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T06:48:09.143Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f","sourceUrl":"https://arxiv.org/abs/2611.11234","contentSha256":"2b2301c08c47db89606aa26b23c161c73c17a53b0751bce83631f3e2ed4d2193","textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observedAt":"2026-11-12T06:48:09.143Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.11234","snapshotId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f","contentSha256":"2b2301c08c47db89606aa26b23c161c73c17a53b0751bce83631f3e2ed4d2193","textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observedAt":"2026-11-12T06:48:09.143Z","basis":"publisher-page-metadata","sourceLinks":41,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-cs-ai-lomasa-a-local-multi-agent-sales-assistant-for-b2b-ecaaaea6","formats":{"html":"/readings/arxiv-cs-ai-lomasa-a-local-multi-agent-sales-assistant-for-b2b-ecaaaea6","markdown":"/readings/arxiv-cs-ai-lomasa-a-local-multi-agent-sales-assistant-for-b2b-ecaaaea6.md","json":"/readings/arxiv-cs-ai-lomasa-a-local-multi-agent-sales-assistant-for-b2b-ecaaaea6.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-mixture-of-experts","name":"Mixture of experts","explanation":"A model design that sends each input to a few specialized sub-networks instead of the whole model, which saves computation.","referenceUrl":"https://developers.google.com/machine-learning/glossary#mixture-of-experts"},{"id":"term-calibration","name":"Model calibration","explanation":"Whether a model's stated confidence matches how often it turns out to be right.","referenceUrl":"https://en.wikipedia.org/wiki/Calibration_(statistics)"}],"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":"ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95","revision":"ab3e7758-5b4c-48c6-885a-c93962fe0be5","revisionKind":"publication-id","title":"arXiv cs.AI: LoMASA, a Local Multi-Agent Sales Assistant for B2B Quoting","url":"https://escapemeridian.com/readings/arxiv-cs-ai-lomasa-a-local-multi-agent-sales-assistant-for-b2b-ecaaaea6"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.11234","revision":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","revisionKind":"sha256-source-text","title":"LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting","url":"https://arxiv.org/abs/2611.11234"},"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":"ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95","title":"arXiv cs.AI: LoMASA, a Local Multi-Agent Sales Assistant for B2B Quoting","argument":"Yu Liu and Xingbang Du present LoMASA, a training-free multi-agent system designed to help frontline sales staff respond to changing B2B quotes. It runs on one desktop computer: a 35B mixture-of-experts model makes decisions using confidence read from answer-option probabilities, while a 27B dense model writes customer replies and escalation briefs when consulted. Code computes every price, margin and option, and hard pricing rules cannot be overridden. In a test split of 500 simulated episodes built from public supply-chain data, salespeople had to decide 19.5% of decisions; 38.0% were handled automatically, and hard rules escalated 42.5% to a manager. For automatically handled decisions, the authors report accuracy matching a rule baseline and fewer commitments whose margin later fell below the floor, at the price of additional manager escalations. They also report that option probabilities were well calibrated without training, unlike confidence stated in words by the same model. The 27B model rarely changed a decision, a case where the design did not help. These results concern simulated episodes; how the system performs in live sales work remains an open question.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results come from a test split of simulated episodes built from public supply-chain data; they do not establish performance in live sales work.","speaker":{"name":"Liu, Yu, Du, Xingbang","roleAtSource":"Article author"},"source":{"title":"LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.11234","publishedAt":"2026-11-08","dateBasis":"Date basis not provided."},"evidence":[{"text":"Liu and Du present LoMASA as a training-free system for B2B quoting that runs on one desktop computer.","excerpt":"We present LoMASA (Local Multi-Agent Sales Assistant), a training-free multi-agent system that runs entirely on one desktop computer.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"The system uses a 35B mixture-of-experts model for decisions and a 27B dense model for customer replies and escalation briefs; code computes prices, margins and options, and hard pricing rules cannot be overridden.","excerpt":"Code computes every price, margin and option, and hard pricing rules cannot be overridden.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"In a test split of 500 simulated episodes, salespeople made 19.5% of decisions, 38.0% were settled automatically, and 42.5% were escalated by hard rules.","excerpt":"On a test split of 500 simulated episodes with 1512 decisions built from public supply-chain data, the salesperson has to decide only 19.5% of decisions: 38.0% are settled automatically, with a median latency of 2.67 seconds, and 42.5% are escalated to a manager by the hard rules.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"On automatically settled decisions, LoMASA matched the rule baseline's accuracy and made fewer commitments whose margin later fell below the floor; the authors describe additional manager escalations as the price.","excerpt":"On the automatically settled decisions LoMASA is as accurate as a rule baseline (0.894 vs.\\ 0.894) and makes fewer commitments whose margin later falls below the floor (2.61% vs 4.53%); the price is additional escalations to the manager.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"},{"text":"The authors report that option probabilities were well calibrated without training and that the 27B model rarely changed a decision.","excerpt":"Option probabilities are well calibrated without any training, unlike confidence stated in words by the same model. We also report where the design does not help: the 27B model rarely changes a decision.","url":"https://arxiv.org/abs/2611.11234","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.11234","checkedAt":"2026-11-12T06:48:09.143Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How would LoMASA perform on real sales decisions beyond the simulated episodes used in this test?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv cs.AI: LoMASA, a Local Multi-Agent Sales Assistant for B2B Quoting","summary":"Yu Liu and Xingbang Du present LoMASA, a training-free multi-agent system designed to help frontline sales staff respond to changing B2B quotes. It runs on one desktop computer: a 35B mixture-of-experts model makes decisions using confidence read from answer-option probabilities, while a 27B dense model writes customer replies and escalation briefs when consulted. Code computes every price, margin and option, and hard pricing rules cannot be overridden. In a test split of 500 simulated episodes built from public supply-chain data, salespeople had to decide 19.5% of decisions; 38.0% were handled automatically, and hard rules escalated 42.5% to a manager. For automatically handled decisions, the authors report accuracy matching a rule baseline and fewer commitments whose margin later fell below the floor, at the price of additional manager escalations. They also report that option probabilities were well calibrated without training, unlike confidence stated in words by the same model. The 27B model rarely changed a decision, a case where the design did not help. These results concern simulated episodes; how the system performs in live sales work remains an open question.","question":"How would LoMASA perform on real sales decisions beyond the simulated episodes used in this test?","claims":[{"text":"Liu and Du present LoMASA as a training-free system for B2B quoting that runs on one desktop computer.","excerpt":"We present LoMASA (Local Multi-Agent Sales Assistant), a training-free multi-agent system that runs entirely on one desktop computer.","support":"source-stated"},{"text":"The system uses a 35B mixture-of-experts model for decisions and a 27B dense model for customer replies and escalation briefs; code computes prices, margins and options, and hard pricing rules cannot be overridden.","excerpt":"Code computes every price, margin and option, and hard pricing rules cannot be overridden.","support":"source-stated"},{"text":"In a test split of 500 simulated episodes, salespeople made 19.5% of decisions, 38.0% were settled automatically, and 42.5% were escalated by hard rules.","excerpt":"On a test split of 500 simulated episodes with 1512 decisions built from public supply-chain data, the salesperson has to decide only 19.5% of decisions: 38.0% are settled automatically, with a median latency of 2.67 seconds, and 42.5% are escalated to a manager by the hard rules.","support":"source-stated"},{"text":"On automatically settled decisions, LoMASA matched the rule baseline's accuracy and made fewer commitments whose margin later fell below the floor; the authors describe additional manager escalations as the price.","excerpt":"On the automatically settled decisions LoMASA is as accurate as a rule baseline (0.894 vs.\\ 0.894) and makes fewer commitments whose margin later falls below the floor (2.61% vs 4.53%); the price is additional escalations to the manager.","support":"source-stated"},{"text":"The authors report that option probabilities were well calibrated without training and that the 27B model rarely changed a decision.","excerpt":"Option probabilities are well calibrated without any training, unlike confidence stated in words by the same model. We also report where the design does not help: the 27B model rarely changes a decision.","support":"source-stated"}],"limitations":["The reported results come from a test split of simulated episodes built from public supply-chain data; they do not establish performance in live sales work."],"mediaId":"65a86c3d47f5a1f3fd972a3d506401603cf822492b2abde39b764ae0384dd2e3","correctionReason":null,"id":"ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95","publicationId":"ab3e7758-5b4c-48c6-885a-c93962fe0be5","version":1,"sourceUrl":"https://arxiv.org/abs/2611.11234","sourceTitle":"LoMASA: A Local Multi-Agent Sales Assistant with Confidence-Gated Delegation for B2B Quoting","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-08","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-10T08:34:22.042Z","createdAt":"2026-11-12T06:32:18.760Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"65a86c3d47f5a1f3fd972a3d506401603cf822492b2abde39b764ae0384dd2e3","kind":"article","url":"https://arxiv.org/abs/2611.11234","sourceUrl":"https://arxiv.org/abs/2611.11234","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T06:31:20.896Z","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/2611.11234","checkedAt":"2026-11-12T06:48:09.143Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"65a86c3d47f5a1f3fd972a3d506401603cf822492b2abde39b764ae0384dd2e3","confidence":0.9,"provider":"typesafe","model":"jev-1.13.0","inputHash":"7e264bc3b7b2f8c19e5fd8c722a2e29abcf37ccb3765d2fbbd2c4d055ff81e2f","decidedAt":"2026-11-12T06:31:21.816Z"},"provenance":{"textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observationId":"1706be9b94628110e176b26c43c25860461cbc1a4e27d40fc643876ff930aa4e","retrievedAt":"2026-11-12T06:31:20.792Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"95f8eecc-4881-4eb1-8180-9f7013b14c63"},"reviewer":{"model":"gpt-6-luna","runId":"c0b852d3-e7b0-4c98-8e44-065c14f1240e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T06:32:18.760Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Liu, Yu","url":null,"kind":"unknown"},{"name":"Du, Xingbang","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Liu, Yu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Yu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.11234","locator":"head/meta[name=author|citation_author]"},"assertionId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T06:48:09.143Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f","sourceUrl":"https://arxiv.org/abs/2611.11234","contentSha256":"2b2301c08c47db89606aa26b23c161c73c17a53b0751bce83631f3e2ed4d2193","textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observedAt":"2026-11-12T06:48:09.143Z"}},{"actor":{"name":"Du, Xingbang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Du, Xingbang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.11234","locator":"head/meta[name=author|citation_author]"},"assertionId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T06:48:09.143Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f","sourceUrl":"https://arxiv.org/abs/2611.11234","contentSha256":"2b2301c08c47db89606aa26b23c161c73c17a53b0751bce83631f3e2ed4d2193","textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observedAt":"2026-11-12T06:48:09.143Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.11234","snapshotId":"aed211d2a5e13315d1c0f1eba5bb44bc2c81148d655bc328404c210fad37da2f","contentSha256":"2b2301c08c47db89606aa26b23c161c73c17a53b0751bce83631f3e2ed4d2193","textHash":"abb6b56735e5892c740972cf679fc292f5a839f795e3cdca2a38eadb02892f0b","observedAt":"2026-11-12T06:48:09.143Z","basis":"publisher-page-metadata","sourceLinks":41,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/ecaaaea6bfe75335e5912ee596cda67310df0c6c05a0db5f4ef2c1088f7a1a95/source"}}}],"audits":[{"publicationId":"ab3e7758-5b4c-48c6-885a-c93962fe0be5","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly credits the paper to Yu Liu and Xingbang Du on arXiv cs.AI. It keeps results attributed to the authors ('the authors report') rather than presenting them as established facts.","excerpt":"Authors:Yu Liu, Xingbang Du"},{"dimension":"summary","status":"uncertain","explanation":"The reading blurs the paper's central idea, confidence-gated delegation. The source says the 27B model writes replies and briefs AND is consulted when System-1 is unsure. The draft's phrase 'when consulted' never says the trigger is low confidence. Smallest repair: add 'and is consulted when the 35B model is unsure.' Also, 'mixture-of-experts', 'dense model', 'B2B' and 'well calibrated' need tooltips or brief plain glosses. For 'well calibrated', a gloss such as 'its confidence matched how often it was right' would make the significance readable for non-specialists.","excerpt":"A 27B dense model (System-2) writes customer replies and escalation briefs and is consulted when System-1 is unsure."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-08 matches the v1 submission date on the arXiv page. It is not taken from the later feed date.","excerpt":"[Submitted on 8 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Every number and claim is source-stated: the 19.5/38.0/42.5% split, matching accuracy, fewer below-floor commitments at the cost of extra escalations, calibration, and the 27B model rarely changing a decision. The limitation about simulated episodes and the open question are appropriately framed rather than asserted. No invented benefits or costs.","excerpt":"the price is additional escalations to the manager"},{"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":"2026-11-12T06:32:18.760Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}