{"id":"4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37","title":"Kozachek and Srivastava’s arXiv study tests LLM agents in public-service scenarios","argument":"In an arXiv paper, Diana Kozachek and Kinjalk Srivastava study what LLM agents do when instructions leave the task unclear. Their public-service scenarios involve decisions that can affect residents, patients, and students. In open-ended runs, agents read records in most cases, but produced an artifact or chose a task in fewer than one in five; with an explicit task, production was more common. The authors report that domain context did not change how many runs produced something, while reducing boundary-probing turns. Some artifacts contradicted the records. Comparing the scenarios with a context-free sandbox, the authors found that models’ production ranking largely carried over, but their ranking for whether they read records did not. The contrast suggests that producing an output and attending to case records are distinct behaviors in these tests. The authors propose a test institutions can run on their own records before procurement, using a rule-based taxonomy that needs no labeled data. The abstract does not explain why some artifacts contradict records or establish how well the proposed test predicts behavior in deployed public services.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The arXiv page's abstract does not explain why some artifacts contradict the records or establish how the proposed test predicts behavior in deployed public services.","speaker":{"name":"Kozachek, Diana, Srivastava, Kinjalk","roleAtSource":"Article author"},"source":{"title":"Read and Ignore: What Self-Directed LLM Agents Do in Public Service Scenarios","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.15460","publishedAt":"2026-11-11","dateBasis":"Date basis not provided."},"evidence":[{"text":"Kozachek and Srivastava study LLM agents acting under underspecified instructions in public-service scenarios involving residents, patients, and students.","excerpt":"Underspecified instructions leave LLM agents to decide what to do. In public services, staff write these instructions, but the agent's decisions affect residents, patients, and students.","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"In open-ended runs, agents read records more often than they produced an artifact or chose a task; an explicit task was associated with more production.","excerpt":"Agents read the records in $96.9\\%$ of open-ended runs and produce an artifact or a self-chosen task in $18.7\\%$; with an explicit task, $86.0\\%$ produce.","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"Domain context made little difference to the share of runs producing something, while the models’ production ranking largely carried over between settings and their record-reading ranking did not.","excerpt":"Domain context does not change how many runs produce ($18.7\\%$ vs.\\ $16.0\\%$), and the production ranking of the models largely carries over (Spearman $0.79$), while the ranking by whether a model reads the records does not ($0.19$).","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"The authors report fewer boundary-probing turns with context and say some artifacts contradicted the records.","excerpt":"Context cuts boundary-probing turns per run by $60\\%$, and some artifacts contradict the records.","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"The authors propose a test that public-service institutions can run on their own records before procurement.","excerpt":"We contribute a test that a public-service institution can run on its own records before procurement; its rule-based taxonomy needs no labeled data","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.15460","checkedAt":"2026-11-14T10:20:09.472Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What explains why some artifacts contradict the records in the authors’ scenarios?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Kozachek and Srivastava’s arXiv study tests LLM agents in public-service scenarios","summary":"In an arXiv paper, Diana Kozachek and Kinjalk Srivastava study what LLM agents do when instructions leave the task unclear. Their public-service scenarios involve decisions that can affect residents, patients, and students. In open-ended runs, agents read records in most cases, but produced an artifact or chose a task in fewer than one in five; with an explicit task, production was more common. The authors report that domain context did not change how many runs produced something, while reducing boundary-probing turns. Some artifacts contradicted the records. Comparing the scenarios with a context-free sandbox, the authors found that models’ production ranking largely carried over, but their ranking for whether they read records did not. The contrast suggests that producing an output and attending to case records are distinct behaviors in these tests. The authors propose a test institutions can run on their own records before procurement, using a rule-based taxonomy that needs no labeled data. The abstract does not explain why some artifacts contradict records or establish how well the proposed test predicts behavior in deployed public services.","question":"What explains why some artifacts contradict the records in the authors’ scenarios?","claims":[{"text":"Kozachek and Srivastava study LLM agents acting under underspecified instructions in public-service scenarios involving residents, patients, and students.","excerpt":"Underspecified instructions leave LLM agents to decide what to do. In public services, staff write these instructions, but the agent's decisions affect residents, patients, and students.","support":"source-stated"},{"text":"In open-ended runs, agents read records more often than they produced an artifact or chose a task; an explicit task was associated with more production.","excerpt":"Agents read the records in $96.9\\%$ of open-ended runs and produce an artifact or a self-chosen task in $18.7\\%$; with an explicit task, $86.0\\%$ produce.","support":"source-stated"},{"text":"Domain context made little difference to the share of runs producing something, while the models’ production ranking largely carried over between settings and their record-reading ranking did not.","excerpt":"Domain context does not change how many runs produce ($18.7\\%$ vs.\\ $16.0\\%$), and the production ranking of the models largely carries over (Spearman $0.79$), while the ranking by whether a model reads the records does not ($0.19$).","support":"source-stated"},{"text":"The authors report fewer boundary-probing turns with context and say some artifacts contradicted the records.","excerpt":"Context cuts boundary-probing turns per run by $60\\%$, and some artifacts contradict the records.","support":"source-stated"},{"text":"The authors propose a test that public-service institutions can run on their own records before procurement.","excerpt":"We contribute a test that a public-service institution can run on its own records before procurement; its rule-based taxonomy needs no labeled data","support":"source-stated"}],"limitations":["The arXiv page's abstract does not explain why some artifacts contradict the records or establish how the proposed test predicts behavior in deployed public services."],"mediaId":"02bb1a2d1323033fe886a40c8b98acf398a385ff0fa248b3dc7516bd789bd8ed","correctionReason":null,"id":"4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37","publicationId":"b995c9ce-3e94-45f8-8348-3a0463bbaab6","version":1,"sourceUrl":"https://arxiv.org/abs/2611.15460","sourceTitle":"Read and Ignore: What Self-Directed LLM Agents Do in Public Service Scenarios","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-11","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-13T07:43:35.782Z","createdAt":"2026-11-13T17:16:41.881Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"02bb1a2d1323033fe886a40c8b98acf398a385ff0fa248b3dc7516bd789bd8ed","kind":"article","url":"https://arxiv.org/abs/2611.15460","sourceUrl":"https://arxiv.org/abs/2611.15460","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-13T17:03:34.030Z","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/2611.15460","checkedAt":"2026-11-14T10:20:09.472Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v7","mediaId":"02bb1a2d1323033fe886a40c8b98acf398a385ff0fa248b3dc7516bd789bd8ed","confidence":0.7,"provider":"typesafe","model":"jev-1.13.0","inputHash":"6571e47dca2b0ec797c3e7b0ada72692e95e08042bbcb119ca326d049f5c4077","decidedAt":"2026-11-13T17:03:37.448Z"},"provenance":{"textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observationId":"d83b3bde5af225f838932bb70d863c2f6f69a52f0b130f4bb3a17f79d8c5513d","retrievedAt":"2026-11-13T17:03:33.710Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"ecf32746-9b49-4521-82bd-9dc35ad7208e"},"reviewer":{"model":"gpt-6-luna","runId":"29db105f-5b8f-4b25-8747-32b795bd2442"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-13T17:16:41.881Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Kozachek, Diana","url":null,"kind":"unknown"},{"name":"Srivastava, Kinjalk","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Kozachek, Diana","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kozachek, Diana","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.15460","locator":"head/meta[name=author|citation_author]"},"assertionId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T10:20:09.472Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2","sourceUrl":"https://arxiv.org/abs/2611.15460","contentSha256":"84034fbf9422bc98bb32d319ff1b4eb9069170dd735a9275bdeff7dc14c0a04e","textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observedAt":"2026-11-14T10:20:09.472Z"}},{"actor":{"name":"Srivastava, Kinjalk","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Srivastava, Kinjalk","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.15460","locator":"head/meta[name=author|citation_author]"},"assertionId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T10:20:09.472Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2","sourceUrl":"https://arxiv.org/abs/2611.15460","contentSha256":"84034fbf9422bc98bb32d319ff1b4eb9069170dd735a9275bdeff7dc14c0a04e","textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observedAt":"2026-11-14T10:20:09.472Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.15460","snapshotId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2","contentSha256":"84034fbf9422bc98bb32d319ff1b4eb9069170dd735a9275bdeff7dc14c0a04e","textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observedAt":"2026-11-14T10:20:09.472Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37/source"}},"canonicalUrl":"https://escapemeridian.com/readings/kozachek-and-srivastava-s-arxiv-study-tests-llm-agents-in-4bcf9dac","formats":{"html":"/readings/kozachek-and-srivastava-s-arxiv-study-tests-llm-agents-in-4bcf9dac","markdown":"/readings/kozachek-and-srivastava-s-arxiv-study-tests-llm-agents-in-4bcf9dac.md","json":"/readings/kozachek-and-srivastava-s-arxiv-study-tests-llm-agents-in-4bcf9dac.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":[],"conditions":[],"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":"4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37","revision":"b995c9ce-3e94-45f8-8348-3a0463bbaab6","revisionKind":"publication-id","title":"Kozachek and Srivastava’s arXiv study tests LLM agents in public-service scenarios","url":"https://escapemeridian.com/readings/kozachek-and-srivastava-s-arxiv-study-tests-llm-agents-in-4bcf9dac"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.15460","revision":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","revisionKind":"sha256-source-text","title":"Read and Ignore: What Self-Directed LLM Agents Do in Public Service Scenarios","url":"https://arxiv.org/abs/2611.15460"},"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":"4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37","title":"Kozachek and Srivastava’s arXiv study tests LLM agents in public-service scenarios","argument":"In an arXiv paper, Diana Kozachek and Kinjalk Srivastava study what LLM agents do when instructions leave the task unclear. Their public-service scenarios involve decisions that can affect residents, patients, and students. In open-ended runs, agents read records in most cases, but produced an artifact or chose a task in fewer than one in five; with an explicit task, production was more common. The authors report that domain context did not change how many runs produced something, while reducing boundary-probing turns. Some artifacts contradicted the records. Comparing the scenarios with a context-free sandbox, the authors found that models’ production ranking largely carried over, but their ranking for whether they read records did not. The contrast suggests that producing an output and attending to case records are distinct behaviors in these tests. The authors propose a test institutions can run on their own records before procurement, using a rule-based taxonomy that needs no labeled data. The abstract does not explain why some artifacts contradict records or establish how well the proposed test predicts behavior in deployed public services.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The arXiv page's abstract does not explain why some artifacts contradict the records or establish how the proposed test predicts behavior in deployed public services.","speaker":{"name":"Kozachek, Diana, Srivastava, Kinjalk","roleAtSource":"Article author"},"source":{"title":"Read and Ignore: What Self-Directed LLM Agents Do in Public Service Scenarios","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.15460","publishedAt":"2026-11-11","dateBasis":"Date basis not provided."},"evidence":[{"text":"Kozachek and Srivastava study LLM agents acting under underspecified instructions in public-service scenarios involving residents, patients, and students.","excerpt":"Underspecified instructions leave LLM agents to decide what to do. In public services, staff write these instructions, but the agent's decisions affect residents, patients, and students.","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"In open-ended runs, agents read records more often than they produced an artifact or chose a task; an explicit task was associated with more production.","excerpt":"Agents read the records in $96.9\\%$ of open-ended runs and produce an artifact or a self-chosen task in $18.7\\%$; with an explicit task, $86.0\\%$ produce.","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"Domain context made little difference to the share of runs producing something, while the models’ production ranking largely carried over between settings and their record-reading ranking did not.","excerpt":"Domain context does not change how many runs produce ($18.7\\%$ vs.\\ $16.0\\%$), and the production ranking of the models largely carries over (Spearman $0.79$), while the ranking by whether a model reads the records does not ($0.19$).","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"The authors report fewer boundary-probing turns with context and say some artifacts contradicted the records.","excerpt":"Context cuts boundary-probing turns per run by $60\\%$, and some artifacts contradict the records.","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"},{"text":"The authors propose a test that public-service institutions can run on their own records before procurement.","excerpt":"We contribute a test that a public-service institution can run on its own records before procurement; its rule-based taxonomy needs no labeled data","url":"https://arxiv.org/abs/2611.15460","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.15460","checkedAt":"2026-11-14T10:20:09.472Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What explains why some artifacts contradict the records in the authors’ scenarios?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Kozachek and Srivastava’s arXiv study tests LLM agents in public-service scenarios","summary":"In an arXiv paper, Diana Kozachek and Kinjalk Srivastava study what LLM agents do when instructions leave the task unclear. Their public-service scenarios involve decisions that can affect residents, patients, and students. In open-ended runs, agents read records in most cases, but produced an artifact or chose a task in fewer than one in five; with an explicit task, production was more common. The authors report that domain context did not change how many runs produced something, while reducing boundary-probing turns. Some artifacts contradicted the records. Comparing the scenarios with a context-free sandbox, the authors found that models’ production ranking largely carried over, but their ranking for whether they read records did not. The contrast suggests that producing an output and attending to case records are distinct behaviors in these tests. The authors propose a test institutions can run on their own records before procurement, using a rule-based taxonomy that needs no labeled data. The abstract does not explain why some artifacts contradict records or establish how well the proposed test predicts behavior in deployed public services.","question":"What explains why some artifacts contradict the records in the authors’ scenarios?","claims":[{"text":"Kozachek and Srivastava study LLM agents acting under underspecified instructions in public-service scenarios involving residents, patients, and students.","excerpt":"Underspecified instructions leave LLM agents to decide what to do. In public services, staff write these instructions, but the agent's decisions affect residents, patients, and students.","support":"source-stated"},{"text":"In open-ended runs, agents read records more often than they produced an artifact or chose a task; an explicit task was associated with more production.","excerpt":"Agents read the records in $96.9\\%$ of open-ended runs and produce an artifact or a self-chosen task in $18.7\\%$; with an explicit task, $86.0\\%$ produce.","support":"source-stated"},{"text":"Domain context made little difference to the share of runs producing something, while the models’ production ranking largely carried over between settings and their record-reading ranking did not.","excerpt":"Domain context does not change how many runs produce ($18.7\\%$ vs.\\ $16.0\\%$), and the production ranking of the models largely carries over (Spearman $0.79$), while the ranking by whether a model reads the records does not ($0.19$).","support":"source-stated"},{"text":"The authors report fewer boundary-probing turns with context and say some artifacts contradicted the records.","excerpt":"Context cuts boundary-probing turns per run by $60\\%$, and some artifacts contradict the records.","support":"source-stated"},{"text":"The authors propose a test that public-service institutions can run on their own records before procurement.","excerpt":"We contribute a test that a public-service institution can run on its own records before procurement; its rule-based taxonomy needs no labeled data","support":"source-stated"}],"limitations":["The arXiv page's abstract does not explain why some artifacts contradict the records or establish how the proposed test predicts behavior in deployed public services."],"mediaId":"02bb1a2d1323033fe886a40c8b98acf398a385ff0fa248b3dc7516bd789bd8ed","correctionReason":null,"id":"4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37","publicationId":"b995c9ce-3e94-45f8-8348-3a0463bbaab6","version":1,"sourceUrl":"https://arxiv.org/abs/2611.15460","sourceTitle":"Read and Ignore: What Self-Directed LLM Agents Do in Public Service Scenarios","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-11","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-13T07:43:35.782Z","createdAt":"2026-11-13T17:16:41.881Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"02bb1a2d1323033fe886a40c8b98acf398a385ff0fa248b3dc7516bd789bd8ed","kind":"article","url":"https://arxiv.org/abs/2611.15460","sourceUrl":"https://arxiv.org/abs/2611.15460","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-13T17:03:34.030Z","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/2611.15460","checkedAt":"2026-11-14T10:20:09.472Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v7","mediaId":"02bb1a2d1323033fe886a40c8b98acf398a385ff0fa248b3dc7516bd789bd8ed","confidence":0.7,"provider":"typesafe","model":"jev-1.13.0","inputHash":"6571e47dca2b0ec797c3e7b0ada72692e95e08042bbcb119ca326d049f5c4077","decidedAt":"2026-11-13T17:03:37.448Z"},"provenance":{"textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observationId":"d83b3bde5af225f838932bb70d863c2f6f69a52f0b130f4bb3a17f79d8c5513d","retrievedAt":"2026-11-13T17:03:33.710Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"ecf32746-9b49-4521-82bd-9dc35ad7208e"},"reviewer":{"model":"gpt-6-luna","runId":"29db105f-5b8f-4b25-8747-32b795bd2442"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-13T17:16:41.881Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Kozachek, Diana","url":null,"kind":"unknown"},{"name":"Srivastava, Kinjalk","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Kozachek, Diana","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Kozachek, Diana","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.15460","locator":"head/meta[name=author|citation_author]"},"assertionId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T10:20:09.472Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2","sourceUrl":"https://arxiv.org/abs/2611.15460","contentSha256":"84034fbf9422bc98bb32d319ff1b4eb9069170dd735a9275bdeff7dc14c0a04e","textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observedAt":"2026-11-14T10:20:09.472Z"}},{"actor":{"name":"Srivastava, Kinjalk","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Srivastava, Kinjalk","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.15460","locator":"head/meta[name=author|citation_author]"},"assertionId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T10:20:09.472Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2","sourceUrl":"https://arxiv.org/abs/2611.15460","contentSha256":"84034fbf9422bc98bb32d319ff1b4eb9069170dd735a9275bdeff7dc14c0a04e","textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observedAt":"2026-11-14T10:20:09.472Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.15460","snapshotId":"17e8c30a28db8d96871193c13895428e811b18fbbeca04db5c76aa9bcc6890e2","contentSha256":"84034fbf9422bc98bb32d319ff1b4eb9069170dd735a9275bdeff7dc14c0a04e","textHash":"1ba6b00ee52c96a3b30b9f334ca5d4cfaf8738bc77dbb069d3172ac75ab34a8b","observedAt":"2026-11-14T10:20:09.472Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/4bcf9dac9bd70ad7fd2e486c06dc0361ffd77346d7abb61ba2646dc3feea1d37/source"}}}],"audits":[{"publicationId":"b995c9ce-3e94-45f8-8348-3a0463bbaab6","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft correctly attributes the study to Diana Kozachek and Kinjalk Srivastava in an arXiv paper. Findings are framed as the authors' reports, and the proposed test is presented as their contribution.","excerpt":"Authors:Diana Kozachek, Kinjalk Srivastava"},{"dimension":"summary","status":"uncertain","explanation":"The figures and structure are faithful, but central terms are left unexplained for a general reader. 'Boundary-probing turns', 'artifact', 'context-free sandbox' and 'rule-based taxonomy' carry key findings and are never explained in plain language. A reader cannot tell what context reduced or what kind of output contradicted records. Smallest repair: briefly explain these terms or add tooltips without changing their wording. The draft also omits the scale of the study: ten LLMs, four prompt conditions and 2,400 trajectories. The question is a fair open issue.","excerpt":"Context cuts boundary-probing turns per run by $60\\%$, and some artifacts contradict the records."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-11 matches the arXiv v1 submission date. It does not use the later feed date of 13 Nov.","excerpt":"[Submitted on 11 Nov 2026]"},{"dimension":"support","status":"uncertain","explanation":"The numeric claims match the abstract: 96.9% of runs read the records, 18.7% produced something, 86.0% produced with an explicit task, Spearman correlations were 0.79 and 0.19, and context cut boundary-probing by 60%. However, the sentence 'The contrast suggests that producing an output and attending to case records are distinct behaviors in these tests' is an inference. It is neither attributed to the authors nor labeled as Escape Meridian interpretation. Smallest repair: label it as EM interpretation conditioned on the ranking correlations, or remove it. The limitations sentence correctly avoids claiming the test predicts real-world deployment.","excerpt":"the production ranking of the models largely carries over (Spearman $0.79$), while the ranking by whether a model reads the records does not ($0.19$)"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page, which is the original source.","excerpt":null}],"createdAt":"2026-11-13T17:16:41.881Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}