{"id":"88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8","title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","argument":"In their paper, Tianyi Huang and coauthors study a problem in repeated human-AI interaction: one observation can return as several quoted or summarized reports, making repeated evidence look independent. They introduce Latent Quotient Evidence Training (LQET), which trains language models to use supplied report lineage—the history connecting reports to their evidence—rather than report count when updating beliefs. On a controlled binary benchmark, the authors report lower update error and less confidence drift after repeated relays than task-only training. They also report gains in a closed-loop simulation, where model confidence affects how many reports return in later rounds. The results suggest a possible way to reduce the influence of repeated reports that add no evidence, but the paper does not establish performance in real-world human-AI systems. How well would LQET work when report lineage is incomplete or inaccurate?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported benchmark is controlled and binary; the evidence does not establish performance in real-world human-AI interaction. The reported comparisons cover Qwen3-4B and Phi-4-mini, with three paired training seeds per backbone. The evidence describes a closed-loop simulation, not deployment with people or a test of incomplete or inaccurate report lineage.","speaker":{"name":"Huang, Tianyi, Liu, Eric, Chong, Jay, Yin, Emily, Zhao, Wanqi","roleAtSource":"Article author"},"source":{"title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.13008","publishedAt":"2026-10-09","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors describe how relayed versions of one observation can appear to be independent corroboration.","excerpt":"a single observation can be quoted, summarized, and returned to a model as multiple reports, making repetition look like independent corroboration.","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"The authors introduce LQET, a training objective that uses supplied report lineage to guide belief updates.","excerpt":"We introduce Latent Quotient Evidence Training (LQET), a relational objective that combines sparse Bayesian supervision","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"On a controlled binary benchmark, LQET reduced new-evidence update error and confidence drift relative to task-only training on two model backbones.","excerpt":"LQET reduces new-evidence update RMSE by 41-44\\% and log-odds drift after eight relay steps by 89-91\\% relative to task-only training across Qwen3-4B and Phi-4-mini","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"The authors report that gains extended to a closed-loop simulation where model confidence affected the return of relayed reports.","excerpt":"The gains extend to a closed-loop simulation in which model confidence controls how many relayed reports re-enter the model's context in later rounds.","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"These results suggest LQET could help limit the influence of repeated reports that add no evidence; real-world effects remain unestablished.","excerpt":"a route toward dynamic human-AI systems in which influence follows new evidence, not the number of times the same evidence returns.","url":"https://arxiv.org/abs/2610.13008","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.13008","checkedAt":"2026-11-10T20:50:07.895Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well would LQET work when report lineage is incomplete or inaccurate?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","summary":"In their paper, Tianyi Huang and coauthors study a problem in repeated human-AI interaction: one observation can return as several quoted or summarized reports, making repeated evidence look independent. They introduce Latent Quotient Evidence Training (LQET), which trains language models to use supplied report lineage—the history connecting reports to their evidence—rather than report count when updating beliefs. On a controlled binary benchmark, the authors report lower update error and less confidence drift after repeated relays than task-only training. They also report gains in a closed-loop simulation, where model confidence affects how many reports return in later rounds. The results suggest a possible way to reduce the influence of repeated reports that add no evidence, but the paper does not establish performance in real-world human-AI systems. How well would LQET work when report lineage is incomplete or inaccurate?","question":"How well would LQET work when report lineage is incomplete or inaccurate?","claims":[{"text":"The authors describe how relayed versions of one observation can appear to be independent corroboration.","excerpt":"a single observation can be quoted, summarized, and returned to a model as multiple reports, making repetition look like independent corroboration.","support":"source-stated"},{"text":"The authors introduce LQET, a training objective that uses supplied report lineage to guide belief updates.","excerpt":"We introduce Latent Quotient Evidence Training (LQET), a relational objective that combines sparse Bayesian supervision","support":"source-stated"},{"text":"On a controlled binary benchmark, LQET reduced new-evidence update error and confidence drift relative to task-only training on two model backbones.","excerpt":"LQET reduces new-evidence update RMSE by 41-44\\% and log-odds drift after eight relay steps by 89-91\\% relative to task-only training across Qwen3-4B and Phi-4-mini","support":"source-stated"},{"text":"The authors report that gains extended to a closed-loop simulation where model confidence affected the return of relayed reports.","excerpt":"The gains extend to a closed-loop simulation in which model confidence controls how many relayed reports re-enter the model's context in later rounds.","support":"source-stated"},{"text":"These results suggest LQET could help limit the influence of repeated reports that add no evidence; real-world effects remain unestablished.","excerpt":"a route toward dynamic human-AI systems in which influence follows new evidence, not the number of times the same evidence returns.","support":"interpretation"}],"limitations":["The reported benchmark is controlled and binary; the evidence does not establish performance in real-world human-AI interaction.","The reported comparisons cover Qwen3-4B and Phi-4-mini, with three paired training seeds per backbone.","The evidence describes a closed-loop simulation, not deployment with people or a test of incomplete or inaccurate report lineage."],"mediaId":"663ab8cc55c03e343d7bb6504abeee247fd5c1ad31c27e2d0148b022e515e49c","correctionReason":null,"id":"88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8","publicationId":"fb167d8a-032b-4eb4-89d4-21ae7bbdc79a","version":1,"sourceUrl":"https://arxiv.org/abs/2610.13008","sourceTitle":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-09","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-12T08:33:07.207Z","createdAt":"2026-11-10T20:18:28.137Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"663ab8cc55c03e343d7bb6504abeee247fd5c1ad31c27e2d0148b022e515e49c","kind":"article","url":"https://arxiv.org/abs/2610.13008","sourceUrl":"https://arxiv.org/abs/2610.13008","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T03:32:28.230Z","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/2610.13008","checkedAt":"2026-11-10T20:50:07.895Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"663ab8cc55c03e343d7bb6504abeee247fd5c1ad31c27e2d0148b022e515e49c","confidence":0.8400000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"084e80191053fd7effd9e42cfce06956a8fba519e08a2cbb8f5fdf87fd5a4f75","decidedAt":"2026-10-30T03:32:29.229Z"},"provenance":{"textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observationId":"1ae8bd8883c6957e9afc37aa1b102dc5cced517a900a66befd4e4c73655a536c","retrievedAt":"2026-10-30T03:32:28.082Z","kind":"publisher-page","preparer":{"model":"server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna)","runId":"7501581d-b863-42ed-8dca-19a769deaf91"},"reviewer":{"model":"gpt-6-luna","runId":"82466c35-b33a-48bf-8e5f-82a7c6db8b43"}},"audit":{"status":"passed","reviewedAt":"2026-11-10T20:18:28.137Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Tianyi","url":null,"kind":"unknown"},{"name":"Liu, Eric","url":null,"kind":"unknown"},{"name":"Chong, Jay","url":null,"kind":"unknown"},{"name":"Yin, Emily","url":null,"kind":"unknown"},{"name":"Zhao, Wanqi","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Tianyi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Tianyi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Liu, Eric","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Eric","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Chong, Jay","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chong, Jay","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Yin, Emily","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yin, Emily","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Zhao, Wanqi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhao, Wanqi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.13008","snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8/source"}},"canonicalUrl":"https://escapemeridian.com/readings/provenance-aware-belief-updating-in-human-ai-feedback-loops-88e35ebb","formats":{"html":"/readings/provenance-aware-belief-updating-in-human-ai-feedback-loops-88e35ebb","markdown":"/readings/provenance-aware-belief-updating-in-human-ai-feedback-loops-88e35ebb.md","json":"/readings/provenance-aware-belief-updating-in-human-ai-feedback-loops-88e35ebb.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-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"}],"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":"88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8","revision":"fb167d8a-032b-4eb4-89d4-21ae7bbdc79a","revisionKind":"publication-id","title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","url":"https://escapemeridian.com/readings/provenance-aware-belief-updating-in-human-ai-feedback-loops-88e35ebb"},"source":{"type":"source","id":"https://arxiv.org/abs/2610.13008","revision":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","revisionKind":"sha256-source-text","title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","url":"https://arxiv.org/abs/2610.13008"},"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":"88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8","title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","argument":"In their paper, Tianyi Huang and coauthors study a problem in repeated human-AI interaction: one observation can return as several quoted or summarized reports, making repeated evidence look independent. They introduce Latent Quotient Evidence Training (LQET), which trains language models to use supplied report lineage—the history connecting reports to their evidence—rather than report count when updating beliefs. On a controlled binary benchmark, the authors report lower update error and less confidence drift after repeated relays than task-only training. They also report gains in a closed-loop simulation, where model confidence affects how many reports return in later rounds. The results suggest a possible way to reduce the influence of repeated reports that add no evidence, but the paper does not establish performance in real-world human-AI systems. How well would LQET work when report lineage is incomplete or inaccurate?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported benchmark is controlled and binary; the evidence does not establish performance in real-world human-AI interaction. The reported comparisons cover Qwen3-4B and Phi-4-mini, with three paired training seeds per backbone. The evidence describes a closed-loop simulation, not deployment with people or a test of incomplete or inaccurate report lineage.","speaker":{"name":"Huang, Tianyi, Liu, Eric, Chong, Jay, Yin, Emily, Zhao, Wanqi","roleAtSource":"Article author"},"source":{"title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.13008","publishedAt":"2026-10-09","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors describe how relayed versions of one observation can appear to be independent corroboration.","excerpt":"a single observation can be quoted, summarized, and returned to a model as multiple reports, making repetition look like independent corroboration.","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"The authors introduce LQET, a training objective that uses supplied report lineage to guide belief updates.","excerpt":"We introduce Latent Quotient Evidence Training (LQET), a relational objective that combines sparse Bayesian supervision","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"On a controlled binary benchmark, LQET reduced new-evidence update error and confidence drift relative to task-only training on two model backbones.","excerpt":"LQET reduces new-evidence update RMSE by 41-44\\% and log-odds drift after eight relay steps by 89-91\\% relative to task-only training across Qwen3-4B and Phi-4-mini","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"The authors report that gains extended to a closed-loop simulation where model confidence affected the return of relayed reports.","excerpt":"The gains extend to a closed-loop simulation in which model confidence controls how many relayed reports re-enter the model's context in later rounds.","url":"https://arxiv.org/abs/2610.13008","relationship":"Source-stated"},{"text":"These results suggest LQET could help limit the influence of repeated reports that add no evidence; real-world effects remain unestablished.","excerpt":"a route toward dynamic human-AI systems in which influence follows new evidence, not the number of times the same evidence returns.","url":"https://arxiv.org/abs/2610.13008","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.13008","checkedAt":"2026-11-10T20:50:07.895Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How well would LQET work when report lineage is incomplete or inaccurate?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","summary":"In their paper, Tianyi Huang and coauthors study a problem in repeated human-AI interaction: one observation can return as several quoted or summarized reports, making repeated evidence look independent. They introduce Latent Quotient Evidence Training (LQET), which trains language models to use supplied report lineage—the history connecting reports to their evidence—rather than report count when updating beliefs. On a controlled binary benchmark, the authors report lower update error and less confidence drift after repeated relays than task-only training. They also report gains in a closed-loop simulation, where model confidence affects how many reports return in later rounds. The results suggest a possible way to reduce the influence of repeated reports that add no evidence, but the paper does not establish performance in real-world human-AI systems. How well would LQET work when report lineage is incomplete or inaccurate?","question":"How well would LQET work when report lineage is incomplete or inaccurate?","claims":[{"text":"The authors describe how relayed versions of one observation can appear to be independent corroboration.","excerpt":"a single observation can be quoted, summarized, and returned to a model as multiple reports, making repetition look like independent corroboration.","support":"source-stated"},{"text":"The authors introduce LQET, a training objective that uses supplied report lineage to guide belief updates.","excerpt":"We introduce Latent Quotient Evidence Training (LQET), a relational objective that combines sparse Bayesian supervision","support":"source-stated"},{"text":"On a controlled binary benchmark, LQET reduced new-evidence update error and confidence drift relative to task-only training on two model backbones.","excerpt":"LQET reduces new-evidence update RMSE by 41-44\\% and log-odds drift after eight relay steps by 89-91\\% relative to task-only training across Qwen3-4B and Phi-4-mini","support":"source-stated"},{"text":"The authors report that gains extended to a closed-loop simulation where model confidence affected the return of relayed reports.","excerpt":"The gains extend to a closed-loop simulation in which model confidence controls how many relayed reports re-enter the model's context in later rounds.","support":"source-stated"},{"text":"These results suggest LQET could help limit the influence of repeated reports that add no evidence; real-world effects remain unestablished.","excerpt":"a route toward dynamic human-AI systems in which influence follows new evidence, not the number of times the same evidence returns.","support":"interpretation"}],"limitations":["The reported benchmark is controlled and binary; the evidence does not establish performance in real-world human-AI interaction.","The reported comparisons cover Qwen3-4B and Phi-4-mini, with three paired training seeds per backbone.","The evidence describes a closed-loop simulation, not deployment with people or a test of incomplete or inaccurate report lineage."],"mediaId":"663ab8cc55c03e343d7bb6504abeee247fd5c1ad31c27e2d0148b022e515e49c","correctionReason":null,"id":"88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8","publicationId":"fb167d8a-032b-4eb4-89d4-21ae7bbdc79a","version":1,"sourceUrl":"https://arxiv.org/abs/2610.13008","sourceTitle":"Provenance-Aware Belief Updating in Human-AI Feedback Loops","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-09","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-12T08:33:07.207Z","createdAt":"2026-11-10T20:18:28.137Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"663ab8cc55c03e343d7bb6504abeee247fd5c1ad31c27e2d0148b022e515e49c","kind":"article","url":"https://arxiv.org/abs/2610.13008","sourceUrl":"https://arxiv.org/abs/2610.13008","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T03:32:28.230Z","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/2610.13008","checkedAt":"2026-11-10T20:50:07.895Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"663ab8cc55c03e343d7bb6504abeee247fd5c1ad31c27e2d0148b022e515e49c","confidence":0.8400000000000001,"provider":"typesafe","model":"jev-1.13.0","inputHash":"084e80191053fd7effd9e42cfce06956a8fba519e08a2cbb8f5fdf87fd5a4f75","decidedAt":"2026-10-30T03:32:29.229Z"},"provenance":{"textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observationId":"1ae8bd8883c6957e9afc37aa1b102dc5cced517a900a66befd4e4c73655a536c","retrievedAt":"2026-10-30T03:32:28.082Z","kind":"publisher-page","preparer":{"model":"server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna)","runId":"7501581d-b863-42ed-8dca-19a769deaf91"},"reviewer":{"model":"gpt-6-luna","runId":"82466c35-b33a-48bf-8e5f-82a7c6db8b43"}},"audit":{"status":"passed","reviewedAt":"2026-11-10T20:18:28.137Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Tianyi","url":null,"kind":"unknown"},{"name":"Liu, Eric","url":null,"kind":"unknown"},{"name":"Chong, Jay","url":null,"kind":"unknown"},{"name":"Yin, Emily","url":null,"kind":"unknown"},{"name":"Zhao, Wanqi","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Tianyi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Tianyi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Liu, Eric","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Eric","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Chong, Jay","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Chong, Jay","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Yin, Emily","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yin, Emily","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}},{"actor":{"name":"Zhao, Wanqi","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhao, Wanqi","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.13008","locator":"head/meta[name=author|citation_author]"},"assertionId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-10T20:50:07.895Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","sourceUrl":"https://arxiv.org/abs/2610.13008","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.13008","snapshotId":"d41ac445558b683d51adc86a8bf32c277f4c0b332489efed820383bbae0dd6d2","contentSha256":"2e113c1e22e18b7d081604aae73c0d78f4c7bd6ac81b46379e6f94c792eb3011","textHash":"a69c4b7532eff99593415ef1454531aa99eac41459ec664831c062c3e2b1edec","observedAt":"2026-11-10T20:50:07.895Z","basis":"publisher-page-metadata","sourceLinks":45,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/88e35ebbe6b0c3f138afa95a9f872b8281095f5fd37e2c85dace5a7ac64d21a8/source"}}}],"audits":[{"publicationId":"fb167d8a-032b-4eb4-89d4-21ae7bbdc79a","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"The draft attributes the study, LQET and the reported results to Tianyi Huang and coauthors, matching the arXiv author list. Results are framed as the authors' reports, not as established facts.","excerpt":"Authors:Tianyi Huang, Eric Liu, Jay Chong, Emily Yin, Wanqi Zhao"},{"dimension":"summary","status":"pass","explanation":"The draft opens with the actor and the problem: repeated relays of one observation can look like independent corroboration. It explains report lineage in plain language and describes the closed-loop setup. It conveys the proposed significance and its limits. Paraphrasing RMSE as update error and log-odds drift as confidence drift loosens the exact terms, but the claim text keeps the meaning. The question raises a genuine unresolved issue that follows from the abstract.","excerpt":"making repetition look like independent corroboration"},{"dimension":"date","status":"pass","explanation":"originalPublished is 2026-10-09, matching v1 submission on 9 Oct 2026, not the 27 Oct revision or the 29 Oct feed date.","excerpt":"[Submitted on 9 Oct 2026 (v1), last revised 27 Oct 2026 (this version, v2)]"},{"dimension":"support","status":"pass","explanation":"Each claim matches the abstract. This includes the benchmark reductions on Qwen3-4B and Phi-4-mini relative to task-only training, and the closed-loop simulation gains. The interpretive claim is labeled, conditional, and close to the authors' stated route. Limitations correctly note the controlled, simulated scope without overstating real-world benefit.","excerpt":"LQET reduces new-evidence update RMSE by 41-44\\% and log-odds drift after eight relay steps by 89-91\\% relative to task-only training across Qwen3-4B and Phi-4-mini"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified arXiv abstract page for this paper. It is the direct original source and is the most relevant access point.","excerpt":null}],"createdAt":"2026-11-10T20:18:28.137Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}