{"id":"73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9","title":"Xuliang Huang and coauthors propose GIST for selecting offline RLVR training data","argument":"Authors Xuliang Huang, Tianxiang Zhao, Chengwei Qin, and Yao Shu argue that offline reinforcement learning with verifiable rewards (RLVR) should not supervise every response and token equally. Their analysis uses the Gibbs-optimal policy of KL-regularized RLVR to identify where a policy differs from its training target: correct responses matter mainly through low-probability tokens, while incorrect responses matter through frequently generated responses and their high-probability tokens. They propose GIST, a two-level data-selection method that uses sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode, then uses initial token probabilities to select tokens for supervision. The authors report higher average accuracy across GSM8K, MATH-500, and MBPP for all three models under two offline training objectives. In a controlled Qwen3-0.6B/GSM8K comparison, they report that GIST outperforms seven selection baselines using the same candidates while supervising 66.6% fewer tokens than training on all candidates of the same prompts. The reported results concern these evaluations; whether the gains and token savings carry over to other settings remains open.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The evidence supplied is the abstract, which does not identify the three models or name the two offline training objectives. The 66.6% token reduction and comparison with seven baselines are reported for Qwen3-0.6B/GSM8K, not for every evaluation described.","speaker":{"name":"Huang, Xuliang, Zhao, Tianxiang, Qin, Chengwei, Shu, Yao","roleAtSource":"Article author"},"source":{"title":"GIST: Gibbs-Targeted Hierarchical Data Selection for Offline RLVR","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.06395","publishedAt":"2026-11-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors analyze which responses and tokens contribute to mismatch between a policy and the Gibbs-optimal policy of KL-regularized RLVR.","excerpt":"measure the mismatch between a policy and this target with a bidirectional KL divergence","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"The authors report that correct and incorrect responses contribute to the mismatch through different tokens and response patterns.","excerpt":"correct responses contribute mainly through the forward term, concentrated on their low-probability tokens","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"GIST uses sampling temperature and initial policy token probabilities to select responses and tokens for supervision before training.","excerpt":"GIST uses the sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"The authors report higher average accuracy across three benchmarks for all three models under both offline training objectives.","excerpt":"GIST raises the average accuracy over GSM8K, MATH-500, and MBPP for all three models under both objectives.","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"In one controlled comparison, the authors report that GIST beats seven selection baselines while supervising fewer tokens.","excerpt":"it outperforms seven selection baselines that choose from the same candidates while supervising 66.6% fewer tokens","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.06395","checkedAt":"2026-11-09T11:40:11.118Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do GIST's reported accuracy gains and token savings hold beyond the models, objectives, tasks, and controlled comparison described in the abstract?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Xuliang Huang and coauthors propose GIST for selecting offline RLVR training data","summary":"Authors Xuliang Huang, Tianxiang Zhao, Chengwei Qin, and Yao Shu argue that offline reinforcement learning with verifiable rewards (RLVR) should not supervise every response and token equally. Their analysis uses the Gibbs-optimal policy of KL-regularized RLVR to identify where a policy differs from its training target: correct responses matter mainly through low-probability tokens, while incorrect responses matter through frequently generated responses and their high-probability tokens. They propose GIST, a two-level data-selection method that uses sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode, then uses initial token probabilities to select tokens for supervision. The authors report higher average accuracy across GSM8K, MATH-500, and MBPP for all three models under two offline training objectives. In a controlled Qwen3-0.6B/GSM8K comparison, they report that GIST outperforms seven selection baselines using the same candidates while supervising 66.6% fewer tokens than training on all candidates of the same prompts. The reported results concern these evaluations; whether the gains and token savings carry over to other settings remains open.","question":"Do GIST's reported accuracy gains and token savings hold beyond the models, objectives, tasks, and controlled comparison described in the abstract?","claims":[{"text":"The authors analyze which responses and tokens contribute to mismatch between a policy and the Gibbs-optimal policy of KL-regularized RLVR.","excerpt":"measure the mismatch between a policy and this target with a bidirectional KL divergence","support":"source-stated"},{"text":"The authors report that correct and incorrect responses contribute to the mismatch through different tokens and response patterns.","excerpt":"correct responses contribute mainly through the forward term, concentrated on their low-probability tokens","support":"source-stated"},{"text":"GIST uses sampling temperature and initial policy token probabilities to select responses and tokens for supervision before training.","excerpt":"GIST uses the sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode","support":"source-stated"},{"text":"The authors report higher average accuracy across three benchmarks for all three models under both offline training objectives.","excerpt":"GIST raises the average accuracy over GSM8K, MATH-500, and MBPP for all three models under both objectives.","support":"source-stated"},{"text":"In one controlled comparison, the authors report that GIST beats seven selection baselines while supervising fewer tokens.","excerpt":"it outperforms seven selection baselines that choose from the same candidates while supervising 66.6% fewer tokens","support":"source-stated"}],"limitations":["The evidence supplied is the abstract, which does not identify the three models or name the two offline training objectives.","The 66.6% token reduction and comparison with seven baselines are reported for Qwen3-0.6B/GSM8K, not for every evaluation described."],"mediaId":"14229db44999bbe3b4416df794ec9ca114ed9bcd24923dd2bfd1706bad5b4964","correctionReason":null,"id":"73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9","publicationId":"98932b57-4924-474d-82eb-ae8c8b0b9b8b","version":1,"sourceUrl":"https://arxiv.org/abs/2611.06395","sourceTitle":"GIST: Gibbs-Targeted Hierarchical Data Selection for Offline RLVR","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:00:31.762Z","createdAt":"2026-11-09T11:09:58.800Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"14229db44999bbe3b4416df794ec9ca114ed9bcd24923dd2bfd1706bad5b4964","kind":"article","url":"https://arxiv.org/abs/2611.06395","sourceUrl":"https://arxiv.org/abs/2611.06395","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T11:09:13.996Z","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/2611.06395","checkedAt":"2026-11-09T11:40:11.118Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"14229db44999bbe3b4416df794ec9ca114ed9bcd24923dd2bfd1706bad5b4964","confidence":0.85,"provider":"typesafe","model":"jev-1.13.0","inputHash":"662c773ae35276ce1a288f3d62784edbb57827a2c8b166ecfb9d391b1d68ebb5","decidedAt":"2026-11-09T11:09:15.539Z"},"provenance":{"textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observationId":"2584d49706550c5bd850a3a235fee5c97e4650bcdcd62e10ad4140e3e330204a","retrievedAt":"2026-11-09T11:09:13.740Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"a06a6323-7750-4336-85bd-3723fd60c4d2"},"reviewer":{"model":"gpt-6-luna","runId":"c4ee8708-0e2f-472f-8f91-002d026eda8c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T11:09:58.800Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Xuliang","url":null,"kind":"unknown"},{"name":"Zhao, Tianxiang","url":null,"kind":"unknown"},{"name":"Qin, Chengwei","url":null,"kind":"unknown"},{"name":"Shu, Yao","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Xuliang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Xuliang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}},{"actor":{"name":"Zhao, Tianxiang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhao, Tianxiang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}},{"actor":{"name":"Qin, Chengwei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Qin, Chengwei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}},{"actor":{"name":"Shu, Yao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shu, Yao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.06395","snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9/source"}},"canonicalUrl":"https://escapemeridian.com/readings/xuliang-huang-and-coauthors-propose-gist-for-selecting-offline-73e73941","formats":{"html":"/readings/xuliang-huang-and-coauthors-propose-gist-for-selecting-offline-73e73941","markdown":"/readings/xuliang-huang-and-coauthors-propose-gist-for-selecting-offline-73e73941.md","json":"/readings/xuliang-huang-and-coauthors-propose-gist-for-selecting-offline-73e73941.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-reinforcement-learning","name":"Reinforcement learning","explanation":"Training by trial and error, in which a system is rewarded for the outcomes its designers want.","referenceUrl":"https://developers.google.com/machine-learning/glossary#reinforcement-learning-rl"}],"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":"73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9","revision":"98932b57-4924-474d-82eb-ae8c8b0b9b8b","revisionKind":"publication-id","title":"Xuliang Huang and coauthors propose GIST for selecting offline RLVR training data","url":"https://escapemeridian.com/readings/xuliang-huang-and-coauthors-propose-gist-for-selecting-offline-73e73941"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.06395","revision":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","revisionKind":"sha256-source-text","title":"GIST: Gibbs-Targeted Hierarchical Data Selection for Offline RLVR","url":"https://arxiv.org/abs/2611.06395"},"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":"73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9","title":"Xuliang Huang and coauthors propose GIST for selecting offline RLVR training data","argument":"Authors Xuliang Huang, Tianxiang Zhao, Chengwei Qin, and Yao Shu argue that offline reinforcement learning with verifiable rewards (RLVR) should not supervise every response and token equally. Their analysis uses the Gibbs-optimal policy of KL-regularized RLVR to identify where a policy differs from its training target: correct responses matter mainly through low-probability tokens, while incorrect responses matter through frequently generated responses and their high-probability tokens. They propose GIST, a two-level data-selection method that uses sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode, then uses initial token probabilities to select tokens for supervision. The authors report higher average accuracy across GSM8K, MATH-500, and MBPP for all three models under two offline training objectives. In a controlled Qwen3-0.6B/GSM8K comparison, they report that GIST outperforms seven selection baselines using the same candidates while supervising 66.6% fewer tokens than training on all candidates of the same prompts. The reported results concern these evaluations; whether the gains and token savings carry over to other settings remains open.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The evidence supplied is the abstract, which does not identify the three models or name the two offline training objectives. The 66.6% token reduction and comparison with seven baselines are reported for Qwen3-0.6B/GSM8K, not for every evaluation described.","speaker":{"name":"Huang, Xuliang, Zhao, Tianxiang, Qin, Chengwei, Shu, Yao","roleAtSource":"Article author"},"source":{"title":"GIST: Gibbs-Targeted Hierarchical Data Selection for Offline RLVR","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.06395","publishedAt":"2026-11-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors analyze which responses and tokens contribute to mismatch between a policy and the Gibbs-optimal policy of KL-regularized RLVR.","excerpt":"measure the mismatch between a policy and this target with a bidirectional KL divergence","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"The authors report that correct and incorrect responses contribute to the mismatch through different tokens and response patterns.","excerpt":"correct responses contribute mainly through the forward term, concentrated on their low-probability tokens","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"GIST uses sampling temperature and initial policy token probabilities to select responses and tokens for supervision before training.","excerpt":"GIST uses the sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"The authors report higher average accuracy across three benchmarks for all three models under both offline training objectives.","excerpt":"GIST raises the average accuracy over GSM8K, MATH-500, and MBPP for all three models under both objectives.","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"},{"text":"In one controlled comparison, the authors report that GIST beats seven selection baselines while supervising fewer tokens.","excerpt":"it outperforms seven selection baselines that choose from the same candidates while supervising 66.6% fewer tokens","url":"https://arxiv.org/abs/2611.06395","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.06395","checkedAt":"2026-11-09T11:40:11.118Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do GIST's reported accuracy gains and token savings hold beyond the models, objectives, tasks, and controlled comparison described in the abstract?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Xuliang Huang and coauthors propose GIST for selecting offline RLVR training data","summary":"Authors Xuliang Huang, Tianxiang Zhao, Chengwei Qin, and Yao Shu argue that offline reinforcement learning with verifiable rewards (RLVR) should not supervise every response and token equally. Their analysis uses the Gibbs-optimal policy of KL-regularized RLVR to identify where a policy differs from its training target: correct responses matter mainly through low-probability tokens, while incorrect responses matter through frequently generated responses and their high-probability tokens. They propose GIST, a two-level data-selection method that uses sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode, then uses initial token probabilities to select tokens for supervision. The authors report higher average accuracy across GSM8K, MATH-500, and MBPP for all three models under two offline training objectives. In a controlled Qwen3-0.6B/GSM8K comparison, they report that GIST outperforms seven selection baselines using the same candidates while supervising 66.6% fewer tokens than training on all candidates of the same prompts. The reported results concern these evaluations; whether the gains and token savings carry over to other settings remains open.","question":"Do GIST's reported accuracy gains and token savings hold beyond the models, objectives, tasks, and controlled comparison described in the abstract?","claims":[{"text":"The authors analyze which responses and tokens contribute to mismatch between a policy and the Gibbs-optimal policy of KL-regularized RLVR.","excerpt":"measure the mismatch between a policy and this target with a bidirectional KL divergence","support":"source-stated"},{"text":"The authors report that correct and incorrect responses contribute to the mismatch through different tokens and response patterns.","excerpt":"correct responses contribute mainly through the forward term, concentrated on their low-probability tokens","support":"source-stated"},{"text":"GIST uses sampling temperature and initial policy token probabilities to select responses and tokens for supervision before training.","excerpt":"GIST uses the sampling temperature to select correct responses on-policy and incorrect responses near the policy's mode","support":"source-stated"},{"text":"The authors report higher average accuracy across three benchmarks for all three models under both offline training objectives.","excerpt":"GIST raises the average accuracy over GSM8K, MATH-500, and MBPP for all three models under both objectives.","support":"source-stated"},{"text":"In one controlled comparison, the authors report that GIST beats seven selection baselines while supervising fewer tokens.","excerpt":"it outperforms seven selection baselines that choose from the same candidates while supervising 66.6% fewer tokens","support":"source-stated"}],"limitations":["The evidence supplied is the abstract, which does not identify the three models or name the two offline training objectives.","The 66.6% token reduction and comparison with seven baselines are reported for Qwen3-0.6B/GSM8K, not for every evaluation described."],"mediaId":"14229db44999bbe3b4416df794ec9ca114ed9bcd24923dd2bfd1706bad5b4964","correctionReason":null,"id":"73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9","publicationId":"98932b57-4924-474d-82eb-ae8c8b0b9b8b","version":1,"sourceUrl":"https://arxiv.org/abs/2611.06395","sourceTitle":"GIST: Gibbs-Targeted Hierarchical Data Selection for Offline RLVR","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-05T05:00:31.762Z","createdAt":"2026-11-09T11:09:58.800Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"14229db44999bbe3b4416df794ec9ca114ed9bcd24923dd2bfd1706bad5b4964","kind":"article","url":"https://arxiv.org/abs/2611.06395","sourceUrl":"https://arxiv.org/abs/2611.06395","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T11:09:13.996Z","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/2611.06395","checkedAt":"2026-11-09T11:40:11.118Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"14229db44999bbe3b4416df794ec9ca114ed9bcd24923dd2bfd1706bad5b4964","confidence":0.85,"provider":"typesafe","model":"jev-1.13.0","inputHash":"662c773ae35276ce1a288f3d62784edbb57827a2c8b166ecfb9d391b1d68ebb5","decidedAt":"2026-11-09T11:09:15.539Z"},"provenance":{"textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observationId":"2584d49706550c5bd850a3a235fee5c97e4650bcdcd62e10ad4140e3e330204a","retrievedAt":"2026-11-09T11:09:13.740Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"a06a6323-7750-4336-85bd-3723fd60c4d2"},"reviewer":{"model":"gpt-6-luna","runId":"c4ee8708-0e2f-472f-8f91-002d026eda8c"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T11:09:58.800Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Xuliang","url":null,"kind":"unknown"},{"name":"Zhao, Tianxiang","url":null,"kind":"unknown"},{"name":"Qin, Chengwei","url":null,"kind":"unknown"},{"name":"Shu, Yao","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Xuliang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Xuliang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}},{"actor":{"name":"Zhao, Tianxiang","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Zhao, Tianxiang","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}},{"actor":{"name":"Qin, Chengwei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Qin, Chengwei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}},{"actor":{"name":"Shu, Yao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Shu, Yao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.06395","locator":"head/meta[name=author|citation_author]"},"assertionId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T11:40:11.118Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","sourceUrl":"https://arxiv.org/abs/2611.06395","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.06395","snapshotId":"212a85dcd74495972132a9aa4eb0bc0f79a9cf021996ee919cb6805126e290f4","contentSha256":"9e2afa327e2c36bf3234631a4ad43a69e5654b7f62f4e1023a1430d8da07b56d","textHash":"908effa00d9bcf381f00d38ac11a3b8e9a6e6fababa08da6698dcb9e9fc67a33","observedAt":"2026-11-09T11:40:11.118Z","basis":"publisher-page-metadata","sourceLinks":43,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/73e73941a496ff607581535b84a481ef7ed35f8a1c2c55f37e730f25f68d5dd9/source"}}}],"audits":[{"publicationId":"98932b57-4924-474d-82eb-ae8c8b0b9b8b","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading credits the paper and its findings to the four listed arXiv authors and frames the results as author reports. This matches the source.","excerpt":"Authors:Xuliang Huang, Tianxiang Zhao, Chengwei Qin, Yao Shu"},{"dimension":"summary","status":"uncertain","explanation":"The summary follows the abstract, but its core is unexplained jargon: Gibbs-optimal policy, KL-regularized, policy, tokens, on-policy, policy's mode, and the benchmarks GSM8K, MATH-500 and MBPP. RLVR is spelled out but never described in plain terms. A general reader cannot tell what changed or why fewer supervised tokens might matter. It also drops two scope conditions: the analysis holds exactly only at the reference policy, and GIST approximates those principles. Smallest repairs: add one plain sentence on what RLVR training is, add tooltip entries for the listed terms, and restore 'at the reference policy' and 'approximates'.","excerpt":"At the reference policy, this mismatch factorizes exactly and separates the two outcomes"},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-11-04 matches the arXiv v1 submission date. The feed date of Nov 5 is a later announcement.","excerpt":"[Submitted on 4 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches the quoted abstract text. The 66.6% figure is correctly scoped to the Qwen3-0.6B/GSM8K comparison. The limitations correctly note that the three models and two objectives are not named. The open question does not overstate the results.","excerpt":"In a controlled comparison on Qwen3-0.6B/GSM8K, it outperforms seven selection baselines that choose from the same candidates while supervising 66.6% fewer tokens"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct way to reach the source.","excerpt":null}],"createdAt":"2026-11-09T11:09:58.800Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}