{"id":"fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650","title":"Wang and coauthors propose adaptive zooming for GUI grounding","argument":"In an arXiv paper, Bingchao Wang and coauthors argue that GUI grounding on high-resolution screens often fails on small targets, and that common zooming methods do not adapt zoom to target size or check and revisit incorrect answers. They propose AdaGround, a training-free framework that estimates target size to set zoom, uses a Yes/No verifier to keep or reject zooms and trigger another grounding attempt, and can abstain after an existence query. A third component composes spreadsheet cells from their headers.\n\nThe authors report that, on Qwen3-VL models, AdaGround improves accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and that gains carry over to held-out UI-Vision. They also say it surpasses state-of-the-art training-free zooming methods and many trained GUI models. These are results reported in the paper’s abstract; the supplied page does not provide details needed to assess the experiments or how the benchmark results relate to real-world use.\n\nHow does the verifier’s performance affect the reported gains and the decision to abstain?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the arXiv abstract and page text, not the paper’s experimental details; it does not show how the benchmarks were run or support conclusions about real-world performance. The abstract reports results on Qwen3-VL models but does not specify the verifier’s performance or how it affects the gains and abstention decisions.","speaker":{"name":"Wang, Bingchao, Jing, Yongcheng, Wang, Chengyu, Huang, Jun, Yang, Jie, Liu, Wei, Tao, Dacheng","roleAtSource":"Article author"},"source":{"title":"AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI Grounding","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.36874","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors say existing zooming pipelines do not adapt zoom to target size, accept zoomed answers unconditionally, and do not revisit wrong answers.","excerpt":"most existing zooming pipelines do not match the zoom to the target's size, accept the zoomed answer unconditionally and never revisit a wrong answer","url":"https://arxiv.org/abs/2610.36874","relationship":"Source-stated"},{"text":"AdaGround uses an estimate of target size to set zoom parameters and a Yes/No verifier to guide refinement, re-grounding, and abstention.","excerpt":"uses one marked-click Yes/No verifier to keep or reject each zoom, choose between zoom chains, trigger re-grounding and, with an existence query, decide when to abstain","url":"https://arxiv.org/abs/2610.36874","relationship":"Source-stated"},{"text":"The authors report accuracy improvements over the backbone of 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, with gains carrying over to held-out UI-Vision.","excerpt":"improving accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and its gains carry over to held-out UI-Vision","url":"https://arxiv.org/abs/2610.36874","relationship":"Source-stated"},{"text":"The supplied abstract does not provide experimental details sufficient to assess the reported results or their relationship to real-world use.","excerpt":"On Qwen3-VL models, AdaGround surpasses state-of-the-art training-free zooming methods and many trained GUI models","url":"https://arxiv.org/abs/2610.36874","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.36874","checkedAt":"2026-11-14T21:06:10.348Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How does the verifier’s performance affect the reported gains and the decision to abstain?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Wang and coauthors propose adaptive zooming for GUI grounding","summary":"In an arXiv paper, Bingchao Wang and coauthors argue that GUI grounding on high-resolution screens often fails on small targets, and that common zooming methods do not adapt zoom to target size or check and revisit incorrect answers. They propose AdaGround, a training-free framework that estimates target size to set zoom, uses a Yes/No verifier to keep or reject zooms and trigger another grounding attempt, and can abstain after an existence query. A third component composes spreadsheet cells from their headers.\n\nThe authors report that, on Qwen3-VL models, AdaGround improves accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and that gains carry over to held-out UI-Vision. They also say it surpasses state-of-the-art training-free zooming methods and many trained GUI models. These are results reported in the paper’s abstract; the supplied page does not provide details needed to assess the experiments or how the benchmark results relate to real-world use.\n\nHow does the verifier’s performance affect the reported gains and the decision to abstain?","question":"How does the verifier’s performance affect the reported gains and the decision to abstain?","claims":[{"text":"The authors say existing zooming pipelines do not adapt zoom to target size, accept zoomed answers unconditionally, and do not revisit wrong answers.","excerpt":"most existing zooming pipelines do not match the zoom to the target's size, accept the zoomed answer unconditionally and never revisit a wrong answer","support":"source-stated"},{"text":"AdaGround uses an estimate of target size to set zoom parameters and a Yes/No verifier to guide refinement, re-grounding, and abstention.","excerpt":"uses one marked-click Yes/No verifier to keep or reject each zoom, choose between zoom chains, trigger re-grounding and, with an existence query, decide when to abstain","support":"source-stated"},{"text":"The authors report accuracy improvements over the backbone of 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, with gains carrying over to held-out UI-Vision.","excerpt":"improving accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and its gains carry over to held-out UI-Vision","support":"source-stated"},{"text":"The supplied abstract does not provide experimental details sufficient to assess the reported results or their relationship to real-world use.","excerpt":"On Qwen3-VL models, AdaGround surpasses state-of-the-art training-free zooming methods and many trained GUI models","support":"interpretation"}],"limitations":["The supplied evidence is the arXiv abstract and page text, not the paper’s experimental details; it does not show how the benchmarks were run or support conclusions about real-world performance.","The abstract reports results on Qwen3-VL models but does not specify the verifier’s performance or how it affects the gains and abstention decisions."],"mediaId":"a0d94e244ab25515fb29a8871c51e1b7b6a80e8959d2f5939af99fe61e581200","correctionReason":null,"id":"fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650","publicationId":"3689da61-e519-45be-862a-e211e3c4307b","version":1,"sourceUrl":"https://arxiv.org/abs/2610.36874","sourceTitle":"AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI Grounding","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T07:14:37.613Z","createdAt":"2026-11-14T04:03:12.722Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"a0d94e244ab25515fb29a8871c51e1b7b6a80e8959d2f5939af99fe61e581200","kind":"article","url":"https://arxiv.org/abs/2610.36874","sourceUrl":"https://arxiv.org/abs/2610.36874","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T23:20:02.002Z","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/2610.36874","checkedAt":"2026-11-14T21:06:10.348Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"a0d94e244ab25515fb29a8871c51e1b7b6a80e8959d2f5939af99fe61e581200","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"3a7a0bdaba75793994f551f660de4230662077b72f1b122e8220b3a8b1945f04","decidedAt":"2026-10-30T23:20:06.606Z"},"provenance":{"textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observationId":"6d9fcb68cc8f79ceedd11c58b6922e5472ba53ca9aedb4edb3063579f0144341","retrievedAt":"2026-10-30T23:20:00.299Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"21245db4-8fc4-47f8-884a-8f2a3d04d25f"},"reviewer":{"model":"gpt-6-luna","runId":"ff471dd4-f88d-42f1-8f32-85e93355c040"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-14T04:03:12.722Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Wang, Bingchao","url":null,"kind":"unknown"},{"name":"Jing, Yongcheng","url":null,"kind":"unknown"},{"name":"Wang, Chengyu","url":null,"kind":"unknown"},{"name":"Huang, Jun","url":null,"kind":"unknown"},{"name":"Yang, Jie","url":null,"kind":"unknown"},{"name":"Liu, Wei","url":null,"kind":"unknown"},{"name":"Tao, Dacheng","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Wang, Bingchao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Bingchao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Jing, Yongcheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jing, Yongcheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Wang, Chengyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Chengyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Huang, Jun","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Jun","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Yang, Jie","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yang, Jie","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Liu, Wei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Wei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Tao, Dacheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Tao, Dacheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}}],"participationCoverage":{"total":7,"returned":7,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.36874","snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z","basis":"publisher-page-metadata","sourceLinks":48,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650/source"}},"canonicalUrl":"https://escapemeridian.com/readings/wang-and-coauthors-propose-adaptive-zooming-for-gui-grounding-fef1c11e","formats":{"html":"/readings/wang-and-coauthors-propose-adaptive-zooming-for-gui-grounding-fef1c11e","markdown":"/readings/wang-and-coauthors-propose-adaptive-zooming-for-gui-grounding-fef1c11e.md","json":"/readings/wang-and-coauthors-propose-adaptive-zooming-for-gui-grounding-fef1c11e.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-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":"fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650","revision":"3689da61-e519-45be-862a-e211e3c4307b","revisionKind":"publication-id","title":"Wang and coauthors propose adaptive zooming for GUI grounding","url":"https://escapemeridian.com/readings/wang-and-coauthors-propose-adaptive-zooming-for-gui-grounding-fef1c11e"},"source":{"type":"source","id":"https://arxiv.org/abs/2610.36874","revision":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","revisionKind":"sha256-source-text","title":"AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI Grounding","url":"https://arxiv.org/abs/2610.36874"},"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":"fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650","title":"Wang and coauthors propose adaptive zooming for GUI grounding","argument":"In an arXiv paper, Bingchao Wang and coauthors argue that GUI grounding on high-resolution screens often fails on small targets, and that common zooming methods do not adapt zoom to target size or check and revisit incorrect answers. They propose AdaGround, a training-free framework that estimates target size to set zoom, uses a Yes/No verifier to keep or reject zooms and trigger another grounding attempt, and can abstain after an existence query. A third component composes spreadsheet cells from their headers.\n\nThe authors report that, on Qwen3-VL models, AdaGround improves accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and that gains carry over to held-out UI-Vision. They also say it surpasses state-of-the-art training-free zooming methods and many trained GUI models. These are results reported in the paper’s abstract; the supplied page does not provide details needed to assess the experiments or how the benchmark results relate to real-world use.\n\nHow does the verifier’s performance affect the reported gains and the decision to abstain?","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence is the arXiv abstract and page text, not the paper’s experimental details; it does not show how the benchmarks were run or support conclusions about real-world performance. The abstract reports results on Qwen3-VL models but does not specify the verifier’s performance or how it affects the gains and abstention decisions.","speaker":{"name":"Wang, Bingchao, Jing, Yongcheng, Wang, Chengyu, Huang, Jun, Yang, Jie, Liu, Wei, Tao, Dacheng","roleAtSource":"Article author"},"source":{"title":"AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI Grounding","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.36874","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors say existing zooming pipelines do not adapt zoom to target size, accept zoomed answers unconditionally, and do not revisit wrong answers.","excerpt":"most existing zooming pipelines do not match the zoom to the target's size, accept the zoomed answer unconditionally and never revisit a wrong answer","url":"https://arxiv.org/abs/2610.36874","relationship":"Source-stated"},{"text":"AdaGround uses an estimate of target size to set zoom parameters and a Yes/No verifier to guide refinement, re-grounding, and abstention.","excerpt":"uses one marked-click Yes/No verifier to keep or reject each zoom, choose between zoom chains, trigger re-grounding and, with an existence query, decide when to abstain","url":"https://arxiv.org/abs/2610.36874","relationship":"Source-stated"},{"text":"The authors report accuracy improvements over the backbone of 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, with gains carrying over to held-out UI-Vision.","excerpt":"improving accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and its gains carry over to held-out UI-Vision","url":"https://arxiv.org/abs/2610.36874","relationship":"Source-stated"},{"text":"The supplied abstract does not provide experimental details sufficient to assess the reported results or their relationship to real-world use.","excerpt":"On Qwen3-VL models, AdaGround surpasses state-of-the-art training-free zooming methods and many trained GUI models","url":"https://arxiv.org/abs/2610.36874","relationship":"Escape Meridian interpretation"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.36874","checkedAt":"2026-11-14T21:06:10.348Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How does the verifier’s performance affect the reported gains and the decision to abstain?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Wang and coauthors propose adaptive zooming for GUI grounding","summary":"In an arXiv paper, Bingchao Wang and coauthors argue that GUI grounding on high-resolution screens often fails on small targets, and that common zooming methods do not adapt zoom to target size or check and revisit incorrect answers. They propose AdaGround, a training-free framework that estimates target size to set zoom, uses a Yes/No verifier to keep or reject zooms and trigger another grounding attempt, and can abstain after an existence query. A third component composes spreadsheet cells from their headers.\n\nThe authors report that, on Qwen3-VL models, AdaGround improves accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and that gains carry over to held-out UI-Vision. They also say it surpasses state-of-the-art training-free zooming methods and many trained GUI models. These are results reported in the paper’s abstract; the supplied page does not provide details needed to assess the experiments or how the benchmark results relate to real-world use.\n\nHow does the verifier’s performance affect the reported gains and the decision to abstain?","question":"How does the verifier’s performance affect the reported gains and the decision to abstain?","claims":[{"text":"The authors say existing zooming pipelines do not adapt zoom to target size, accept zoomed answers unconditionally, and do not revisit wrong answers.","excerpt":"most existing zooming pipelines do not match the zoom to the target's size, accept the zoomed answer unconditionally and never revisit a wrong answer","support":"source-stated"},{"text":"AdaGround uses an estimate of target size to set zoom parameters and a Yes/No verifier to guide refinement, re-grounding, and abstention.","excerpt":"uses one marked-click Yes/No verifier to keep or reject each zoom, choose between zoom chains, trigger re-grounding and, with an existence query, decide when to abstain","support":"source-stated"},{"text":"The authors report accuracy improvements over the backbone of 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, with gains carrying over to held-out UI-Vision.","excerpt":"improving accuracy over the backbone by 7.4 to 20.6 points on OSWorld-G and ScreenSpot-Pro, and its gains carry over to held-out UI-Vision","support":"source-stated"},{"text":"The supplied abstract does not provide experimental details sufficient to assess the reported results or their relationship to real-world use.","excerpt":"On Qwen3-VL models, AdaGround surpasses state-of-the-art training-free zooming methods and many trained GUI models","support":"interpretation"}],"limitations":["The supplied evidence is the arXiv abstract and page text, not the paper’s experimental details; it does not show how the benchmarks were run or support conclusions about real-world performance.","The abstract reports results on Qwen3-VL models but does not specify the verifier’s performance or how it affects the gains and abstention decisions."],"mediaId":"a0d94e244ab25515fb29a8871c51e1b7b6a80e8959d2f5939af99fe61e581200","correctionReason":null,"id":"fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650","publicationId":"3689da61-e519-45be-862a-e211e3c4307b","version":1,"sourceUrl":"https://arxiv.org/abs/2610.36874","sourceTitle":"AdaGround: Token-Normalized Adaptive Zooming with Verifier-in-the-Loop Refinement for GUI Grounding","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T07:14:37.613Z","createdAt":"2026-11-14T04:03:12.722Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"a0d94e244ab25515fb29a8871c51e1b7b6a80e8959d2f5939af99fe61e581200","kind":"article","url":"https://arxiv.org/abs/2610.36874","sourceUrl":"https://arxiv.org/abs/2610.36874","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T23:20:02.002Z","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/2610.36874","checkedAt":"2026-11-14T21:06:10.348Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"a0d94e244ab25515fb29a8871c51e1b7b6a80e8959d2f5939af99fe61e581200","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"3a7a0bdaba75793994f551f660de4230662077b72f1b122e8220b3a8b1945f04","decidedAt":"2026-10-30T23:20:06.606Z"},"provenance":{"textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observationId":"6d9fcb68cc8f79ceedd11c58b6922e5472ba53ca9aedb4edb3063579f0144341","retrievedAt":"2026-10-30T23:20:00.299Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"21245db4-8fc4-47f8-884a-8f2a3d04d25f"},"reviewer":{"model":"gpt-6-luna","runId":"ff471dd4-f88d-42f1-8f32-85e93355c040"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-14T04:03:12.722Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Wang, Bingchao","url":null,"kind":"unknown"},{"name":"Jing, Yongcheng","url":null,"kind":"unknown"},{"name":"Wang, Chengyu","url":null,"kind":"unknown"},{"name":"Huang, Jun","url":null,"kind":"unknown"},{"name":"Yang, Jie","url":null,"kind":"unknown"},{"name":"Liu, Wei","url":null,"kind":"unknown"},{"name":"Tao, Dacheng","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Wang, Bingchao","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Bingchao","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Jing, Yongcheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Jing, Yongcheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Wang, Chengyu","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Wang, Chengyu","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Huang, Jun","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Jun","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Yang, Jie","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Yang, Jie","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Liu, Wei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Wei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}},{"actor":{"name":"Tao, Dacheng","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Tao, Dacheng","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36874","locator":"head/meta[name=author|citation_author]"},"assertionId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T21:06:10.348Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","sourceUrl":"https://arxiv.org/abs/2610.36874","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z"}}],"participationCoverage":{"total":7,"returned":7,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.36874","snapshotId":"0b99d2b721712389c80b3936b98d5552d14275ea09dd317baaa9d1f2f9f3acbc","contentSha256":"38f65f057e3dee1992c440428ff8a0b18d35dc333814e4a4cc6f10e2eae00bc6","textHash":"0a097a6c7366bbf4c4c05cd6b2fa702c7d259bef7dcb18bd064a1336a45b306c","observedAt":"2026-11-14T21:06:10.348Z","basis":"publisher-page-metadata","sourceLinks":48,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/fef1c11e87a717d8b2301839eaaa3acf9d7cf1871b9c26083e49f992acad5650/source"}}}],"audits":[{"publicationId":"3689da61-e519-45be-862a-e211e3c4307b","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the problem framing, method and results to Bingchao Wang and coauthors in an arXiv paper. It presents benchmark gains as reported by the authors, not as established facts.","excerpt":"Authors:Bingchao Wang, Yongcheng Jing, Chengyu Wang, Jun Huang, Jie Yang, Wei Liu, Dacheng Tao"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but its central term, GUI grounding, is never explained in plain language. A general reader will not know it means an AI model locating the on-screen element it is told to act on. Training-free, backbone and Qwen3-VL are also unexplained, and the reading never says why the task matters. Smallest repair: add one plain sentence defining GUI grounding, and say what training-free means here (no model retraining). Then tooltip the other terms.","excerpt":"GUI grounding on high-resolution screens fails mostly on small targets, and zooming in is a common training-free remedy."},{"dimension":"date","status":"pass","explanation":"The originalPublished date of 2026-10-27 matches the arXiv v1 submission date. The later feed date is correctly not used.","excerpt":"[Submitted on 27 Oct 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim is source-stated and has a matching excerpt. The limitation about missing experimental detail is properly framed as interpretation. The question about verifier performance follows from the verifier's central role in the method and does not presuppose an answer. No benefits or costs are invented.","excerpt":"uses one marked-click Yes/No verifier to keep or reject each zoom, choose between zoom chains, trigger re-grounding and, with an existence query, decide when to abstain"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the original arXiv abstract page. This is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-14T04:03:12.722Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}