{"paper":{"title":"The Cartesian Shortcut: Re-evaluate Vision Reasoning in Polar Coordinate Space","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Current multimodal models achieve high visual reasoning scores by exploiting grid-based coordinates rather than understanding spatial relationships directly.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Brian Potetz, Chun-Ta Lu, Howard Zhou, Leonidas Guibas, Xia Hu, Zhenrui Yue, Zhicheng Wang","submitted_at":"2026-05-11T02:16:48Z","abstract_excerpt":"As current Multimodal Large Language Models rapidly saturate canonical visual reasoning benchmarks, a key question emerges: do these strong scores genuinely reflect robust visual understanding? We identify a pervasive vulnerability, the Cartesian Shortcut: visual reasoning benchmarks prevalently build on orthogonal grid-based layouts that can be readily discretized into explicit textual coordinates. Models systematically exploit this property, heavily leveraging text-based deductive reasoning to assist visual problem-solving. To systematically dismantle this shortcut, we introduce Polaris-Benc"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"frontier models achieving 70--83% on Cartesian layouts collapse to 31--39% on Polar equivalents, with degradation persisting even under complete logical equivalence.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"Re-formulating the 53 tasks in polar coordinates preserves identical logical constraints, task semantics, and difficulty levels without introducing unrelated visual or reasoning challenges.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MLLMs scoring 70-83% on Cartesian visual tasks drop to 31-39% on logically equivalent polar versions, exposing reliance on grid discretization shortcuts instead of topology-invariant reasoning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Current multimodal models achieve high visual reasoning scores by exploiting grid-based coordinates rather than understanding spatial relationships directly.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1d022bee5ea0b0d7c81f9907a34cec144035915baa3faf74646dfeb08409fbbd"},"source":{"id":"2605.09883","kind":"arxiv","version":2},"verdict":{"id":"f780797c-db83-4d4e-9bcd-12fd49b333ac","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T04:27:29.877728Z","strongest_claim":"frontier models achieving 70--83% on Cartesian layouts collapse to 31--39% on Polar equivalents, with degradation persisting even under complete logical equivalence.","one_line_summary":"MLLMs scoring 70-83% on Cartesian visual tasks drop to 31-39% on logically equivalent polar versions, exposing reliance on grid discretization shortcuts instead of topology-invariant reasoning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"Re-formulating the 53 tasks in polar coordinates preserves identical logical constraints, task semantics, and difficulty levels without introducing unrelated visual or reasoning challenges.","pith_extraction_headline":"Current multimodal models achieve high visual reasoning scores by exploiting grid-based coordinates rather than understanding spatial relationships directly."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09883/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T07:02:01.307430Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T16:34:30.941152Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:17.319580Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:51:27.939939Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"22823f736d95f2816c8dfba27a63b480d8dbb0cc161c220fce79e14a0e4df8da"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"73d59aaa90c4a5e5c14f1de4fcfd4a0c4c82cbc1a9f786a0ea9c9b99483cca57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}