{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:U2GPRN34O2GYARZJ26TNK44DDO","short_pith_number":"pith:U2GPRN34","canonical_record":{"source":{"id":"2605.14068","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T19:46:22Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"0d9c0595ceb9bb930acd406796590d6f7502ec2fc2934690b5f53833a1b18430","abstract_canon_sha256":"4dd9faab3424e6d7b65351d4013f730184889dac501b2a3eec4572b170e7b9b8"},"schema_version":"1.0"},"canonical_sha256":"a68cf8b77c768d804729d7a6d573831b8f3f6eae26abd474e5d76efecae0380b","source":{"kind":"arxiv","id":"2605.14068","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14068","created_at":"2026-05-17T23:39:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14068v1","created_at":"2026-05-17T23:39:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14068","created_at":"2026-05-17T23:39:12Z"},{"alias_kind":"pith_short_12","alias_value":"U2GPRN34O2GY","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"U2GPRN34O2GYARZJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"U2GPRN34","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:U2GPRN34O2GYARZJ26TNK44DDO","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14068","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T19:46:22Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"0d9c0595ceb9bb930acd406796590d6f7502ec2fc2934690b5f53833a1b18430","abstract_canon_sha256":"4dd9faab3424e6d7b65351d4013f730184889dac501b2a3eec4572b170e7b9b8"},"schema_version":"1.0"},"canonical_sha256":"a68cf8b77c768d804729d7a6d573831b8f3f6eae26abd474e5d76efecae0380b","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:12.453369Z","signature_b64":"g/0p3Y2LKwLz1cv8CiEUrQrxe+CKf05Z6YUS1AoU7IavQ1rljVvfJSHBv6UTo3NQaH3dao6/oJ4umkT3bC0dBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a68cf8b77c768d804729d7a6d573831b8f3f6eae26abd474e5d76efecae0380b","last_reissued_at":"2026-05-17T23:39:12.452745Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:12.452745Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14068","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"V0pnaxKGhBmGnUFJRVUqlrppDh+DsMdsLtAgvCGMnbX6iBS6ch830KnJCzwC4KyRt1pO/YsQvpS2G+gfEciTDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-04T20:16:09.154926Z"},"content_sha256":"4982d7bfd7b7b7cd04d99eab1bf270d8c46560785dc14e7428f9bbb66b151b96","schema_version":"1.0","event_id":"sha256:4982d7bfd7b7b7cd04d99eab1bf270d8c46560785dc14e7428f9bbb66b151b96"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:U2GPRN34O2GYARZJ26TNK44DDO","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"CurveBench: A Benchmark for Exact Topological Reasoning over Nested Jordan Curves","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Vision-language models recover exact containment trees from nested Jordan curves at only 71 percent accuracy on easy cases and 19 percent on hard cases.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Amirreza Mohseni, Mona Mohammadi, Morteza Saghafian, Naser Talebizadeh Saradari","submitted_at":"2026-05-13T19:46:22Z","abstract_excerpt":"We introduce CurveBench, a benchmark for hierarchical topological reasoning from visual input. CurveBench consists of \\textbf{756 images} of pairwise non-intersecting Jordan curves across easy, polygonal, topographic-inspired, maze-like, and dense counting configurations. Each image is annotated with a rooted tree encoding the containment relations between planar regions. We formulate the task as structured prediction: given an image, a model must recover the full rooted containment tree induced by the curves. Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1 Pro, achieves only 71.1% tree-generation accuracy on CurveBench-Easy and 19.1% on CurveBench-Hard.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the generated images and their tree annotations faithfully isolate topological containment reasoning without providing unintended low-level visual shortcuts or dataset-specific biases that models could exploit.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CurveBench benchmark reveals that even leading VLMs like Gemini 3.1 Pro reach only 71.1% accuracy recovering containment trees on easy nested-curve images and 19.1% on hard versions, while fine-tuning lifts an open 8B model to 33.3% on easy cases.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Vision-language models recover exact containment trees from nested Jordan curves at only 71 percent accuracy on easy cases and 19 percent on hard cases.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8c959dfc07e0289e15e4f7997f90f45006c7d07fc7f9ea0745f0e98877071448"},"source":{"id":"2605.14068","kind":"arxiv","version":1},"verdict":{"id":"f70b9cb0-297f-4075-af14-507829a0b5bc","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:30:49.710142Z","strongest_claim":"Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1 Pro, achieves only 71.1% tree-generation accuracy on CurveBench-Easy and 19.1% on CurveBench-Hard.","one_line_summary":"CurveBench benchmark reveals that even leading VLMs like Gemini 3.1 Pro reach only 71.1% accuracy recovering containment trees on easy nested-curve images and 19.1% on hard versions, while fine-tuning lifts an open 8B model to 33.3% on easy cases.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the generated images and their tree annotations faithfully isolate topological containment reasoning without providing unintended low-level visual shortcuts or dataset-specific biases that models could exploit.","pith_extraction_headline":"Vision-language models recover exact containment trees from nested Jordan curves at only 71 percent accuracy on easy cases and 19 percent on hard cases."},"references":{"count":44,"sample":[{"doi":"","year":null,"title":"Topographic Map Symbols , year =","work_id":"97537c87-e528-4e49-8aa6-c4a856a8e769","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Jordan Theorem , year =","work_id":"1b97f40c-b303-44f0-be1b-993511763d99","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Journal of Visual Languages and Computing , volume =","work_id":"9b902288-1e92-4908-a0e0-08556ec84269","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Findings of the Association for Computational Linguistics: ACL 2022 , pages =","work_id":"80430d48-4070-43de-b886-0d69668c33c3","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Mathew, Minesh and Karatzas, Dimosthenis and Jawahar, C. V. , title =. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages =. 2021 , url =","work_id":"0289afd3-d25b-479c-a341-b704cb979319","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":44,"snapshot_sha256":"80adee1f36ded0afa842dac20456da112d1fe35319c75a9dde85151a0b83d11c","internal_anchors":10},"formal_canon":{"evidence_count":1,"snapshot_sha256":"977f3def4313ba2ec4124af84af93ea5a60e0a73a6b56c7952340acbff025b50"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"f70b9cb0-297f-4075-af14-507829a0b5bc"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:12Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"PIcra3X1RKEYgQHQh5Ju74e/ZwH0nB2HO22Zas0XFs1Gn0eh18Z5GJz8EMx7idtil+AiaD0It6BwGsk5TvR+Cg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-04T20:16:09.155482Z"},"content_sha256":"97a23fa8e969c70697c9f4c7dbc1b304723ae00c90f27b76b5e22d02fba8cf25","schema_version":"1.0","event_id":"sha256:97a23fa8e969c70697c9f4c7dbc1b304723ae00c90f27b76b5e22d02fba8cf25"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/U2GPRN34O2GYARZJ26TNK44DDO/bundle.json","state_url":"https://pith.science/pith/U2GPRN34O2GYARZJ26TNK44DDO/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/U2GPRN34O2GYARZJ26TNK44DDO/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-04T20:16:09Z","links":{"resolver":"https://pith.science/pith/U2GPRN34O2GYARZJ26TNK44DDO","bundle":"https://pith.science/pith/U2GPRN34O2GYARZJ26TNK44DDO/bundle.json","state":"https://pith.science/pith/U2GPRN34O2GYARZJ26TNK44DDO/state.json","well_known_bundle":"https://pith.science/.well-known/pith/U2GPRN34O2GYARZJ26TNK44DDO/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:U2GPRN34O2GYARZJ26TNK44DDO","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"4dd9faab3424e6d7b65351d4013f730184889dac501b2a3eec4572b170e7b9b8","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T19:46:22Z","title_canon_sha256":"0d9c0595ceb9bb930acd406796590d6f7502ec2fc2934690b5f53833a1b18430"},"schema_version":"1.0","source":{"id":"2605.14068","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14068","created_at":"2026-05-17T23:39:12Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14068v1","created_at":"2026-05-17T23:39:12Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14068","created_at":"2026-05-17T23:39:12Z"},{"alias_kind":"pith_short_12","alias_value":"U2GPRN34O2GY","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"U2GPRN34O2GYARZJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"U2GPRN34","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:97a23fa8e969c70697c9f4c7dbc1b304723ae00c90f27b76b5e22d02fba8cf25","target":"graph","created_at":"2026-05-17T23:39:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1 Pro, achieves only 71.1% tree-generation accuracy on CurveBench-Easy and 19.1% on CurveBench-Hard."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the generated images and their tree annotations faithfully isolate topological containment reasoning without providing unintended low-level visual shortcuts or dataset-specific biases that models could exploit."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"CurveBench benchmark reveals that even leading VLMs like Gemini 3.1 Pro reach only 71.1% accuracy recovering containment trees on easy nested-curve images and 19.1% on hard versions, while fine-tuning lifts an open 8B model to 33.3% on easy cases."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Vision-language models recover exact containment trees from nested Jordan curves at only 71 percent accuracy on easy cases and 19 percent on hard cases."}],"snapshot_sha256":"8c959dfc07e0289e15e4f7997f90f45006c7d07fc7f9ea0745f0e98877071448"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"977f3def4313ba2ec4124af84af93ea5a60e0a73a6b56c7952340acbff025b50"},"paper":{"abstract_excerpt":"We introduce CurveBench, a benchmark for hierarchical topological reasoning from visual input. CurveBench consists of \\textbf{756 images} of pairwise non-intersecting Jordan curves across easy, polygonal, topographic-inspired, maze-like, and dense counting configurations. Each image is annotated with a rooted tree encoding the containment relations between planar regions. We formulate the task as structured prediction: given an image, a model must recover the full rooted containment tree induced by the curves. Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1","authors_text":"Amirreza Mohseni, Mona Mohammadi, Morteza Saghafian, Naser Talebizadeh Saradari","cross_cats":["cs.LG"],"headline":"Vision-language models recover exact containment trees from nested Jordan curves at only 71 percent accuracy on easy cases and 19 percent on hard cases.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T19:46:22Z","title":"CurveBench: A Benchmark for Exact Topological Reasoning over Nested Jordan Curves"},"references":{"count":44,"internal_anchors":10,"resolved_work":44,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Topographic Map Symbols , year =","work_id":"97537c87-e528-4e49-8aa6-c4a856a8e769","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Jordan Theorem , year =","work_id":"1b97f40c-b303-44f0-be1b-993511763d99","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Journal of Visual Languages and Computing , volume =","work_id":"9b902288-1e92-4908-a0e0-08556ec84269","year":2014},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Findings of the Association for Computational Linguistics: ACL 2022 , pages =","work_id":"80430d48-4070-43de-b886-0d69668c33c3","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Mathew, Minesh and Karatzas, Dimosthenis and Jawahar, C. V. , title =. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages =. 2021 , url =","work_id":"0289afd3-d25b-479c-a341-b704cb979319","year":2021}],"snapshot_sha256":"80adee1f36ded0afa842dac20456da112d1fe35319c75a9dde85151a0b83d11c"},"source":{"id":"2605.14068","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T05:30:49.710142Z","id":"f70b9cb0-297f-4075-af14-507829a0b5bc","model_set":{"reader":"grok-4.3"},"one_line_summary":"CurveBench benchmark reveals that even leading VLMs like Gemini 3.1 Pro reach only 71.1% accuracy recovering containment trees on easy nested-curve images and 19.1% on hard versions, while fine-tuning lifts an open 8B model to 33.3% on easy cases.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Vision-language models recover exact containment trees from nested Jordan curves at only 71 percent accuracy on easy cases and 19 percent on hard cases.","strongest_claim":"Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1 Pro, achieves only 71.1% tree-generation accuracy on CurveBench-Easy and 19.1% on CurveBench-Hard.","weakest_assumption":"That the generated images and their tree annotations faithfully isolate topological containment reasoning without providing unintended low-level visual shortcuts or dataset-specific biases that models could exploit."}},"verdict_id":"f70b9cb0-297f-4075-af14-507829a0b5bc"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:4982d7bfd7b7b7cd04d99eab1bf270d8c46560785dc14e7428f9bbb66b151b96","target":"record","created_at":"2026-05-17T23:39:12Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"4dd9faab3424e6d7b65351d4013f730184889dac501b2a3eec4572b170e7b9b8","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T19:46:22Z","title_canon_sha256":"0d9c0595ceb9bb930acd406796590d6f7502ec2fc2934690b5f53833a1b18430"},"schema_version":"1.0","source":{"id":"2605.14068","kind":"arxiv","version":1}},"canonical_sha256":"a68cf8b77c768d804729d7a6d573831b8f3f6eae26abd474e5d76efecae0380b","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a68cf8b77c768d804729d7a6d573831b8f3f6eae26abd474e5d76efecae0380b","first_computed_at":"2026-05-17T23:39:12.452745Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:12.452745Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"g/0p3Y2LKwLz1cv8CiEUrQrxe+CKf05Z6YUS1AoU7IavQ1rljVvfJSHBv6UTo3NQaH3dao6/oJ4umkT3bC0dBA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:12.453369Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14068","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:4982d7bfd7b7b7cd04d99eab1bf270d8c46560785dc14e7428f9bbb66b151b96","sha256:97a23fa8e969c70697c9f4c7dbc1b304723ae00c90f27b76b5e22d02fba8cf25"],"state_sha256":"7c36749178056d205ec521d6d538c7efb2d72c0a8edd4d265e717ded07ddf5e7"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"20YheIxUClXva+14wr59PcfVt8S8KaXok5Hak5GG9X85WEp/fpkJdvt2VoyHXgguUSthiLItlLa66Ogv2jTyCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-04T20:16:09.157929Z","bundle_sha256":"0f8879b871ec7c99a7e303a30e0a0fb485a3dbb42856ac1bb23c5bcec9e0ee76"}}