{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:B7A6DTJJBHY32HYGZNGQ6CBKMD","short_pith_number":"pith:B7A6DTJJ","schema_version":"1.0","canonical_sha256":"0fc1e1cd2909f1bd1f06cb4d0f082a60ea35f9a30a7de7b6ff5bd56eb20a9393","source":{"kind":"arxiv","id":"2605.22903","version":1},"attestation_state":"computed","paper":{"title":"Seeing without Looking: Do Vision-Language Benchmarks Really Test Vision?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Jiawei Zhou, Luzhe Sun, Matthew R. Walter, Zixuan Lan","submitted_at":"2026-05-21T17:35:04Z","abstract_excerpt":"Benchmark accuracy is often implicitly assumed to reflect grounded visual understanding in vision-language models (VLMs), yet it remains unclear to what extent such scores truly reflect reliance on visual evidence. Motivated by a surprising observation that removing a substantial fraction of image tokens only degrades model performance very slightly on a widely used hallucination benchmark, we systematically investigate this mismatch in a set of open-source VLMs. Our analysis spans multiple levels of granularity, spanning global visual degradation, localized occlusion, question reformulation, "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.22903","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-21T17:35:04Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"43e2b10bc66e7716d03acffde29e0916315107cb7e0aac36ce9e7b21101219be","abstract_canon_sha256":"cdbaa4efa1cd368137545686e055d129c00bccee915951a58a83f17abd9c71db"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:29.319773Z","signature_b64":"Gt/HGOf2r0Vv4s84arMCGb/kDzHMN/kFn6iUysz5+QD2Z00bSp+ATGF7FQPBfc6rnIQjlyTdN6+beIyJLNUrDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0fc1e1cd2909f1bd1f06cb4d0f082a60ea35f9a30a7de7b6ff5bd56eb20a9393","last_reissued_at":"2026-05-25T02:01:29.319099Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:29.319099Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Seeing without Looking: Do Vision-Language Benchmarks Really Test Vision?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Jiawei Zhou, Luzhe Sun, Matthew R. Walter, Zixuan Lan","submitted_at":"2026-05-21T17:35:04Z","abstract_excerpt":"Benchmark accuracy is often implicitly assumed to reflect grounded visual understanding in vision-language models (VLMs), yet it remains unclear to what extent such scores truly reflect reliance on visual evidence. Motivated by a surprising observation that removing a substantial fraction of image tokens only degrades model performance very slightly on a widely used hallucination benchmark, we systematically investigate this mismatch in a set of open-source VLMs. Our analysis spans multiple levels of granularity, spanning global visual degradation, localized occlusion, question reformulation, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22903","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.22903/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.22903","created_at":"2026-05-25T02:01:29.319192+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22903v1","created_at":"2026-05-25T02:01:29.319192+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22903","created_at":"2026-05-25T02:01:29.319192+00:00"},{"alias_kind":"pith_short_12","alias_value":"B7A6DTJJBHY3","created_at":"2026-05-25T02:01:29.319192+00:00"},{"alias_kind":"pith_short_16","alias_value":"B7A6DTJJBHY32HYG","created_at":"2026-05-25T02:01:29.319192+00:00"},{"alias_kind":"pith_short_8","alias_value":"B7A6DTJJ","created_at":"2026-05-25T02:01:29.319192+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD","json":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD.json","graph_json":"https://pith.science/api/pith-number/B7A6DTJJBHY32HYGZNGQ6CBKMD/graph.json","events_json":"https://pith.science/api/pith-number/B7A6DTJJBHY32HYGZNGQ6CBKMD/events.json","paper":"https://pith.science/paper/B7A6DTJJ"},"agent_actions":{"view_html":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD","download_json":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD.json","view_paper":"https://pith.science/paper/B7A6DTJJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22903&json=true","fetch_graph":"https://pith.science/api/pith-number/B7A6DTJJBHY32HYGZNGQ6CBKMD/graph.json","fetch_events":"https://pith.science/api/pith-number/B7A6DTJJBHY32HYGZNGQ6CBKMD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD/action/storage_attestation","attest_author":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD/action/author_attestation","sign_citation":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD/action/citation_signature","submit_replication":"https://pith.science/pith/B7A6DTJJBHY32HYGZNGQ6CBKMD/action/replication_record"}},"created_at":"2026-05-25T02:01:29.319192+00:00","updated_at":"2026-05-25T02:01:29.319192+00:00"}