{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:EKDVNIQLS6KMI543JWW7XDRSYQ","short_pith_number":"pith:EKDVNIQL","canonical_record":{"source":{"id":"2604.19412","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-21T12:40:07Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"236c17ec5c947c00d7f2364361cb32c2e4c50e31bae5a7a7fe8c7133bb0d47ae","abstract_canon_sha256":"96e80e797b6e0f909fee306c9344418079a006c61ca361dd7ff0ac8b8e6df07d"},"schema_version":"1.0"},"canonical_sha256":"228756a20b9794c4779b4dadfb8e32c40ea8bba6a7add73b326a4ff8822e87e2","source":{"kind":"arxiv","id":"2604.19412","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.19412","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"arxiv_version","alias_value":"2604.19412v1","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.19412","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"pith_short_12","alias_value":"EKDVNIQLS6KM","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"pith_short_16","alias_value":"EKDVNIQLS6KMI543","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"pith_short_8","alias_value":"EKDVNIQL","created_at":"2026-07-31T01:11:10Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:EKDVNIQLS6KMI543JWW7XDRSYQ","target":"record","payload":{"canonical_record":{"source":{"id":"2604.19412","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-21T12:40:07Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"236c17ec5c947c00d7f2364361cb32c2e4c50e31bae5a7a7fe8c7133bb0d47ae","abstract_canon_sha256":"96e80e797b6e0f909fee306c9344418079a006c61ca361dd7ff0ac8b8e6df07d"},"schema_version":"1.0"},"canonical_sha256":"228756a20b9794c4779b4dadfb8e32c40ea8bba6a7add73b326a4ff8822e87e2","receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"228756a20b9794c4779b4dadfb8e32c40ea8bba6a7add73b326a4ff8822e87e2","last_reissued_at":"2026-07-31T01:11:10.253622Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-31T01:11:10.253622Z"},"source_kind":"arxiv","source_id":"2604.19412","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":"2026-07-31T01:11:10Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"8+J99pxK/YAFx3QvZgqSTRG2VdroC6CoxE1zVPOaUxmB5+aBtlSH7R/zE4pkxyzwbNzPmnD16QpzgH+TsR8nAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-09T21:44:53.668335Z"},"content_sha256":"b37a1675e5b64f83809c89fac5b96c6a21be44dab17758a8f385f9a620a297c7","schema_version":"1.0","event_id":"sha256:b37a1675e5b64f83809c89fac5b96c6a21be44dab17758a8f385f9a620a297c7"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:EKDVNIQLS6KMI543JWW7XDRSYQ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"VCE: A zero-cost hallucination mitigation method of LVLMs via visual contrastive editing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"VCE reduces object hallucinations in large vision-language models by decomposing activation patterns from contrastive image pairs via singular value decomposition and applying targeted edits to suppress hallucination subspaces.","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Guiyao Tie, Hao Sun, Hongfei Wang, Pan Zhou, Xiaoye Qu, Xuelong Li, Yanbin Huang, Yisen Li, Zhaofan Zou","submitted_at":"2026-04-21T12:40:07Z","abstract_excerpt":"Large vision-language models (LVLMs) frequently suffer from Object Hallucination (OH), wherein they generate descriptions containing objects that are not actually present in the input image. This phenomenon is particularly problematic in real-world applications such as medical imaging and autonomous driving, where accuracy is critical. Recent studies suggest that the hallucination problem may stem from language priors: biases learned during pretraining that cause LVLMs to generate words based on their statistical co-occurrence. To mitigate this problem, we propose Visual Contrastive Editing (V"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experimental results demonstrate that VCE effectively reduces object hallucination across multiple benchmarks while maintaining the model's original computational efficiency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That object hallucination can be isolated as distinct subspaces in model activations via SVD on responses to contrastive visual perturbations, and that targeted edits can attenuate these without unintended side effects on other capabilities.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VCE mitigates object hallucination in LVLMs by decomposing activation patterns from contrastive visual inputs via SVD to suppress hallucination subspaces through targeted parameter edits.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"VCE reduces object hallucinations in large vision-language models by decomposing activation patterns from contrastive image pairs via singular value decomposition and applying targeted edits to suppress hallucination subspaces.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d3e41715b5ccc08ee10623b5e2f66090d92c5ceeefc938d001c4c461d340ed37"},"source":{"id":"2604.19412","kind":"arxiv","version":1},"verdict":{"id":"0437d622-63f6-48d4-bc2e-5ab02200d580","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T03:13:25.677313Z","strongest_claim":"Experimental results demonstrate that VCE effectively reduces object hallucination across multiple benchmarks while maintaining the model's original computational efficiency.","one_line_summary":"VCE mitigates object hallucination in LVLMs by decomposing activation patterns from contrastive visual inputs via SVD to suppress hallucination subspaces through targeted parameter edits.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That object hallucination can be isolated as distinct subspaces in model activations via SVD on responses to contrastive visual perturbations, and that targeted edits can attenuate these without unintended side effects on other capabilities.","pith_extraction_headline":"VCE reduces object hallucinations in large vision-language models by decomposing activation patterns from contrastive image pairs via singular value decomposition and applying targeted edits to suppress hallucination subspaces."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.19412/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-20T02:59:56.413513Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"58f7f784542cba7c07bfbdc8c6be97220c7dcc9bb074991bd08e37b5afd8b6af"},"references":{"count":27,"sample":[{"doi":"","year":2026,"title":"VCE: A zero-cost hallucination mitigation method of LVLMs via visual contrastive editing","work_id":"6ed56563-2068-49c5-b3ef-e40e00ca35ba","ref_index":1,"cited_arxiv_id":"2604.19412","is_internal_anchor":true},{"doi":"","year":null,"title":"[13]. In this work, we propose Visual Contrastive Editing (VCE), a framework that diagnoses hallucination by syn- thesizing contrastive image pairs—perturbations that alter localized visual features w","work_id":"201e1e2b-626f-417d-9128-ac9b6a4338ef","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Problem Definition Mathematically, an LVLM can be defined as a conditional probability distribution: p(y|x,v;θ) = TY t=1 pθ yt |y <t,x,v ,(1) wherey={y 1, y2","work_id":"be8d3038-08f0-48dd-9ac3-6a7b540b9398","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Experiment Setups Datasets.We evaluate the VCE method on popular hallucina- tion benchmarks: POPE [18] and CHAIR [19]","work_id":"b5de6fac-3db2-492f-9b98-240be4975919","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"CONCLUSION In this work, we presented Visual Contrastive Editing (VCE), a zero-cost, post-hoc framework for mitigating object hal- lucination in large vision–language models. By leveraging contrastive","work_id":"3faf9150-2ce5-4499-aa89-2685f40c47be","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":27,"snapshot_sha256":"bcd04bcaef6ab79b9593ddf26271bfb3c0ed4c9fc738a3c39e535a5d077b5649","internal_anchors":3},"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"},"verdict_id":"0437d622-63f6-48d4-bc2e-5ab02200d580"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-31T01:11:10Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"yhzJAD6E6wVE0huQwX3oZCo3LdP8bxrCjOzRZ3GUiC+3RupJCfKdUOc+OhwOb2mtcxf2epQn04N7MPayQMi0Ag==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-09T21:44:53.669830Z"},"content_sha256":"449197ce5234da4a4c9e042203dc24aced57430a9b35261830d6c6104736daf8","schema_version":"1.0","event_id":"sha256:449197ce5234da4a4c9e042203dc24aced57430a9b35261830d6c6104736daf8"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/EKDVNIQLS6KMI543JWW7XDRSYQ/bundle.json","state_url":"https://pith.science/pith/EKDVNIQLS6KMI543JWW7XDRSYQ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/EKDVNIQLS6KMI543JWW7XDRSYQ/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":"2026-08-09T21:44:53Z","links":{"resolver":"https://pith.science/pith/EKDVNIQLS6KMI543JWW7XDRSYQ","bundle":"https://pith.science/pith/EKDVNIQLS6KMI543JWW7XDRSYQ/bundle.json","state":"https://pith.science/pith/EKDVNIQLS6KMI543JWW7XDRSYQ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/EKDVNIQLS6KMI543JWW7XDRSYQ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:EKDVNIQLS6KMI543JWW7XDRSYQ","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":"96e80e797b6e0f909fee306c9344418079a006c61ca361dd7ff0ac8b8e6df07d","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-21T12:40:07Z","title_canon_sha256":"236c17ec5c947c00d7f2364361cb32c2e4c50e31bae5a7a7fe8c7133bb0d47ae"},"schema_version":"1.0","source":{"id":"2604.19412","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.19412","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"arxiv_version","alias_value":"2604.19412v1","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.19412","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"pith_short_12","alias_value":"EKDVNIQLS6KM","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"pith_short_16","alias_value":"EKDVNIQLS6KMI543","created_at":"2026-07-31T01:11:10Z"},{"alias_kind":"pith_short_8","alias_value":"EKDVNIQL","created_at":"2026-07-31T01:11:10Z"}],"graph_snapshots":[{"event_id":"sha256:449197ce5234da4a4c9e042203dc24aced57430a9b35261830d6c6104736daf8","target":"graph","created_at":"2026-07-31T01:11:10Z","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":"Experimental results demonstrate that VCE effectively reduces object hallucination across multiple benchmarks while maintaining the model's original computational efficiency."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That object hallucination can be isolated as distinct subspaces in model activations via SVD on responses to contrastive visual perturbations, and that targeted edits can attenuate these without unintended side effects on other capabilities."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"VCE mitigates object hallucination in LVLMs by decomposing activation patterns from contrastive visual inputs via SVD to suppress hallucination subspaces through targeted parameter edits."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"VCE reduces object hallucinations in large vision-language models by decomposing activation patterns from contrastive image pairs via singular value decomposition and applying targeted edits to suppress hallucination subspaces."}],"snapshot_sha256":"d3e41715b5ccc08ee10623b5e2f66090d92c5ceeefc938d001c4c461d340ed37"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-20T02:59:56.413513Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2604.19412/integrity.json","findings":[],"snapshot_sha256":"58f7f784542cba7c07bfbdc8c6be97220c7dcc9bb074991bd08e37b5afd8b6af","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large vision-language models (LVLMs) frequently suffer from Object Hallucination (OH), wherein they generate descriptions containing objects that are not actually present in the input image. This phenomenon is particularly problematic in real-world applications such as medical imaging and autonomous driving, where accuracy is critical. Recent studies suggest that the hallucination problem may stem from language priors: biases learned during pretraining that cause LVLMs to generate words based on their statistical co-occurrence. To mitigate this problem, we propose Visual Contrastive Editing (V","authors_text":"Guiyao Tie, Hao Sun, Hongfei Wang, Pan Zhou, Xiaoye Qu, Xuelong Li, Yanbin Huang, Yisen Li, Zhaofan Zou","cross_cats":["cs.CL"],"headline":"VCE reduces object hallucinations in large vision-language models by decomposing activation patterns from contrastive image pairs via singular value decomposition and applying targeted edits to suppress hallucination subspaces.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-21T12:40:07Z","title":"VCE: A zero-cost hallucination mitigation method of LVLMs via visual contrastive editing"},"references":{"count":27,"internal_anchors":3,"resolved_work":27,"sample":[{"cited_arxiv_id":"2604.19412","doi":"","is_internal_anchor":true,"ref_index":1,"title":"VCE: A zero-cost hallucination mitigation method of LVLMs via visual contrastive editing","work_id":"6ed56563-2068-49c5-b3ef-e40e00ca35ba","year":2026},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"[13]. In this work, we propose Visual Contrastive Editing (VCE), a framework that diagnoses hallucination by syn- thesizing contrastive image pairs—perturbations that alter localized visual features w","work_id":"201e1e2b-626f-417d-9128-ac9b6a4338ef","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Problem Definition Mathematically, an LVLM can be defined as a conditional probability distribution: p(y|x,v;θ) = TY t=1 pθ yt |y <t,x,v ,(1) wherey={y 1, y2","work_id":"be8d3038-08f0-48dd-9ac3-6a7b540b9398","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Experiment Setups Datasets.We evaluate the VCE method on popular hallucina- tion benchmarks: POPE [18] and CHAIR [19]","work_id":"b5de6fac-3db2-492f-9b98-240be4975919","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"CONCLUSION In this work, we presented Visual Contrastive Editing (VCE), a zero-cost, post-hoc framework for mitigating object hal- lucination in large vision–language models. By leveraging contrastive","work_id":"3faf9150-2ce5-4499-aa89-2685f40c47be","year":null}],"snapshot_sha256":"bcd04bcaef6ab79b9593ddf26271bfb3c0ed4c9fc738a3c39e535a5d077b5649"},"source":{"id":"2604.19412","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-10T03:13:25.677313Z","id":"0437d622-63f6-48d4-bc2e-5ab02200d580","model_set":{"reader":"grok-4.3"},"one_line_summary":"VCE mitigates object hallucination in LVLMs by decomposing activation patterns from contrastive visual inputs via SVD to suppress hallucination subspaces through targeted parameter edits.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"VCE reduces object hallucinations in large vision-language models by decomposing activation patterns from contrastive image pairs via singular value decomposition and applying targeted edits to suppress hallucination subspaces.","strongest_claim":"Experimental results demonstrate that VCE effectively reduces object hallucination across multiple benchmarks while maintaining the model's original computational efficiency.","weakest_assumption":"That object hallucination can be isolated as distinct subspaces in model activations via SVD on responses to contrastive visual perturbations, and that targeted edits can attenuate these without unintended side effects on other capabilities."}},"verdict_id":"0437d622-63f6-48d4-bc2e-5ab02200d580"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b37a1675e5b64f83809c89fac5b96c6a21be44dab17758a8f385f9a620a297c7","target":"record","created_at":"2026-07-31T01:11:10Z","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":"96e80e797b6e0f909fee306c9344418079a006c61ca361dd7ff0ac8b8e6df07d","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-21T12:40:07Z","title_canon_sha256":"236c17ec5c947c00d7f2364361cb32c2e4c50e31bae5a7a7fe8c7133bb0d47ae"},"schema_version":"1.0","source":{"id":"2604.19412","kind":"arxiv","version":1}},"canonical_sha256":"228756a20b9794c4779b4dadfb8e32c40ea8bba6a7add73b326a4ff8822e87e2","receipt":{"builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"228756a20b9794c4779b4dadfb8e32c40ea8bba6a7add73b326a4ff8822e87e2","first_computed_at":"2026-07-31T01:11:10.253622Z","kind":"pith_receipt","last_reissued_at":"2026-07-31T01:11:10.253622Z","receipt_version":"0.3","signature_status":"unsigned_v0"},"source_id":"2604.19412","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b37a1675e5b64f83809c89fac5b96c6a21be44dab17758a8f385f9a620a297c7","sha256:449197ce5234da4a4c9e042203dc24aced57430a9b35261830d6c6104736daf8"],"state_sha256":"560585496702927090035e96613fc3f3a22f23c72950710ec5dc1dfea956c66f"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/2Za2n31cmDluEOfIo9zNGwvt4G4KEQs5LdZ31biP2RBJKmDUmOejbQyUI/RPMBChR7V/6PwXAM4HKZAWRpSAQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-09T21:44:53.676292Z","bundle_sha256":"3c9439bd3a1142c3c67fed8ddf44f601a4a02385ced0c0b4f8d3577e14cb552c"}}