{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:L525QO6HXRPQ7YJSME3KYZZJXU","short_pith_number":"pith:L525QO6H","canonical_record":{"source":{"id":"2605.13030","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T05:35:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4638e80f5d3dc6fa2643b2dd8143335c4340d61f0d5361932e800fd9dbe05a27","abstract_canon_sha256":"cc0f645c4a8979914ffc7b2766af804a1d9190e235faba4ae737f3adf2398f4f"},"schema_version":"1.0"},"canonical_sha256":"5f75d83bc7bc5f0fe1326136ac6729bd2c4f7690fd3a9de9f69837faaa937f52","source":{"kind":"arxiv","id":"2605.13030","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13030","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13030v1","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13030","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"pith_short_12","alias_value":"L525QO6HXRPQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"L525QO6HXRPQ7YJS","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"L525QO6H","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:L525QO6HXRPQ7YJSME3KYZZJXU","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13030","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T05:35:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4638e80f5d3dc6fa2643b2dd8143335c4340d61f0d5361932e800fd9dbe05a27","abstract_canon_sha256":"cc0f645c4a8979914ffc7b2766af804a1d9190e235faba4ae737f3adf2398f4f"},"schema_version":"1.0"},"canonical_sha256":"5f75d83bc7bc5f0fe1326136ac6729bd2c4f7690fd3a9de9f69837faaa937f52","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:08:59.774015Z","signature_b64":"lZ3VTndVRuIAtdgasjX5umdUtsNYUs27Hk9yLAfDK0jmcZucP592N1qFjtdd3EM+ZnXSoqkpM/28kU6rD5gyCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5f75d83bc7bc5f0fe1326136ac6729bd2c4f7690fd3a9de9f69837faaa937f52","last_reissued_at":"2026-05-18T03:08:59.773322Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:08:59.773322Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13030","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":"2026-05-18T03:08:59Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Dkwha1G6wfrO9nDhgfFyRbw+IGZrZs06G8+tH8aWQHejFhg2nnVxSJnp0RBWstkq8O/slUfe6U/r2/CxeZS5Aw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T17:05:45.215546Z"},"content_sha256":"92b11294ffcdfcd71e4d55fd0e3edea539c14934190b46e3dc075879a1d098fc","schema_version":"1.0","event_id":"sha256:92b11294ffcdfcd71e4d55fd0e3edea539c14934190b46e3dc075879a1d098fc"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:L525QO6HXRPQ7YJSME3KYZZJXU","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"FeatCal: Feature Calibration for Post-Merging Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Feature drift in merged models decomposes into upstream propagation and local mismatch that can be corrected layer by layer using closed-form calibration on a small set.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Hongxia Yang, Jianmin Wu, Pengkai Wang, Shuo Cai, Sirui Huang, Su Lu, Wenjun Wang, Yanggan Gu, Yuanyi Wang, Zihao Wang","submitted_at":"2026-05-13T05:35:03Z","abstract_excerpt":"Model merging combines task experts into one model and avoids joint training, retraining, or deploying many expert models, but the merged model often still underperforms task experts. We study this performance gap through feature drift, the difference between features produced by the merged model and by the expert on the same input. Our theory decomposes this drift into upstream propagation and local mismatch, tracks how it propagates and combines through later layers in forward order, and links final feature drift to output drift. This view motivates FeatCal, which uses a small calibration se"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On CLIP-ViT-B/32 Task Arithmetic, FeatCal reaches 85.5% versus 77.0%/78.8% for Surgery/ProbSurgery; on FLAN-T5-base GLUE it reaches 85.2% versus 83.7%/82.2%, with 8 examples per task yielding 82.9% and 256 examples taking 53 seconds.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the proposed decomposition of feature drift into upstream propagation and local mismatch accurately captures the dominant cause of output degradation, and that layer-wise closed-form calibration on a small set will generalize without harming unrelated capabilities.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FeatCal reduces feature drift in merged models via layer-wise closed-form calibration on a small dataset, outperforming prior post-merging methods on CLIP and GLUE benchmarks with high sample efficiency.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Feature drift in merged models decomposes into upstream propagation and local mismatch that can be corrected layer by layer using closed-form calibration on a small set.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"748ce4b4556dbb10d0dcedc36e6b0561b84038e0381e701ce90c2ab15b030929"},"source":{"id":"2605.13030","kind":"arxiv","version":1},"verdict":{"id":"5523a285-bb67-4b51-8fe2-468adbc5bce3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:23:54.310104Z","strongest_claim":"On CLIP-ViT-B/32 Task Arithmetic, FeatCal reaches 85.5% versus 77.0%/78.8% for Surgery/ProbSurgery; on FLAN-T5-base GLUE it reaches 85.2% versus 83.7%/82.2%, with 8 examples per task yielding 82.9% and 256 examples taking 53 seconds.","one_line_summary":"FeatCal reduces feature drift in merged models via layer-wise closed-form calibration on a small dataset, outperforming prior post-merging methods on CLIP and GLUE benchmarks with high sample efficiency.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the proposed decomposition of feature drift into upstream propagation and local mismatch accurately captures the dominant cause of output degradation, and that layer-wise closed-form calibration on a small set will generalize without harming unrelated capabilities.","pith_extraction_headline":"Feature drift in merged models decomposes into upstream propagation and local mismatch that can be corrected layer by layer using closed-form calibration on a small set."},"references":{"count":65,"sample":[{"doi":"","year":2022,"title":"Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, and Ludwig Schmidt","work_id":"ec315768-c936-4eb1-acab-3b435d6f5e0f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Matena and Colin A","work_id":"8c19aec7-6be8-42cd-9057-331be9af4408","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Editing models with task arithmetic","work_id":"a4ad53a2-6272-4f48-bf46-c647b9e5fd8b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"TIES-Merging: Resolving interference when merging models","work_id":"eec3bf3a-20db-439a-96ab-b2194552e4f7","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"AdaMerging: Adaptive model merging for multi-task learning","work_id":"621ee4a9-c72a-4503-bd70-5133ddd80462","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":65,"snapshot_sha256":"9bce2ecfce5cb90d89c745bc3af7f79888f4c3978dbb754b6431137599eb1a5c","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"17656895f12acd6fd26e5b2bc1530ce1c5f009a323c0d64031eec8196fdcc356"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"5523a285-bb67-4b51-8fe2-468adbc5bce3"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:08:59Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"7bCWv3i3lBGvfGuw2vnziruElkrP89NVS5k05lCjqPWH3mEao3ZKkcX7qIxTYW0JouDhNXxx1/1NY7dBiwIZCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-02T17:05:45.216303Z"},"content_sha256":"f639b46d658bbb4dccb6be3894ebe7c3b68e349206d590fa407293ae3936e2fd","schema_version":"1.0","event_id":"sha256:f639b46d658bbb4dccb6be3894ebe7c3b68e349206d590fa407293ae3936e2fd"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/L525QO6HXRPQ7YJSME3KYZZJXU/bundle.json","state_url":"https://pith.science/pith/L525QO6HXRPQ7YJSME3KYZZJXU/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/L525QO6HXRPQ7YJSME3KYZZJXU/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":"2026-06-02T17:05:45Z","links":{"resolver":"https://pith.science/pith/L525QO6HXRPQ7YJSME3KYZZJXU","bundle":"https://pith.science/pith/L525QO6HXRPQ7YJSME3KYZZJXU/bundle.json","state":"https://pith.science/pith/L525QO6HXRPQ7YJSME3KYZZJXU/state.json","well_known_bundle":"https://pith.science/.well-known/pith/L525QO6HXRPQ7YJSME3KYZZJXU/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:L525QO6HXRPQ7YJSME3KYZZJXU","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":"cc0f645c4a8979914ffc7b2766af804a1d9190e235faba4ae737f3adf2398f4f","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T05:35:03Z","title_canon_sha256":"4638e80f5d3dc6fa2643b2dd8143335c4340d61f0d5361932e800fd9dbe05a27"},"schema_version":"1.0","source":{"id":"2605.13030","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13030","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13030v1","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13030","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"pith_short_12","alias_value":"L525QO6HXRPQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"L525QO6HXRPQ7YJS","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"L525QO6H","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:f639b46d658bbb4dccb6be3894ebe7c3b68e349206d590fa407293ae3936e2fd","target":"graph","created_at":"2026-05-18T03:08:59Z","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":"On CLIP-ViT-B/32 Task Arithmetic, FeatCal reaches 85.5% versus 77.0%/78.8% for Surgery/ProbSurgery; on FLAN-T5-base GLUE it reaches 85.2% versus 83.7%/82.2%, with 8 examples per task yielding 82.9% and 256 examples taking 53 seconds."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the proposed decomposition of feature drift into upstream propagation and local mismatch accurately captures the dominant cause of output degradation, and that layer-wise closed-form calibration on a small set will generalize without harming unrelated capabilities."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"FeatCal reduces feature drift in merged models via layer-wise closed-form calibration on a small dataset, outperforming prior post-merging methods on CLIP and GLUE benchmarks with high sample efficiency."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Feature drift in merged models decomposes into upstream propagation and local mismatch that can be corrected layer by layer using closed-form calibration on a small set."}],"snapshot_sha256":"748ce4b4556dbb10d0dcedc36e6b0561b84038e0381e701ce90c2ab15b030929"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"17656895f12acd6fd26e5b2bc1530ce1c5f009a323c0d64031eec8196fdcc356"},"paper":{"abstract_excerpt":"Model merging combines task experts into one model and avoids joint training, retraining, or deploying many expert models, but the merged model often still underperforms task experts. We study this performance gap through feature drift, the difference between features produced by the merged model and by the expert on the same input. Our theory decomposes this drift into upstream propagation and local mismatch, tracks how it propagates and combines through later layers in forward order, and links final feature drift to output drift. This view motivates FeatCal, which uses a small calibration se","authors_text":"Hongxia Yang, Jianmin Wu, Pengkai Wang, Shuo Cai, Sirui Huang, Su Lu, Wenjun Wang, Yanggan Gu, Yuanyi Wang, Zihao Wang","cross_cats":["cs.AI"],"headline":"Feature drift in merged models decomposes into upstream propagation and local mismatch that can be corrected layer by layer using closed-form calibration on a small set.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T05:35:03Z","title":"FeatCal: Feature Calibration for Post-Merging Models"},"references":{"count":65,"internal_anchors":3,"resolved_work":65,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, and Ludwig Schmidt","work_id":"ec315768-c936-4eb1-acab-3b435d6f5e0f","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Matena and Colin A","work_id":"8c19aec7-6be8-42cd-9057-331be9af4408","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Editing models with task arithmetic","work_id":"a4ad53a2-6272-4f48-bf46-c647b9e5fd8b","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"TIES-Merging: Resolving interference when merging models","work_id":"eec3bf3a-20db-439a-96ab-b2194552e4f7","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"AdaMerging: Adaptive model merging for multi-task learning","work_id":"621ee4a9-c72a-4503-bd70-5133ddd80462","year":2024}],"snapshot_sha256":"9bce2ecfce5cb90d89c745bc3af7f79888f4c3978dbb754b6431137599eb1a5c"},"source":{"id":"2605.13030","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:23:54.310104Z","id":"5523a285-bb67-4b51-8fe2-468adbc5bce3","model_set":{"reader":"grok-4.3"},"one_line_summary":"FeatCal reduces feature drift in merged models via layer-wise closed-form calibration on a small dataset, outperforming prior post-merging methods on CLIP and GLUE benchmarks with high sample efficiency.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Feature drift in merged models decomposes into upstream propagation and local mismatch that can be corrected layer by layer using closed-form calibration on a small set.","strongest_claim":"On CLIP-ViT-B/32 Task Arithmetic, FeatCal reaches 85.5% versus 77.0%/78.8% for Surgery/ProbSurgery; on FLAN-T5-base GLUE it reaches 85.2% versus 83.7%/82.2%, with 8 examples per task yielding 82.9% and 256 examples taking 53 seconds.","weakest_assumption":"That the proposed decomposition of feature drift into upstream propagation and local mismatch accurately captures the dominant cause of output degradation, and that layer-wise closed-form calibration on a small set will generalize without harming unrelated capabilities."}},"verdict_id":"5523a285-bb67-4b51-8fe2-468adbc5bce3"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:92b11294ffcdfcd71e4d55fd0e3edea539c14934190b46e3dc075879a1d098fc","target":"record","created_at":"2026-05-18T03:08:59Z","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":"cc0f645c4a8979914ffc7b2766af804a1d9190e235faba4ae737f3adf2398f4f","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T05:35:03Z","title_canon_sha256":"4638e80f5d3dc6fa2643b2dd8143335c4340d61f0d5361932e800fd9dbe05a27"},"schema_version":"1.0","source":{"id":"2605.13030","kind":"arxiv","version":1}},"canonical_sha256":"5f75d83bc7bc5f0fe1326136ac6729bd2c4f7690fd3a9de9f69837faaa937f52","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"5f75d83bc7bc5f0fe1326136ac6729bd2c4f7690fd3a9de9f69837faaa937f52","first_computed_at":"2026-05-18T03:08:59.773322Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:08:59.773322Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"lZ3VTndVRuIAtdgasjX5umdUtsNYUs27Hk9yLAfDK0jmcZucP592N1qFjtdd3EM+ZnXSoqkpM/28kU6rD5gyCA==","signature_status":"signed_v1","signed_at":"2026-05-18T03:08:59.774015Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13030","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:92b11294ffcdfcd71e4d55fd0e3edea539c14934190b46e3dc075879a1d098fc","sha256:f639b46d658bbb4dccb6be3894ebe7c3b68e349206d590fa407293ae3936e2fd"],"state_sha256":"58e2a1f1047972ca1c822499f5200911c945e4aec2d9b0f26ccddf6d6253c31b"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/cj4a1X37BIMI25/bKpgaFfRoRwW76RT6uYUg5TylN4t7spGQSyYUBiqzozjbBGJak1fBPfebnEwfH4HF8jkCw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-02T17:05:45.218630Z","bundle_sha256":"23d15f4531d328f82bbe3e3d54fdbec4d182eba8c7d4ac4a564e7b8b282ccd45"}}