{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:H3N2OKQQFCA5T7TDBBJ5Z7CXXA","short_pith_number":"pith:H3N2OKQQ","schema_version":"1.0","canonical_sha256":"3edba72a102881d9fe630853dcfc57b820da5dd704778e9073a795540814cefa","source":{"kind":"arxiv","id":"2606.27771","version":1},"attestation_state":"computed","paper":{"title":"NormGuard: Reward-Preserving Norm Constraints in Flow-Matching Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Changqian Yu, Cheng Da, Huan Yang, Kun Gai, Lianyu Pang, Tianlin Pan, Wenhan Luo","submitted_at":"2026-06-26T06:56:54Z","abstract_excerpt":"Reinforcement learning (RL) post-training improves the reward alignment of flow-based generators, but often degrades perceptual quality in ways that are not captured by the reward proxy. We identify a simple structural signature of this drift: across three post-training methods (NFT, AWM, DPO), RL fine-tuning inflates the per-step velocity norm $\\|v_\\theta\\|$ by $5\\%$ to $15\\%$ relative to the reference. A form of norm inflation has been studied in classifier-free guidance (CFG), where rescaling the velocity back to a reference norm at inference time can mitigate the resulting artifacts. Howev"},"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":"2606.27771","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-26T06:56:54Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"a6fe8cb2306fc4b90b607c4fb30ec0224206f4f0c762cf76694ede352fd222e7","abstract_canon_sha256":"67aea9673e958eeeae31d34615504be0f393e3daa734e2ebd796d8c89864553f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-29T01:14:47.936394Z","signature_b64":"eiHzGSVBN3IGYEAd3NMzjQWwZdKuL9BcnDdRzRXpzrTlwEKesdAhMJYVrDD8oXb0aZ/Fhkx6NSoXVBCOv6h/Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3edba72a102881d9fe630853dcfc57b820da5dd704778e9073a795540814cefa","last_reissued_at":"2026-06-29T01:14:47.935980Z","signature_status":"signed_v1","first_computed_at":"2026-06-29T01:14:47.935980Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"NormGuard: Reward-Preserving Norm Constraints in Flow-Matching Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Changqian Yu, Cheng Da, Huan Yang, Kun Gai, Lianyu Pang, Tianlin Pan, Wenhan Luo","submitted_at":"2026-06-26T06:56:54Z","abstract_excerpt":"Reinforcement learning (RL) post-training improves the reward alignment of flow-based generators, but often degrades perceptual quality in ways that are not captured by the reward proxy. We identify a simple structural signature of this drift: across three post-training methods (NFT, AWM, DPO), RL fine-tuning inflates the per-step velocity norm $\\|v_\\theta\\|$ by $5\\%$ to $15\\%$ relative to the reference. A form of norm inflation has been studied in classifier-free guidance (CFG), where rescaling the velocity back to a reference norm at inference time can mitigate the resulting artifacts. Howev"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.27771","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/2606.27771/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":"2606.27771","created_at":"2026-06-29T01:14:47.936041+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.27771v1","created_at":"2026-06-29T01:14:47.936041+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.27771","created_at":"2026-06-29T01:14:47.936041+00:00"},{"alias_kind":"pith_short_12","alias_value":"H3N2OKQQFCA5","created_at":"2026-06-29T01:14:47.936041+00:00"},{"alias_kind":"pith_short_16","alias_value":"H3N2OKQQFCA5T7TD","created_at":"2026-06-29T01:14:47.936041+00:00"},{"alias_kind":"pith_short_8","alias_value":"H3N2OKQQ","created_at":"2026-06-29T01:14:47.936041+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/H3N2OKQQFCA5T7TDBBJ5Z7CXXA","json":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA.json","graph_json":"https://pith.science/api/pith-number/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/graph.json","events_json":"https://pith.science/api/pith-number/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/events.json","paper":"https://pith.science/paper/H3N2OKQQ"},"agent_actions":{"view_html":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA","download_json":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA.json","view_paper":"https://pith.science/paper/H3N2OKQQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.27771&json=true","fetch_graph":"https://pith.science/api/pith-number/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/graph.json","fetch_events":"https://pith.science/api/pith-number/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/action/storage_attestation","attest_author":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/action/author_attestation","sign_citation":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/action/citation_signature","submit_replication":"https://pith.science/pith/H3N2OKQQFCA5T7TDBBJ5Z7CXXA/action/replication_record"}},"created_at":"2026-06-29T01:14:47.936041+00:00","updated_at":"2026-06-29T01:14:47.936041+00:00"}