{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:D75O2IKGNKCKW5XHQADKDWGECG","short_pith_number":"pith:D75O2IKG","schema_version":"1.0","canonical_sha256":"1ffaed21466a84ab76e78006a1d8c411a3bebae8bc55f71293dd1b390651cdf7","source":{"kind":"arxiv","id":"2511.19356","version":3},"attestation_state":"computed","paper":{"title":"Rethinking Reward Signals in Video GRPO: When Scores Become Targets","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chi Zhang, Haibin Huang, Rui Li, Xuelong Li, Yuanzhi Liang, Ziqi Ni","submitted_at":"2025-11-24T17:56:03Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) enables stable and preference-oriented updates via group-wise comparisons for post-training video generation. However, GRPO directly optimizes reward-induced advantages. Under sustained optimization, the reward score can lose fidelity as a proxy for true video quality, consistent with the phenomenon described by Goodhart's Law. This leads to two recurring issues: (i) shortcut-driven optimization under composite objectives and (ii) reward saturation within prompt groups. To address these issues, we introduce TaRoS, a Target-Robust Reward Signaling frame"},"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":"2511.19356","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-11-24T17:56:03Z","cross_cats_sorted":[],"title_canon_sha256":"f04499cc2fe213d17289ceb65724354c5be0721b4721111a834e069342883c7e","abstract_canon_sha256":"5b07e18d5a28c1f7f614e522654c4509ac3e4bb6d3b96e39108460bcec8c418e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-17T01:21:44.511860Z","signature_b64":"EnXPCPxMqcd3aDTK+baH8Xa2X6nOctIEfxOQi0hQeaWUp4PArQn7V0cmQ+f+VN1WEUAEkwaIvKW20/xAPouWAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1ffaed21466a84ab76e78006a1d8c411a3bebae8bc55f71293dd1b390651cdf7","last_reissued_at":"2026-07-17T01:21:44.510937Z","signature_status":"signed_v1","first_computed_at":"2026-07-17T01:21:44.510937Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking Reward Signals in Video GRPO: When Scores Become Targets","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Chi Zhang, Haibin Huang, Rui Li, Xuelong Li, Yuanzhi Liang, Ziqi Ni","submitted_at":"2025-11-24T17:56:03Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) enables stable and preference-oriented updates via group-wise comparisons for post-training video generation. However, GRPO directly optimizes reward-induced advantages. Under sustained optimization, the reward score can lose fidelity as a proxy for true video quality, consistent with the phenomenon described by Goodhart's Law. This leads to two recurring issues: (i) shortcut-driven optimization under composite objectives and (ii) reward saturation within prompt groups. To address these issues, we introduce TaRoS, a Target-Robust Reward Signaling frame"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.19356","kind":"arxiv","version":3},"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/2511.19356/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":"2511.19356","created_at":"2026-07-17T01:21:44.511378+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.19356v3","created_at":"2026-07-17T01:21:44.511378+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.19356","created_at":"2026-07-17T01:21:44.511378+00:00"},{"alias_kind":"pith_short_12","alias_value":"D75O2IKGNKCK","created_at":"2026-07-17T01:21:44.511378+00:00"},{"alias_kind":"pith_short_16","alias_value":"D75O2IKGNKCKW5XH","created_at":"2026-07-17T01:21:44.511378+00:00"},{"alias_kind":"pith_short_8","alias_value":"D75O2IKG","created_at":"2026-07-17T01:21:44.511378+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.29602","citing_title":"CogniVerse: Revolutionizing Multi-Modal Retrieval-Augmented Generation with Cognitive Reflection and Geometric Reasoning","ref_index":75,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15458","citing_title":"Video Models Can Reason with Verifiable Rewards","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19234","citing_title":"Learning to Credit the Right Steps: Objective-aware Process Optimization for Visual Generation","ref_index":22,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG","json":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG.json","graph_json":"https://pith.science/api/pith-number/D75O2IKGNKCKW5XHQADKDWGECG/graph.json","events_json":"https://pith.science/api/pith-number/D75O2IKGNKCKW5XHQADKDWGECG/events.json","paper":"https://pith.science/paper/D75O2IKG"},"agent_actions":{"view_html":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG","download_json":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG.json","view_paper":"https://pith.science/paper/D75O2IKG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.19356&json=true","fetch_graph":"https://pith.science/api/pith-number/D75O2IKGNKCKW5XHQADKDWGECG/graph.json","fetch_events":"https://pith.science/api/pith-number/D75O2IKGNKCKW5XHQADKDWGECG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG/action/storage_attestation","attest_author":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG/action/author_attestation","sign_citation":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG/action/citation_signature","submit_replication":"https://pith.science/pith/D75O2IKGNKCKW5XHQADKDWGECG/action/replication_record"}},"created_at":"2026-07-17T01:21:44.511378+00:00","updated_at":"2026-07-17T01:21:44.511378+00:00"}