{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NUOI7CQO345ATWPL5IUQKTVQYB","short_pith_number":"pith:NUOI7CQO","schema_version":"1.0","canonical_sha256":"6d1c8f8a0edf3a09d9ebea29054eb0c06fa606e5dba76028ea0fb1f81af8fb82","source":{"kind":"arxiv","id":"2506.09096","version":3},"attestation_state":"computed","paper":{"title":"Intra-Trajectory Consistency for Reward Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bo Du, Chaoyang Zhou, Dacheng Tao, Di Wang, Rong-Cheng Tu, Shunyu Liu, Zengmao Wang","submitted_at":"2025-06-10T12:59:14Z","abstract_excerpt":"Reward models are critical for improving large language models (LLMs), particularly in reinforcement learning from human feedback (RLHF) or inference-time verification. Current reward modeling typically relies on scores of overall responses to learn the outcome rewards for the responses. However, since the response-level scores are coarse-grained supervision signals, the reward model struggles to identify the specific components within a response trajectory that truly correlate with the scores, leading to poor generalization on unseen responses. In this paper, we propose to leverage generation"},"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":"2506.09096","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-06-10T12:59:14Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"29e3570183e42d9aa8c530baa966832c80e9a37cca12d87b2c87bed40eecfb93","abstract_canon_sha256":"cb983dba1c2471c31b2b4996ff14cb1f8f3aeb18d8255e35332da483d0954eed"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:21:49.222619Z","signature_b64":"RwNuIyqTe7V63JfM4Wwl417IrEXyQKiSYJgDsd3Rf3AI6PZfD2a9BwxoM/rQwyGlyap3KsHIkJ5fuPJFedwFDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d1c8f8a0edf3a09d9ebea29054eb0c06fa606e5dba76028ea0fb1f81af8fb82","last_reissued_at":"2026-07-05T11:21:49.222151Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:21:49.222151Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Intra-Trajectory Consistency for Reward Modeling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bo Du, Chaoyang Zhou, Dacheng Tao, Di Wang, Rong-Cheng Tu, Shunyu Liu, Zengmao Wang","submitted_at":"2025-06-10T12:59:14Z","abstract_excerpt":"Reward models are critical for improving large language models (LLMs), particularly in reinforcement learning from human feedback (RLHF) or inference-time verification. Current reward modeling typically relies on scores of overall responses to learn the outcome rewards for the responses. However, since the response-level scores are coarse-grained supervision signals, the reward model struggles to identify the specific components within a response trajectory that truly correlate with the scores, leading to poor generalization on unseen responses. In this paper, we propose to leverage generation"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09096","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/2506.09096/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":"2506.09096","created_at":"2026-07-05T11:21:49.222206+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.09096v3","created_at":"2026-07-05T11:21:49.222206+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09096","created_at":"2026-07-05T11:21:49.222206+00:00"},{"alias_kind":"pith_short_12","alias_value":"NUOI7CQO345A","created_at":"2026-07-05T11:21:49.222206+00:00"},{"alias_kind":"pith_short_16","alias_value":"NUOI7CQO345ATWPL","created_at":"2026-07-05T11:21:49.222206+00:00"},{"alias_kind":"pith_short_8","alias_value":"NUOI7CQO","created_at":"2026-07-05T11:21:49.222206+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/NUOI7CQO345ATWPL5IUQKTVQYB","json":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB.json","graph_json":"https://pith.science/api/pith-number/NUOI7CQO345ATWPL5IUQKTVQYB/graph.json","events_json":"https://pith.science/api/pith-number/NUOI7CQO345ATWPL5IUQKTVQYB/events.json","paper":"https://pith.science/paper/NUOI7CQO"},"agent_actions":{"view_html":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB","download_json":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB.json","view_paper":"https://pith.science/paper/NUOI7CQO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.09096&json=true","fetch_graph":"https://pith.science/api/pith-number/NUOI7CQO345ATWPL5IUQKTVQYB/graph.json","fetch_events":"https://pith.science/api/pith-number/NUOI7CQO345ATWPL5IUQKTVQYB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB/action/storage_attestation","attest_author":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB/action/author_attestation","sign_citation":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB/action/citation_signature","submit_replication":"https://pith.science/pith/NUOI7CQO345ATWPL5IUQKTVQYB/action/replication_record"}},"created_at":"2026-07-05T11:21:49.222206+00:00","updated_at":"2026-07-05T11:21:49.222206+00:00"}