{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GHAVRVZFRHPXZWCD5G7OLUSV26","short_pith_number":"pith:GHAVRVZF","schema_version":"1.0","canonical_sha256":"31c158d72589df7cd843e9bee5d255d7ba1eaf21749dfe79ebe22c32118a2256","source":{"kind":"arxiv","id":"2602.05897","version":2},"attestation_state":"computed","paper":{"title":"Stop Rewarding Hallucinated Steps: Faithfulness-Aware Step-Level Reinforcement Learning for Small Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chao Wang, Hexuan Deng, Min Zhang, Ruiyu Fang, Shuangyong Song, Shuo Nie, Xuebo Liu, Xuelong Li, Yu Li","submitted_at":"2026-02-05T17:15:12Z","abstract_excerpt":"As large language models become smaller and more efficient, small reasoning models (SRMs) are crucial for enabling chain-of-thought (CoT) reasoning in resource-constrained settings. However, they are prone to faithfulness hallucinations, especially in intermediate reasoning steps. Existing mitigation methods based on online reinforcement learning rely on outcome-based rewards or coarse-grained CoT evaluation, which can inadvertently reinforce unfaithful reasoning when the final answer is correct. To address these limitations, we propose Faithfulness-Aware Step-Level Reinforcement Learning (Fai"},"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":"2602.05897","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-02-05T17:15:12Z","cross_cats_sorted":[],"title_canon_sha256":"48fe5685bf33dce4f8533a249f14fdb33a96f569bafa0be200e320e546dbddb3","abstract_canon_sha256":"7afa869029282f14689b1e12f2c495207340c78cd61e3345759d3e71ca739d28"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:36.452574Z","signature_b64":"SNo7dRYTZWULPjqVkdYTkl2CGDs580hbrEcy0FbO3siZ2sSzJr20dukUw1Gfn7vHjvoTJfKVTGIcLWFgogagCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"31c158d72589df7cd843e9bee5d255d7ba1eaf21749dfe79ebe22c32118a2256","last_reissued_at":"2026-05-28T01:04:36.452046Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:36.452046Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stop Rewarding Hallucinated Steps: Faithfulness-Aware Step-Level Reinforcement Learning for Small Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chao Wang, Hexuan Deng, Min Zhang, Ruiyu Fang, Shuangyong Song, Shuo Nie, Xuebo Liu, Xuelong Li, Yu Li","submitted_at":"2026-02-05T17:15:12Z","abstract_excerpt":"As large language models become smaller and more efficient, small reasoning models (SRMs) are crucial for enabling chain-of-thought (CoT) reasoning in resource-constrained settings. However, they are prone to faithfulness hallucinations, especially in intermediate reasoning steps. Existing mitigation methods based on online reinforcement learning rely on outcome-based rewards or coarse-grained CoT evaluation, which can inadvertently reinforce unfaithful reasoning when the final answer is correct. To address these limitations, we propose Faithfulness-Aware Step-Level Reinforcement Learning (Fai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.05897","kind":"arxiv","version":2},"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/2602.05897/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":"2602.05897","created_at":"2026-05-28T01:04:36.452114+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.05897v2","created_at":"2026-05-28T01:04:36.452114+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.05897","created_at":"2026-05-28T01:04:36.452114+00:00"},{"alias_kind":"pith_short_12","alias_value":"GHAVRVZFRHPX","created_at":"2026-05-28T01:04:36.452114+00:00"},{"alias_kind":"pith_short_16","alias_value":"GHAVRVZFRHPXZWCD","created_at":"2026-05-28T01:04:36.452114+00:00"},{"alias_kind":"pith_short_8","alias_value":"GHAVRVZF","created_at":"2026-05-28T01:04:36.452114+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/GHAVRVZFRHPXZWCD5G7OLUSV26","json":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26.json","graph_json":"https://pith.science/api/pith-number/GHAVRVZFRHPXZWCD5G7OLUSV26/graph.json","events_json":"https://pith.science/api/pith-number/GHAVRVZFRHPXZWCD5G7OLUSV26/events.json","paper":"https://pith.science/paper/GHAVRVZF"},"agent_actions":{"view_html":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26","download_json":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26.json","view_paper":"https://pith.science/paper/GHAVRVZF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.05897&json=true","fetch_graph":"https://pith.science/api/pith-number/GHAVRVZFRHPXZWCD5G7OLUSV26/graph.json","fetch_events":"https://pith.science/api/pith-number/GHAVRVZFRHPXZWCD5G7OLUSV26/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26/action/storage_attestation","attest_author":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26/action/author_attestation","sign_citation":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26/action/citation_signature","submit_replication":"https://pith.science/pith/GHAVRVZFRHPXZWCD5G7OLUSV26/action/replication_record"}},"created_at":"2026-05-28T01:04:36.452114+00:00","updated_at":"2026-05-28T01:04:36.452114+00:00"}