{"paper":{"title":"Verifiable Process Rewards for Agentic Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Converting oracles into dense turn-level rewards improves credit assignment for long-horizon LLM agent reasoning.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Chao Yu, Huaijie Wang, Huining Yuan, Jiaxuan Gao, Xiangmin Yi, Xiao-Ping Zhang, Yi Wu, Yu Wang, Zelai Xu","submitted_at":"2026-05-11T10:30:53Z","abstract_excerpt":"Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning abilities of large language models (LLMs), but most existing approaches rely on sparse outcome-level feedback. This sparsity creates a credit assignment challenge in long-horizon agentic reasoning: a trajectory may fail despite containing many correct intermediate decisions, or succeed despite containing flawed ones. In this work, we study a class of densely-verifiable agentic reasoning problems, where intermediate actions can be objectively checked by symbolic or algorithmic oracles. We propose Verifiable Process"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"dense verifier-grounded rewards can improve long-horizon credit assignment by providing more localized learning signals, with the benefit depending on the reliability of the verifier. Empirically, VPR outperforms outcome-level reward and rollout-based process reward baselines across controlled environments, and more importantly, transfers to both general and agentic reasoning benchmarks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That reliable symbolic, algorithmic, or posterior-based oracles exist for verifying intermediate actions in the studied densely-verifiable agentic reasoning problems, and that the learned skills transfer meaningfully beyond the specific training environments.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Verifiable Process Rewards (VPR) converts symbolic oracles into dense turn-level supervision for reinforcement learning in agentic reasoning, outperforming outcome-only rewards and transferring to general benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Converting oracles into dense turn-level rewards improves credit assignment for long-horizon LLM agent reasoning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5a80f2e6b22834041601d60e6a06411bcac1561c146f1d549fb15dd32f872c33"},"source":{"id":"2605.10325","kind":"arxiv","version":2},"verdict":{"id":"dffaaed6-739f-4f55-a565-297817bdf206","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T04:54:42.341354Z","strongest_claim":"dense verifier-grounded rewards can improve long-horizon credit assignment by providing more localized learning signals, with the benefit depending on the reliability of the verifier. Empirically, VPR outperforms outcome-level reward and rollout-based process reward baselines across controlled environments, and more importantly, transfers to both general and agentic reasoning benchmarks.","one_line_summary":"Verifiable Process Rewards (VPR) converts symbolic oracles into dense turn-level supervision for reinforcement learning in agentic reasoning, outperforming outcome-only rewards and transferring to general benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That reliable symbolic, algorithmic, or posterior-based oracles exist for verifying intermediate actions in the studied densely-verifiable agentic reasoning problems, and that the learned skills transfer meaningfully beyond the specific training environments.","pith_extraction_headline":"Converting oracles into dense turn-level rewards improves credit assignment for long-horizon LLM agent reasoning."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10325/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T06:02:01.181511Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T15:35:52.703771Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T11:31:18.913231Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:28:02.357747Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"6af5b09c74a15e3f5f3775966c56be195131b94fa01518981aacabe99cfc8cb5"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ab14a2bba61a9cf04afef2acb9fb54b785580336a47c1a077436088b88f1d0c8"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}