{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:Y5BR26G7LCNH3OGEJYBPXFAPA6","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"769839a6b2f431823f2ce011bc079382dda2ac288ba14a42e289655beab38918","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T08:33:02Z","title_canon_sha256":"381fa8f479b09790d076d448166f439f82aaf7e52afc4bf0e30ededcc7af5a81"},"schema_version":"1.0","source":{"id":"2605.14558","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14558","created_at":"2026-05-17T23:39:05Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14558v1","created_at":"2026-05-17T23:39:05Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14558","created_at":"2026-05-17T23:39:05Z"},{"alias_kind":"pith_short_12","alias_value":"Y5BR26G7LCNH","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"Y5BR26G7LCNH3OGE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"Y5BR26G7","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:c67e1fa3a048192babe0dbf22e2afb2b87cf6223de6ba24f7b199f9d0db1934f","target":"graph","created_at":"2026-05-17T23:39:05Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"token-level training signals, quantified by their correlations with reward variance of different rollouts sampled from a given prompt, concentrate sharply on action tokens rather than reasoning tokens, even though action tokens account for only a small fraction of the trajectory"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"that down-weighting reasoning tokens and boosting high-uncertainty action tokens will not degrade the quality of the reasoning chain itself or introduce new instabilities in long-horizon trajectories"},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Token-level signals concentrate on action tokens in agentic RL, so reweighting gradients toward them outperforms uniform policy gradients."}],"snapshot_sha256":"89df5c0110227a37e42f43dd209e275d1d2e766ec454baf5a00e52c833d09f07"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2658e067e3a21f2e238b50c054966600406cf2f3e106331f0fd19623b5045d36"},"paper":{"abstract_excerpt":"Agentic reinforcement learning trains large language models using multi-turn trajectories that interleave long reasoning traces with short environment-facing actions. Common policy-gradient methods, such as PPO and GRPO, treat each token in a trajectory equally, leading to uniform credit assignment. In this paper, we critically demonstrate that such uniform credit assignment largely misallocates token-level training signals. From an energy-based modeling perspective, we show that token-level training signals, quantified by their correlations with reward variance of different rollouts sampled f","authors_text":"David Wipf, Henry Peng Zou, Junhua Liu, Junyou Zhu, Langzhou He, Philip S. Yu, Qitian Wu, Wei-Chieh Huang, Yue Zhou, Zhengyao Gu","cross_cats":["cs.AI","cs.CL"],"headline":"Token-level signals concentrate on action tokens in agentic RL, so reweighting gradients toward them outperforms uniform policy gradients.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T08:33:02Z","title":"Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy"},"references":{"count":39,"internal_anchors":17,"resolved_work":39,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Lmrl gym: Benchmarks for multi-turn reinforcement learn- ing with language models","work_id":"bad1baad-5c3f-4456-ac96-29c8f5e78bfb","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Chan, Hao Sun, Samuel Holt, and Mihaela van der Schaar","work_id":"c8748d0b-32ee-428e-a008-b6fee31a2bff","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"arXiv preprint arXiv:2502.01600 , year=","work_id":"7e929792-6a2a-42ff-a1db-763f890d8b4e","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Process reinforcement through implicit rewards,","work_id":"9ea5754d-d348-48ee-9421-97a9f4c6d3eb","year":null},{"cited_arxiv_id":"2502.01456","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Process Reinforcement through Implicit Rewards","work_id":"c31a2126-86f9-44f3-91f3-208d0fc1463a","year":null}],"snapshot_sha256":"8887903501c992eb3daf01e2640401d8063f10ac6db147c7fe14f36e48e00b83"},"source":{"id":"2605.14558","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T01:43:12.935990Z","id":"0fc00fb7-6634-456c-8f63-2e929df9641c","model_set":{"reader":"grok-4.3"},"one_line_summary":"ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Token-level signals concentrate on action tokens in agentic RL, so reweighting gradients toward them outperforms uniform policy gradients.","strongest_claim":"token-level training signals, quantified by their correlations with reward variance of different rollouts sampled from a given prompt, concentrate sharply on action tokens rather than reasoning tokens, even though action tokens account for only a small fraction of the trajectory","weakest_assumption":"that down-weighting reasoning tokens and boosting high-uncertainty action tokens will not degrade the quality of the reasoning chain itself or introduce new instabilities in long-horizon trajectories"}},"verdict_id":"0fc00fb7-6634-456c-8f63-2e929df9641c"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:4ded5eb164d1f9c3ff776767746d4431716bd7ce53f530c6f0d5d50c25514c0d","target":"record","created_at":"2026-05-17T23:39:05Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"769839a6b2f431823f2ce011bc079382dda2ac288ba14a42e289655beab38918","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T08:33:02Z","title_canon_sha256":"381fa8f479b09790d076d448166f439f82aaf7e52afc4bf0e30ededcc7af5a81"},"schema_version":"1.0","source":{"id":"2605.14558","kind":"arxiv","version":1}},"canonical_sha256":"c7431d78df589a7db8c44e02fb940f07806e1939ab66a013470fbf4fc8a89aab","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c7431d78df589a7db8c44e02fb940f07806e1939ab66a013470fbf4fc8a89aab","first_computed_at":"2026-05-17T23:39:05.628152Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:05.628152Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"AMh57IO1yLCmyvqcVWopqckg5PFtwAtn2gjTxBEjEPVoYbKxNh81jPR3AIrA8kMI7D3QyjLgiu3J5enTHosvCw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:05.628770Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14558","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:4ded5eb164d1f9c3ff776767746d4431716bd7ce53f530c6f0d5d50c25514c0d","sha256:c67e1fa3a048192babe0dbf22e2afb2b87cf6223de6ba24f7b199f9d0db1934f"],"state_sha256":"e297fa6a3570353fd98997fc35c7e37cf643a0cb1db02c28096171b41c58448f"}