{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:QZVCWZPCXSQYQRXNNRUXFWCNGR","short_pith_number":"pith:QZVCWZPC","schema_version":"1.0","canonical_sha256":"866a2b65e2bca18846ed6c6972d84d344d8356b8094bb826ed3cdcdbff15f6ac","source":{"kind":"arxiv","id":"2605.13486","version":1},"attestation_state":"computed","paper":{"title":"R^2-Mem: Reflective Experience for Memory Search","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"R^2-Mem distills abstract experiences from scored past trajectories to guide LLM agents away from repeated search errors without reinforcement learning.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Junkang Wu, Wenyu Mao, Xiangnan He, Xiang Wang, Xinyuan Wang","submitted_at":"2026-05-13T13:09:36Z","abstract_excerpt":"Deep search has recently emerged as a promising paradigm for enabling agents to retrieve fine-grained historical information without heavy memory pre-managed. However, existing deep search agents for memory system repeat past error behaviors because they fail to learn from the prior high- and low-quality search trajectories. To address this limitation, we propose R^2-Mem, a reflective experience framework for memory search systems. In the offline stage, a Rubric-guided Evaluator scores low- and high-quality steps in historical trajectories, and a self-Reflection Learner distills the correspond"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.13486","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T13:09:36Z","cross_cats_sorted":[],"title_canon_sha256":"649b4779aabb9fccb77f698a33dbc2290f0ffebd8a0bbe469d355409cdfa7fd3","abstract_canon_sha256":"44514237febfe7091fed30834c9849bf363845afb57451f69902fcfce6bf9715"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:41.266299Z","signature_b64":"/0njiFI62x7OqLHDoxhLC1Fdvr/H/mGUFwtlOfVebteRgbJ80qR72fMpjczGIQSkrCL+0ay9XzfCuMWX9gWfDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"866a2b65e2bca18846ed6c6972d84d344d8356b8094bb826ed3cdcdbff15f6ac","last_reissued_at":"2026-05-18T02:44:41.265847Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:41.265847Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"R^2-Mem: Reflective Experience for Memory Search","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"R^2-Mem distills abstract experiences from scored past trajectories to guide LLM agents away from repeated search errors without reinforcement learning.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Junkang Wu, Wenyu Mao, Xiangnan He, Xiang Wang, Xinyuan Wang","submitted_at":"2026-05-13T13:09:36Z","abstract_excerpt":"Deep search has recently emerged as a promising paradigm for enabling agents to retrieve fine-grained historical information without heavy memory pre-managed. However, existing deep search agents for memory system repeat past error behaviors because they fail to learn from the prior high- and low-quality search trajectories. To address this limitation, we propose R^2-Mem, a reflective experience framework for memory search systems. In the offline stage, a Rubric-guided Evaluator scores low- and high-quality steps in historical trajectories, and a self-Reflection Learner distills the correspond"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"R^2-Mem provides a RL-free and low-cost solution for self-improving LLM agents, improving F1 scores by up to 22.6%, while reducing token consumption by 12.9% and search iterations by 20.2%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the rubric-guided evaluator accurately distinguishes high- and low-quality steps and that the distilled experiences generalize to new queries without overfitting to the offline trajectories.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"R^2-Mem distills rubric-scored experiences from high- and low-quality search trajectories to guide LLM agents, raising F1 by up to 22.6% while cutting tokens 12.9% and iterations 20.2%.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"R^2-Mem distills abstract experiences from scored past trajectories to guide LLM agents away from repeated search errors without reinforcement learning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e3d6067f3144751a532cc637a305d66483791fbca725fce6d16dd8e62a0e7823"},"source":{"id":"2605.13486","kind":"arxiv","version":1},"verdict":{"id":"1cfbb5d1-5aeb-4659-b8a8-451689d5436c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:06:53.510192Z","strongest_claim":"R^2-Mem provides a RL-free and low-cost solution for self-improving LLM agents, improving F1 scores by up to 22.6%, while reducing token consumption by 12.9% and search iterations by 20.2%.","one_line_summary":"R^2-Mem distills rubric-scored experiences from high- and low-quality search trajectories to guide LLM agents, raising F1 by up to 22.6% while cutting tokens 12.9% and iterations 20.2%.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the rubric-guided evaluator accurately distinguishes high- and low-quality steps and that the distilled experiences generalize to new queries without overfitting to the offline trajectories.","pith_extraction_headline":"R^2-Mem distills abstract experiences from scored past trajectories to guide LLM agents away from repeated search errors without reinforcement learning."},"references":{"count":11,"sample":[{"doi":"10.18653/v1/2024.findings-acl.137","year":2024,"title":"M 3-embedding: Multi-linguality, multi-functionality, multi-granularity text embeddings through self-knowledge distillation","work_id":"6f249b44-4b0a-47dd-b5f4-1877fd9269ad","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1162/tacl_a_00023","year":2026,"title":"The N arrative QA Reading Comprehension Challenge","work_id":"82006945-c336-4164-9c68-f07125cc331d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"URLhttps://openreview.net/forum?id=jL7fwchScm. William F. Shen, Xinchi Qiu, Chenxi Whitehouse, Lisa Alazraki, Shashwat Goel, Francesco Barbieri, Timon Willi, Akhil Mathur, and Ilias Leontiadis. Rethin","work_id":"549989be-410e-4209-811a-07402cf7e4bc","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Memory-R1: Enhancing Large Language Model Agents to Manage and Utilize Memories via Reinforcement Learning","work_id":"2a070440-2167-4398-be97-c2d4c3ee3541","ref_index":4,"cited_arxiv_id":"2508.19828","is_internal_anchor":true},{"doi":"","year":null,"title":"Analyze why this planning trace is high-quality or low-quality","work_id":"0f37153a-dcc8-4592-9381-2f0f48dbb2fb","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":11,"snapshot_sha256":"eb909a04f21ec584e015dfb95e71530fa4e9bc19bd0236adc5c7f0a671f78872","internal_anchors":1},"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":"2605.13486","created_at":"2026-05-18T02:44:41.265912+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.13486v1","created_at":"2026-05-18T02:44:41.265912+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13486","created_at":"2026-05-18T02:44:41.265912+00:00"},{"alias_kind":"pith_short_12","alias_value":"QZVCWZPCXSQY","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"QZVCWZPCXSQYQRXN","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"QZVCWZPC","created_at":"2026-05-18T12:33:37.589309+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/QZVCWZPCXSQYQRXNNRUXFWCNGR","json":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR.json","graph_json":"https://pith.science/api/pith-number/QZVCWZPCXSQYQRXNNRUXFWCNGR/graph.json","events_json":"https://pith.science/api/pith-number/QZVCWZPCXSQYQRXNNRUXFWCNGR/events.json","paper":"https://pith.science/paper/QZVCWZPC"},"agent_actions":{"view_html":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR","download_json":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR.json","view_paper":"https://pith.science/paper/QZVCWZPC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.13486&json=true","fetch_graph":"https://pith.science/api/pith-number/QZVCWZPCXSQYQRXNNRUXFWCNGR/graph.json","fetch_events":"https://pith.science/api/pith-number/QZVCWZPCXSQYQRXNNRUXFWCNGR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR/action/storage_attestation","attest_author":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR/action/author_attestation","sign_citation":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR/action/citation_signature","submit_replication":"https://pith.science/pith/QZVCWZPCXSQYQRXNNRUXFWCNGR/action/replication_record"}},"created_at":"2026-05-18T02:44:41.265912+00:00","updated_at":"2026-05-18T02:44:41.265912+00:00"}