{"paper":{"title":"$S^3$-R1: Learning to Retrieve and Answer Step-by-Step with Synthetic Data","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Coupling synthetic multi-hop questions with rewards for search steps and answers enables models to learn effective retrieval strategies and generalize up to 10% better out of domain.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Akhil Udathu, Atharva Parulekar, Harsh Goel, Pradnesh Kalkar, Susmija Jabbireddy","submitted_at":"2026-05-02T05:01:05Z","abstract_excerpt":"Reinforcement learning (RL) post-training has enabled newer capabilities in models, such as agentic tool-use for search. However, these models struggle primarily due to limitations with sparse outcome-based rewards and a lack of training data that encapsulates questions of differing hardness, which results in models not performing deeper searches with tools to collect evidence for question-answering. To address these limitations, we introduce S^3-R1 (Synthetic data and stabilized Search R1), a framework that couples a data-centric approach with denser learning signals. We first develop a synth"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our evaluations show that S^3-R1 outperforms existing baselines by learning more effective search and synthesis strategies, yielding up to a 10% improvement in robust generalization on out-of-domain datasets.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The synthetic generation and retrieval-based verification pipeline produces questions of genuinely intermediate difficulty that transfer to real user queries without introducing distribution shift or annotation artifacts that inflate measured gains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"S^3-R1 generates synthetic intermediate-difficulty multi-hop questions and applies dense rewards for search quality plus answer correctness, yielding up to 10% better out-of-domain generalization than baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Coupling synthetic multi-hop questions with rewards for search steps and answers enables models to learn effective retrieval strategies and generalize up to 10% better out of domain.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1e0be816271a93b5b2ce8a405475b9b96d5fd22abc48d355f12fbd768549b879"},"source":{"id":"2605.01248","kind":"arxiv","version":3},"verdict":{"id":"5a42145d-1340-44a4-9004-dae1b835fed5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T15:05:21.542628Z","strongest_claim":"Our evaluations show that S^3-R1 outperforms existing baselines by learning more effective search and synthesis strategies, yielding up to a 10% improvement in robust generalization on out-of-domain datasets.","one_line_summary":"S^3-R1 generates synthetic intermediate-difficulty multi-hop questions and applies dense rewards for search quality plus answer correctness, yielding up to 10% better out-of-domain generalization than baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The synthetic generation and retrieval-based verification pipeline produces questions of genuinely intermediate difficulty that transfer to real user queries without introducing distribution shift or annotation artifacts that inflate measured gains.","pith_extraction_headline":"Coupling synthetic multi-hop questions with rewards for search steps and answers enables models to learn effective retrieval strategies and generalize up to 10% better out of domain."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.01248/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T18:36:11.355256Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T17:28:38.166668Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"304b5c8b06090ffc52ef51d631cf34569b03302993c92030bfeae2c01d150f4b"},"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"}