{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IJFJSDQODJBX3RRX75DHW35VFF","short_pith_number":"pith:IJFJSDQO","schema_version":"1.0","canonical_sha256":"424a990e0e1a437dc637ff467b6fb52971932a423c02404d22b9ece45a8517bd","source":{"kind":"arxiv","id":"2602.05370","version":3},"attestation_state":"computed","paper":{"title":"Mining or Synthesis? Rethinking Exploration Efficiency in Iterative Alignment of Mathematical Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Guhan Chen, Hejin Wang, Jiansheng Wei, Jing Li, Jun Rao, Min Zhang, Xiaojun Meng, Xuebo Liu, Zixiong Yu","submitted_at":"2026-02-05T06:47:40Z","abstract_excerpt":"Iterative Direct Preference Optimization (DPO) has emerged as a widely used paradigm for aligning Large Language Models on reasoning tasks. Existing approaches typically rely on Best-of-N sampling ($N\\geq8$) to mine positive trajectories from the distribution tail. In this work, we show that in mathematical reasoning, increasing $N$ yields diminishing returns while increasing verifier-induced false-positive risk and the distribution shift required for policy updates. To address this, we introduce PACE (Proximal Alignment via Corrective Exploration), a generation-based corrective framework that"},"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":"2602.05370","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-02-05T06:47:40Z","cross_cats_sorted":[],"title_canon_sha256":"2df72fa5de501a7f53e3cae46ab94157c3b5d90adc26a0fe9c1c026381187f73","abstract_canon_sha256":"cfcdccda2ede0b9ccf6f8d4b19427305cb3e82d5477dbde2ac8cf5dd5a561e11"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:05:03.527141Z","signature_b64":"D1DKi27Y59h2OSm3i5/g4O/1j5h2zo32ASMR6sdoHtBugZB8qHjV2pT0pPja+6BNzWOcDy2OBQ3wo7nGgGpdCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"424a990e0e1a437dc637ff467b6fb52971932a423c02404d22b9ece45a8517bd","last_reissued_at":"2026-05-29T01:05:03.526134Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:05:03.526134Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mining or Synthesis? Rethinking Exploration Efficiency in Iterative Alignment of Mathematical Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Guhan Chen, Hejin Wang, Jiansheng Wei, Jing Li, Jun Rao, Min Zhang, Xiaojun Meng, Xuebo Liu, Zixiong Yu","submitted_at":"2026-02-05T06:47:40Z","abstract_excerpt":"Iterative Direct Preference Optimization (DPO) has emerged as a widely used paradigm for aligning Large Language Models on reasoning tasks. Existing approaches typically rely on Best-of-N sampling ($N\\geq8$) to mine positive trajectories from the distribution tail. In this work, we show that in mathematical reasoning, increasing $N$ yields diminishing returns while increasing verifier-induced false-positive risk and the distribution shift required for policy updates. To address this, we introduce PACE (Proximal Alignment via Corrective Exploration), a generation-based corrective framework that"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.05370","kind":"arxiv","version":3},"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/2602.05370/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":"2602.05370","created_at":"2026-05-29T01:05:03.526259+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.05370v3","created_at":"2026-05-29T01:05:03.526259+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.05370","created_at":"2026-05-29T01:05:03.526259+00:00"},{"alias_kind":"pith_short_12","alias_value":"IJFJSDQODJBX","created_at":"2026-05-29T01:05:03.526259+00:00"},{"alias_kind":"pith_short_16","alias_value":"IJFJSDQODJBX3RRX","created_at":"2026-05-29T01:05:03.526259+00:00"},{"alias_kind":"pith_short_8","alias_value":"IJFJSDQO","created_at":"2026-05-29T01:05:03.526259+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/IJFJSDQODJBX3RRX75DHW35VFF","json":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF.json","graph_json":"https://pith.science/api/pith-number/IJFJSDQODJBX3RRX75DHW35VFF/graph.json","events_json":"https://pith.science/api/pith-number/IJFJSDQODJBX3RRX75DHW35VFF/events.json","paper":"https://pith.science/paper/IJFJSDQO"},"agent_actions":{"view_html":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF","download_json":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF.json","view_paper":"https://pith.science/paper/IJFJSDQO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.05370&json=true","fetch_graph":"https://pith.science/api/pith-number/IJFJSDQODJBX3RRX75DHW35VFF/graph.json","fetch_events":"https://pith.science/api/pith-number/IJFJSDQODJBX3RRX75DHW35VFF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF/action/storage_attestation","attest_author":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF/action/author_attestation","sign_citation":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF/action/citation_signature","submit_replication":"https://pith.science/pith/IJFJSDQODJBX3RRX75DHW35VFF/action/replication_record"}},"created_at":"2026-05-29T01:05:03.526259+00:00","updated_at":"2026-05-29T01:05:03.526259+00:00"}