{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:H4M7WT3WJKADRQIGFG3R4SH7A7","short_pith_number":"pith:H4M7WT3W","schema_version":"1.0","canonical_sha256":"3f19fb4f764a8038c10629b71e48ff07fba97f39cf9fadad9aaa01454263ed1a","source":{"kind":"arxiv","id":"2605.22217","version":1},"attestation_state":"computed","paper":{"title":"Survive or Collapse: The Asymmetric Roles of Data Gating and Reward Grounding in Self-Play RL","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chengzhi Liu, Gaowen Liu, Jayanth Srinivasa, Sophia Xiao Pu, William Yang Wang, Xin Eric Wang, Zhaotian Weng","submitted_at":"2026-05-21T09:19:23Z","abstract_excerpt":"Self-play reinforcement learning trains language models on their own generated tasks, co-evolving a proposer and solver without human labels. Recent systems report strong reasoning gains, but collapse and instability are widely observed and poorly understood. The dominant response treats this as a reward-design problem. We argue instead that self-play stability is governed by two distinct levers: a data-level gate that decides which proposer-generated tasks enter the training pool, and the reward signal that updates the policy on tasks already admitted. Through controlled experiments on a Pyth"},"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":"2605.22217","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-21T09:19:23Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"41fdbbacecf77e264af1e7a7a377763fa4c2aec3e8c17b83d17f77d172eb496c","abstract_canon_sha256":"e341ea0ca36fe33c88963fc62e4941e62344e2cc5a0b5678233f825cbb7c6c89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-22T01:04:32.675826Z","signature_b64":"zi6rRvQP1S9N2cPhxzoQ6ha+E2A7B4Ff8Q7v9rJSZ5OMl0kNEdDcgf3pNr5NoEn2kKjfF3lnCiGUSjQMHTDwBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3f19fb4f764a8038c10629b71e48ff07fba97f39cf9fadad9aaa01454263ed1a","last_reissued_at":"2026-05-22T01:04:32.674978Z","signature_status":"signed_v1","first_computed_at":"2026-05-22T01:04:32.674978Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Survive or Collapse: The Asymmetric Roles of Data Gating and Reward Grounding in Self-Play RL","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chengzhi Liu, Gaowen Liu, Jayanth Srinivasa, Sophia Xiao Pu, William Yang Wang, Xin Eric Wang, Zhaotian Weng","submitted_at":"2026-05-21T09:19:23Z","abstract_excerpt":"Self-play reinforcement learning trains language models on their own generated tasks, co-evolving a proposer and solver without human labels. Recent systems report strong reasoning gains, but collapse and instability are widely observed and poorly understood. The dominant response treats this as a reward-design problem. We argue instead that self-play stability is governed by two distinct levers: a data-level gate that decides which proposer-generated tasks enter the training pool, and the reward signal that updates the policy on tasks already admitted. Through controlled experiments on a Pyth"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.22217","kind":"arxiv","version":1},"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/2605.22217/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":"2605.22217","created_at":"2026-05-22T01:04:32.675071+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.22217v1","created_at":"2026-05-22T01:04:32.675071+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.22217","created_at":"2026-05-22T01:04:32.675071+00:00"},{"alias_kind":"pith_short_12","alias_value":"H4M7WT3WJKAD","created_at":"2026-05-22T01:04:32.675071+00:00"},{"alias_kind":"pith_short_16","alias_value":"H4M7WT3WJKADRQIG","created_at":"2026-05-22T01:04:32.675071+00:00"},{"alias_kind":"pith_short_8","alias_value":"H4M7WT3W","created_at":"2026-05-22T01:04:32.675071+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/H4M7WT3WJKADRQIGFG3R4SH7A7","json":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7.json","graph_json":"https://pith.science/api/pith-number/H4M7WT3WJKADRQIGFG3R4SH7A7/graph.json","events_json":"https://pith.science/api/pith-number/H4M7WT3WJKADRQIGFG3R4SH7A7/events.json","paper":"https://pith.science/paper/H4M7WT3W"},"agent_actions":{"view_html":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7","download_json":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7.json","view_paper":"https://pith.science/paper/H4M7WT3W","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.22217&json=true","fetch_graph":"https://pith.science/api/pith-number/H4M7WT3WJKADRQIGFG3R4SH7A7/graph.json","fetch_events":"https://pith.science/api/pith-number/H4M7WT3WJKADRQIGFG3R4SH7A7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7/action/storage_attestation","attest_author":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7/action/author_attestation","sign_citation":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7/action/citation_signature","submit_replication":"https://pith.science/pith/H4M7WT3WJKADRQIGFG3R4SH7A7/action/replication_record"}},"created_at":"2026-05-22T01:04:32.675071+00:00","updated_at":"2026-05-22T01:04:32.675071+00:00"}