{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:LRNCKBYC45PQ2FTPITTZ2HETAO","short_pith_number":"pith:LRNCKBYC","schema_version":"1.0","canonical_sha256":"5c5a250702e75f0d166f44e79d1c9303bff0d6cda6f0eeb8e2977081e8aad043","source":{"kind":"arxiv","id":"2602.06717","version":2},"attestation_state":"computed","paper":{"title":"F-GRPO: Don't Let Your Policy Learn the Obvious and Forget the Rare","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexey Gorbatovski, Alexey Malakhov, Boris Shaposhnikov, Daniil Gavrilov, Daniil Plyusov, Daria Korotyshova, Viacheslav Sinii","submitted_at":"2026-02-06T14:07:30Z","abstract_excerpt":"Reinforcement Learning with Verifiable Rewards (RLVR) is commonly based on group sampling to estimate advantages and stabilize policy updates. In practice, computational limits often rule out very large groups, so training proceeds with finite rollout sets that can reinforce only the correct behavior they expose. At practical group sizes, updates can miss rare-correct trajectories while still containing mixed rewards, concentrating probability on more common sampled solutions. We derive the probability of such prompt-local tail-miss events as a function of group size, showing non-monotonic beh"},"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":"2602.06717","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-06T14:07:30Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b61bfe4dea16770959d19fff89edd9913716a5e0d8a8d6923fdfd16f316cfc40","abstract_canon_sha256":"7ea809f04e4533ff76af64da061924b151be809d01ebce5af1770e5ff550a83b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:05.328833Z","signature_b64":"BUPP7Fys2t60pLlWOAGIeppZEYLUYPRa1+MI676ta8F2RwCmslqLzLufcHU/lW+qlUz640LzIcYMjlMvs4KICQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5c5a250702e75f0d166f44e79d1c9303bff0d6cda6f0eeb8e2977081e8aad043","last_reissued_at":"2026-05-26T02:04:05.328090Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:05.328090Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"F-GRPO: Don't Let Your Policy Learn the Obvious and Forget the Rare","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alexey Gorbatovski, Alexey Malakhov, Boris Shaposhnikov, Daniil Gavrilov, Daniil Plyusov, Daria Korotyshova, Viacheslav Sinii","submitted_at":"2026-02-06T14:07:30Z","abstract_excerpt":"Reinforcement Learning with Verifiable Rewards (RLVR) is commonly based on group sampling to estimate advantages and stabilize policy updates. In practice, computational limits often rule out very large groups, so training proceeds with finite rollout sets that can reinforce only the correct behavior they expose. At practical group sizes, updates can miss rare-correct trajectories while still containing mixed rewards, concentrating probability on more common sampled solutions. We derive the probability of such prompt-local tail-miss events as a function of group size, showing non-monotonic beh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.06717","kind":"arxiv","version":2},"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/2602.06717/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":"2602.06717","created_at":"2026-05-26T02:04:05.328202+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.06717v2","created_at":"2026-05-26T02:04:05.328202+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.06717","created_at":"2026-05-26T02:04:05.328202+00:00"},{"alias_kind":"pith_short_12","alias_value":"LRNCKBYC45PQ","created_at":"2026-05-26T02:04:05.328202+00:00"},{"alias_kind":"pith_short_16","alias_value":"LRNCKBYC45PQ2FTP","created_at":"2026-05-26T02:04:05.328202+00:00"},{"alias_kind":"pith_short_8","alias_value":"LRNCKBYC","created_at":"2026-05-26T02:04:05.328202+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.00365","citing_title":"Uniform-Correct Policy Optimization: Breaking RLVR's Indifference to Diversity","ref_index":43,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07039","citing_title":"PACEvolve++: Improving Test-time Learning for Evolutionary Search Agents","ref_index":27,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO","json":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO.json","graph_json":"https://pith.science/api/pith-number/LRNCKBYC45PQ2FTPITTZ2HETAO/graph.json","events_json":"https://pith.science/api/pith-number/LRNCKBYC45PQ2FTPITTZ2HETAO/events.json","paper":"https://pith.science/paper/LRNCKBYC"},"agent_actions":{"view_html":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO","download_json":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO.json","view_paper":"https://pith.science/paper/LRNCKBYC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.06717&json=true","fetch_graph":"https://pith.science/api/pith-number/LRNCKBYC45PQ2FTPITTZ2HETAO/graph.json","fetch_events":"https://pith.science/api/pith-number/LRNCKBYC45PQ2FTPITTZ2HETAO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO/action/storage_attestation","attest_author":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO/action/author_attestation","sign_citation":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO/action/citation_signature","submit_replication":"https://pith.science/pith/LRNCKBYC45PQ2FTPITTZ2HETAO/action/replication_record"}},"created_at":"2026-05-26T02:04:05.328202+00:00","updated_at":"2026-05-26T02:04:05.328202+00:00"}