{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:UKG6RHDKYB2RW5DM46JYUJJOJY","short_pith_number":"pith:UKG6RHDK","schema_version":"1.0","canonical_sha256":"a28de89c6ac0751b746ce7938a252e4e1b13e25daccbf909d83ebc80ae680adb","source":{"kind":"arxiv","id":"2605.30451","version":1},"attestation_state":"computed","paper":{"title":"VeriGate: Verifier-Gated Step-Level Supervision for GRPO","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aakriti Agrawal, Furong Huang, Minghui Liu","submitted_at":"2026-05-28T18:20:32Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) is an effective recipe for training reasoning models with verifier-based outcome rewards, but its supervision is sparse: when all sampled trajectories for a prompt receive the same verifier reward, the group-relative advantage collapses to zero and learning stalls. Outcome-only rewards also provide no step-level credit assignment, limiting exploration and making it harder to learn robust reasoning. We present VeriGate (Verifier-Gated Step-Level GRPO), a verifier-gated extension of GRPO that addresses these limitations with three design choices. First, "},"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":"2605.30451","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-28T18:20:32Z","cross_cats_sorted":[],"title_canon_sha256":"96fececb83b2c09d5e577a25bf258e9ad9db9ae2c2fd629627ffcb4c619a4958","abstract_canon_sha256":"5ff7906b3ad95cc74a426adf7fe2bdd41f4c6072115d27f32de087925eda792d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T01:02:55.005643Z","signature_b64":"/le32pXuEz3ptWRHIfWf7GSwIZYOxWa+X9EDY3kRlRpwDvjOOppSBCoAbeizutOTyPdWnrkr7HOSIiacfQv8DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a28de89c6ac0751b746ce7938a252e4e1b13e25daccbf909d83ebc80ae680adb","last_reissued_at":"2026-06-01T01:02:55.004571Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T01:02:55.004571Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"VeriGate: Verifier-Gated Step-Level Supervision for GRPO","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aakriti Agrawal, Furong Huang, Minghui Liu","submitted_at":"2026-05-28T18:20:32Z","abstract_excerpt":"Group Relative Policy Optimization (GRPO) is an effective recipe for training reasoning models with verifier-based outcome rewards, but its supervision is sparse: when all sampled trajectories for a prompt receive the same verifier reward, the group-relative advantage collapses to zero and learning stalls. Outcome-only rewards also provide no step-level credit assignment, limiting exploration and making it harder to learn robust reasoning. We present VeriGate (Verifier-Gated Step-Level GRPO), a verifier-gated extension of GRPO that addresses these limitations with three design choices. First, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.30451","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/2605.30451/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":"2605.30451","created_at":"2026-06-01T01:02:55.004754+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.30451v1","created_at":"2026-06-01T01:02:55.004754+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.30451","created_at":"2026-06-01T01:02:55.004754+00:00"},{"alias_kind":"pith_short_12","alias_value":"UKG6RHDKYB2R","created_at":"2026-06-01T01:02:55.004754+00:00"},{"alias_kind":"pith_short_16","alias_value":"UKG6RHDKYB2RW5DM","created_at":"2026-06-01T01:02:55.004754+00:00"},{"alias_kind":"pith_short_8","alias_value":"UKG6RHDK","created_at":"2026-06-01T01:02:55.004754+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/UKG6RHDKYB2RW5DM46JYUJJOJY","json":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY.json","graph_json":"https://pith.science/api/pith-number/UKG6RHDKYB2RW5DM46JYUJJOJY/graph.json","events_json":"https://pith.science/api/pith-number/UKG6RHDKYB2RW5DM46JYUJJOJY/events.json","paper":"https://pith.science/paper/UKG6RHDK"},"agent_actions":{"view_html":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY","download_json":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY.json","view_paper":"https://pith.science/paper/UKG6RHDK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.30451&json=true","fetch_graph":"https://pith.science/api/pith-number/UKG6RHDKYB2RW5DM46JYUJJOJY/graph.json","fetch_events":"https://pith.science/api/pith-number/UKG6RHDKYB2RW5DM46JYUJJOJY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY/action/storage_attestation","attest_author":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY/action/author_attestation","sign_citation":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY/action/citation_signature","submit_replication":"https://pith.science/pith/UKG6RHDKYB2RW5DM46JYUJJOJY/action/replication_record"}},"created_at":"2026-06-01T01:02:55.004754+00:00","updated_at":"2026-06-01T01:02:55.004754+00:00"}