{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7QRCYGXDBE3TVICUWSICZUL4CM","short_pith_number":"pith:7QRCYGXD","schema_version":"1.0","canonical_sha256":"fc222c1ae309373aa054b4902cd17c130f5b323035be9bdae8a70c39ada32a66","source":{"kind":"arxiv","id":"2606.09304","version":1},"attestation_state":"computed","paper":{"title":"SG-OPD: Sign-Gated On-Policy Distillation via Sign-Consistency Gating and Phased Teacher Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Haoran Xu, Hongyu Wang, Jiaze Li, Xiaofeng Zhang, Xiaosong Yuan, Yifei Gao","submitted_at":"2026-06-08T10:11:58Z","abstract_excerpt":"On-policy distillation (OPD) trains a student on its own trajectories with dense per-token supervision from a stronger teacher, and often outperforms off-policy distillation and standard reinforcement learning. However, we find that its effectiveness implicitly relies on two assumptions that frequently break in practice: trajectory-level alignment between the student and the teacher, and uniform token-level reliability of the teacher's preferences. We therefore propose Sign-Gated On-Policy Distillation (SG-OPD), which uses a binary verifier as a trust signal for the teacher at two complementar"},"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":"2606.09304","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-06-08T10:11:58Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"94afbce500e64b94f43cdab39ed016edd224928f53e511e7062636a342be92f9","abstract_canon_sha256":"c1db695b5ebb5abad0411c8d941e60cc3800329d543a0b32e194205d0eff0541"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T02:08:13.798366Z","signature_b64":"jPIGeWNr933KsIDg4vTWt8rQ9pjDcXlkLtFdKM8il4lHax6xnAn334+P+gq6MbeVtEvNtDg9JRCSF3xgnfOaAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fc222c1ae309373aa054b4902cd17c130f5b323035be9bdae8a70c39ada32a66","last_reissued_at":"2026-06-09T02:08:13.797549Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T02:08:13.797549Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SG-OPD: Sign-Gated On-Policy Distillation via Sign-Consistency Gating and Phased Teacher Sampling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Haoran Xu, Hongyu Wang, Jiaze Li, Xiaofeng Zhang, Xiaosong Yuan, Yifei Gao","submitted_at":"2026-06-08T10:11:58Z","abstract_excerpt":"On-policy distillation (OPD) trains a student on its own trajectories with dense per-token supervision from a stronger teacher, and often outperforms off-policy distillation and standard reinforcement learning. However, we find that its effectiveness implicitly relies on two assumptions that frequently break in practice: trajectory-level alignment between the student and the teacher, and uniform token-level reliability of the teacher's preferences. We therefore propose Sign-Gated On-Policy Distillation (SG-OPD), which uses a binary verifier as a trust signal for the teacher at two complementar"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.09304","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/2606.09304/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":"2606.09304","created_at":"2026-06-09T02:08:13.797683+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.09304v1","created_at":"2026-06-09T02:08:13.797683+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.09304","created_at":"2026-06-09T02:08:13.797683+00:00"},{"alias_kind":"pith_short_12","alias_value":"7QRCYGXDBE3T","created_at":"2026-06-09T02:08:13.797683+00:00"},{"alias_kind":"pith_short_16","alias_value":"7QRCYGXDBE3TVICU","created_at":"2026-06-09T02:08:13.797683+00:00"},{"alias_kind":"pith_short_8","alias_value":"7QRCYGXD","created_at":"2026-06-09T02:08:13.797683+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/7QRCYGXDBE3TVICUWSICZUL4CM","json":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM.json","graph_json":"https://pith.science/api/pith-number/7QRCYGXDBE3TVICUWSICZUL4CM/graph.json","events_json":"https://pith.science/api/pith-number/7QRCYGXDBE3TVICUWSICZUL4CM/events.json","paper":"https://pith.science/paper/7QRCYGXD"},"agent_actions":{"view_html":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM","download_json":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM.json","view_paper":"https://pith.science/paper/7QRCYGXD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.09304&json=true","fetch_graph":"https://pith.science/api/pith-number/7QRCYGXDBE3TVICUWSICZUL4CM/graph.json","fetch_events":"https://pith.science/api/pith-number/7QRCYGXDBE3TVICUWSICZUL4CM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM/action/storage_attestation","attest_author":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM/action/author_attestation","sign_citation":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM/action/citation_signature","submit_replication":"https://pith.science/pith/7QRCYGXDBE3TVICUWSICZUL4CM/action/replication_record"}},"created_at":"2026-06-09T02:08:13.797683+00:00","updated_at":"2026-06-09T02:08:13.797683+00:00"}