{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:IBATJDLOU3LETEJXTQF67NBOJN","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"6e600b008c47516456ddd1fa78f0bb9609bf7056c0cd27b6bb77a80754a25007","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-08T14:38:53Z","title_canon_sha256":"1ddfef6338d71579cc32c9d572291023565d114402883f18cd7f0ae117a97825"},"schema_version":"1.0","source":{"id":"2605.07804","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.07804","created_at":"2026-05-29T02:05:46Z"},{"alias_kind":"arxiv_version","alias_value":"2605.07804v2","created_at":"2026-05-29T02:05:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.07804","created_at":"2026-05-29T02:05:46Z"},{"alias_kind":"pith_short_12","alias_value":"IBATJDLOU3LE","created_at":"2026-05-29T02:05:46Z"},{"alias_kind":"pith_short_16","alias_value":"IBATJDLOU3LETEJX","created_at":"2026-05-29T02:05:46Z"},{"alias_kind":"pith_short_8","alias_value":"IBATJDLO","created_at":"2026-05-29T02:05:46Z"}],"graph_snapshots":[{"event_id":"sha256:4b983366bd81be1c73789c379d9e414b28fa032e5dfc4bb4fbd6a1e419c6a424","target":"graph","created_at":"2026-05-29T02:05:46Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Prune-OPD reduces training time by 37.6%--68.0% while preserving, and often improving, performance on challenging benchmarks (AMC, AIME, HMMT) by dynamically aligning computation with supervision reliability across diverse teacher-student combinations."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That top-k overlap between student and teacher predictions is a reliable real-time indicator of when dense teacher rewards lose local exploitability, and that monotonic down-weighting plus truncation does not discard critical learning signals needed for long-horizon improvement."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Prune-OPD dynamically prunes unreliable teacher rewards in on-policy distillation by monitoring prefix drift via top-k overlap, reducing training time 37.6-68% on AMC/AIME/HMMT while preserving or improving performance."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Prune-OPD makes on-policy distillation for long-horizon reasoning more efficient by pruning unreliable teacher rewards in real time."}],"snapshot_sha256":"47398ec2ba48b8a26806dd2bbf5c591a01be0727c8a7a91bc7843afc7f74178d"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"417f5c0c6787c8438a67801a88bebd7d79258251d92a7fd3d4856ae90d9331d2"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T10:22:02.734357Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T04:50:05.693007Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T15:31:18.842395Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T11:32:17.673878Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.07804/integrity.json","findings":[],"snapshot_sha256":"3a3d2a13ab9d23d2f5a72728efba70f7d75dfeb91beb8e649f3c0fc591e500e5","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"On-policy distillation (OPD) leverages dense teacher rewards to enhance reasoning models. However, scaling OPD to long-horizon tasks exposes a critical flaw: as the student's generated prefix inevitably diverges from the teacher's thought process, the teacher's dense reward loses local exploitability. Continuing to generate and evaluate tokens on these ``drifted'' trajectories not only degrades reward quality but also incurs massive computational waste. To address this, we introduce \\textbf{Prune-OPD}, a framework that dynamically aligns training budgets with supervision quality. By continuous","authors_text":"Jing Tang, Minrui Xu, Xiaodan Liang, Yifan Song, Yiwei Wang, Yongxin Wang, Zhicheng Yang, Zhijiang Guo","cross_cats":["cs.AI"],"headline":"Prune-OPD makes on-policy distillation for long-horizon reasoning more efficient by pruning unreliable teacher rewards in real time.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-08T14:38:53Z","title":"Prune-OPD: Efficient and Reliable On-Policy Distillation for Long-Horizon Reasoning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.07804","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-11T03:15:11.248353Z","id":"ee240024-508d-47e7-a45d-187ed1cfa5ba","model_set":{"reader":"grok-4.3"},"one_line_summary":"Prune-OPD dynamically prunes unreliable teacher rewards in on-policy distillation by monitoring prefix drift via top-k overlap, reducing training time 37.6-68% on AMC/AIME/HMMT while preserving or improving performance.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Prune-OPD makes on-policy distillation for long-horizon reasoning more efficient by pruning unreliable teacher rewards in real time.","strongest_claim":"Prune-OPD reduces training time by 37.6%--68.0% while preserving, and often improving, performance on challenging benchmarks (AMC, AIME, HMMT) by dynamically aligning computation with supervision reliability across diverse teacher-student combinations.","weakest_assumption":"That top-k overlap between student and teacher predictions is a reliable real-time indicator of when dense teacher rewards lose local exploitability, and that monotonic down-weighting plus truncation does not discard critical learning signals needed for long-horizon improvement."}},"verdict_id":"ee240024-508d-47e7-a45d-187ed1cfa5ba"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:60050daf6895f909bd1e2f42df7075f7254016db34627682ec27b71b5a4462ab","target":"record","created_at":"2026-05-29T02:05:46Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"6e600b008c47516456ddd1fa78f0bb9609bf7056c0cd27b6bb77a80754a25007","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-08T14:38:53Z","title_canon_sha256":"1ddfef6338d71579cc32c9d572291023565d114402883f18cd7f0ae117a97825"},"schema_version":"1.0","source":{"id":"2605.07804","kind":"arxiv","version":2}},"canonical_sha256":"4041348d6ea6d64991379c0befb42e4b55be755766871ddb865528ac4527a08d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4041348d6ea6d64991379c0befb42e4b55be755766871ddb865528ac4527a08d","first_computed_at":"2026-05-29T02:05:46.258958Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-29T02:05:46.258958Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"t5wBwYhB6tV4kwQZ8etzSIx2qVGyOonUmZUOMKCCg20Ca2qyzTRbgfsgyWFp/xaq6iGP9Zu54CBk9xho8FsVBQ==","signature_status":"signed_v1","signed_at":"2026-05-29T02:05:46.259561Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.07804","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:60050daf6895f909bd1e2f42df7075f7254016db34627682ec27b71b5a4462ab","sha256:4b983366bd81be1c73789c379d9e414b28fa032e5dfc4bb4fbd6a1e419c6a424"],"state_sha256":"172061ccff88b94c687de90cb867b959fe2701066e0b0d9373e96ed58ba01f7e"}