{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WYRZ4OQH6GPYKCZD62O74PACQY","short_pith_number":"pith:WYRZ4OQH","schema_version":"1.0","canonical_sha256":"b6239e3a07f19f850b23f69dfe3c02862ac4e4169dc71e5f4c262ed35e0bcbf3","source":{"kind":"arxiv","id":"2605.26844","version":1},"attestation_state":"computed","paper":{"title":"Not All Disagreement Is Learnable: Token Teachability in On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Congkai Xie, Hongxia Yang, Jianmin Wu, Pengkai Wang, Su Lu, Yanggan Gu, Yifan Yang, Yuanyi Wang, Zhaoyi Yan","submitted_at":"2026-05-26T10:56:46Z","abstract_excerpt":"On-policy distillation (OPD) trains a student on its own rollouts with token-level teacher supervision. Recent selective OPD methods exploit the non-uniformity of OPD signals by prioritizing high-entropy or high-disagreement tokens. We revisit this principle and ask: which token-level teacher signals are actually learnable? Using a fixed-context diagnostic that measures same-context teacher-student KL reduction, we show that raw KL disagreement is a coarse proxy for learning value. It conflates learnable disagreement, where the teacher assigns corrective mass to the student's top-K candidates,"},"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":"2605.26844","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-26T10:56:46Z","cross_cats_sorted":[],"title_canon_sha256":"ea935a34a1f520ee0112c90cd00fd8d6d6716d3fb702383d677f97feeb0d675d","abstract_canon_sha256":"7857fac283e5d7f293951d61ff916d5b41367a4247fb0b6c04d3a287e71551f6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:06:15.359769Z","signature_b64":"WHVo31OIcxQMRI+o7l1RSpnQhcplsty5Hqg882BZ5E0Gd16/eN3fbpCrVdu3zXRnOpc2RIBnMWngvfpuNbHTCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b6239e3a07f19f850b23f69dfe3c02862ac4e4169dc71e5f4c262ed35e0bcbf3","last_reissued_at":"2026-05-27T01:06:15.359067Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:06:15.359067Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Not All Disagreement Is Learnable: Token Teachability in On-Policy Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Congkai Xie, Hongxia Yang, Jianmin Wu, Pengkai Wang, Su Lu, Yanggan Gu, Yifan Yang, Yuanyi Wang, Zhaoyi Yan","submitted_at":"2026-05-26T10:56:46Z","abstract_excerpt":"On-policy distillation (OPD) trains a student on its own rollouts with token-level teacher supervision. Recent selective OPD methods exploit the non-uniformity of OPD signals by prioritizing high-entropy or high-disagreement tokens. We revisit this principle and ask: which token-level teacher signals are actually learnable? Using a fixed-context diagnostic that measures same-context teacher-student KL reduction, we show that raw KL disagreement is a coarse proxy for learning value. It conflates learnable disagreement, where the teacher assigns corrective mass to the student's top-K candidates,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26844","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/2605.26844/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":"2605.26844","created_at":"2026-05-27T01:06:15.359171+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26844v1","created_at":"2026-05-27T01:06:15.359171+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26844","created_at":"2026-05-27T01:06:15.359171+00:00"},{"alias_kind":"pith_short_12","alias_value":"WYRZ4OQH6GPY","created_at":"2026-05-27T01:06:15.359171+00:00"},{"alias_kind":"pith_short_16","alias_value":"WYRZ4OQH6GPYKCZD","created_at":"2026-05-27T01:06:15.359171+00:00"},{"alias_kind":"pith_short_8","alias_value":"WYRZ4OQH","created_at":"2026-05-27T01:06:15.359171+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/WYRZ4OQH6GPYKCZD62O74PACQY","json":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY.json","graph_json":"https://pith.science/api/pith-number/WYRZ4OQH6GPYKCZD62O74PACQY/graph.json","events_json":"https://pith.science/api/pith-number/WYRZ4OQH6GPYKCZD62O74PACQY/events.json","paper":"https://pith.science/paper/WYRZ4OQH"},"agent_actions":{"view_html":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY","download_json":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY.json","view_paper":"https://pith.science/paper/WYRZ4OQH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26844&json=true","fetch_graph":"https://pith.science/api/pith-number/WYRZ4OQH6GPYKCZD62O74PACQY/graph.json","fetch_events":"https://pith.science/api/pith-number/WYRZ4OQH6GPYKCZD62O74PACQY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY/action/storage_attestation","attest_author":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY/action/author_attestation","sign_citation":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY/action/citation_signature","submit_replication":"https://pith.science/pith/WYRZ4OQH6GPYKCZD62O74PACQY/action/replication_record"}},"created_at":"2026-05-27T01:06:15.359171+00:00","updated_at":"2026-05-27T01:06:15.359171+00:00"}