{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IPG6I7ICRXQRSF3XO263NNEGSQ","short_pith_number":"pith:IPG6I7IC","schema_version":"1.0","canonical_sha256":"43cde47d028de119177776bdb6b4869433bceaab8984b39c03a3b169b2814203","source":{"kind":"arxiv","id":"2605.18529","version":1},"attestation_state":"computed","paper":{"title":"AMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit Assignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Guojun Yin, Jiajun Chai, Pu Jian, Shanbin Zhang, Wei Lin, Xiaohan Wang, Yingzhuo Deng, Zhenlin Wei, Zhexin Hu","submitted_at":"2026-05-18T15:14:34Z","abstract_excerpt":"The alignment of Large Language Models (LLMs) for complex reasoning heavily relies on Reinforcement Learning with Verifiable Rewards (RLVR). However, standard algorithms like GRPO apply sequence-level rewards uniformly to all tokens, creating a severe credit-assignment bottleneck. While on-policy self-distillation attempts to resolve this by conditioning a self-teacher on privileged contexts, direct exposure to raw oracle solutions often induces over-conditioned teacher distributions, implicit answer leakage, and late-stage training collapse. To overcome these limitations, we propose Asymmetri"},"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":"2605.18529","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-18T15:14:34Z","cross_cats_sorted":[],"title_canon_sha256":"d8d55af960275e500181025b473147c484150be13f266859ae12025821f8a399","abstract_canon_sha256":"d44c8404f0d1d0c5a2fd0e71fa00cc2094db6a769e2d748d49ebea63c10432ea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:06:06.123345Z","signature_b64":"wjGA/QsBmvpvLSuRgYrd3oqyJt+/IDCoiBROrsKxJSr6sWyDHw8qqZ9wS2tKJNPXs7VQItyNl14XvyEhp+VzBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43cde47d028de119177776bdb6b4869433bceaab8984b39c03a3b169b2814203","last_reissued_at":"2026-05-20T00:06:06.122530Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:06:06.122530Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AMR-SD: Asymmetric Meta-Reflective Self-Distillation for Token-Level Credit Assignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Guojun Yin, Jiajun Chai, Pu Jian, Shanbin Zhang, Wei Lin, Xiaohan Wang, Yingzhuo Deng, Zhenlin Wei, Zhexin Hu","submitted_at":"2026-05-18T15:14:34Z","abstract_excerpt":"The alignment of Large Language Models (LLMs) for complex reasoning heavily relies on Reinforcement Learning with Verifiable Rewards (RLVR). However, standard algorithms like GRPO apply sequence-level rewards uniformly to all tokens, creating a severe credit-assignment bottleneck. While on-policy self-distillation attempts to resolve this by conditioning a self-teacher on privileged contexts, direct exposure to raw oracle solutions often induces over-conditioned teacher distributions, implicit answer leakage, and late-stage training collapse. To overcome these limitations, we propose Asymmetri"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.18529","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/2605.18529/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":"2605.18529","created_at":"2026-05-20T00:06:06.122666+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.18529v1","created_at":"2026-05-20T00:06:06.122666+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.18529","created_at":"2026-05-20T00:06:06.122666+00:00"},{"alias_kind":"pith_short_12","alias_value":"IPG6I7ICRXQR","created_at":"2026-05-20T00:06:06.122666+00:00"},{"alias_kind":"pith_short_16","alias_value":"IPG6I7ICRXQRSF3X","created_at":"2026-05-20T00:06:06.122666+00:00"},{"alias_kind":"pith_short_8","alias_value":"IPG6I7IC","created_at":"2026-05-20T00:06:06.122666+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/IPG6I7ICRXQRSF3XO263NNEGSQ","json":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ.json","graph_json":"https://pith.science/api/pith-number/IPG6I7ICRXQRSF3XO263NNEGSQ/graph.json","events_json":"https://pith.science/api/pith-number/IPG6I7ICRXQRSF3XO263NNEGSQ/events.json","paper":"https://pith.science/paper/IPG6I7IC"},"agent_actions":{"view_html":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ","download_json":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ.json","view_paper":"https://pith.science/paper/IPG6I7IC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.18529&json=true","fetch_graph":"https://pith.science/api/pith-number/IPG6I7ICRXQRSF3XO263NNEGSQ/graph.json","fetch_events":"https://pith.science/api/pith-number/IPG6I7ICRXQRSF3XO263NNEGSQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ/action/storage_attestation","attest_author":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ/action/author_attestation","sign_citation":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ/action/citation_signature","submit_replication":"https://pith.science/pith/IPG6I7ICRXQRSF3XO263NNEGSQ/action/replication_record"}},"created_at":"2026-05-20T00:06:06.122666+00:00","updated_at":"2026-05-20T00:06:06.122666+00:00"}