{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RU2E63QX7KG7VLT337R46BYZIR","short_pith_number":"pith:RU2E63QX","schema_version":"1.0","canonical_sha256":"8d344f6e17fa8dfaae7bdfe3cf071944680cb02565fa1b3cbf19c638fa88d1ad","source":{"kind":"arxiv","id":"2512.01970","version":3},"attestation_state":"computed","paper":{"title":"Atomic Skills are the Prerequisite: When Reinforcement Learning Synthesizes Compositional Reasoning, and When It Only Amplifies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Liangming Pan, Ruiwen Zhou, Sitao Cheng, Victor Zhong, William Yang Wang, Xinyi Wang, Xunjian Yin, Yuxuan Li","submitted_at":"2025-12-01T18:27:25Z","abstract_excerpt":"Does Reinforcement Learning (RL) merely amplify existing skills, or synthesize novel skills? We investigate this question through the lens of Complementary Reasoning: the critical practical capability of integrating internal knowledge with external context, a prerequisite for reliable Continual Learning and Retrieval-Augmented Generation. To avoid pre-training contamination, we construct a controlled semanticsynthetic dataset of biographies and decompose this capability into two atomic skills: Parametric Reasoning (retrieving facts encoded in model weights) and Contextual Reasoning (processing"},"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":"2512.01970","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-12-01T18:27:25Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"15d1a0a1124fe0df8e9264b0e57efa1cccd47ee05f97c6b1fe366e3729bd3869","abstract_canon_sha256":"024917eb33086dfea99facb88def3d105718dd00ac8a65a039dd962a56afd5e4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:33.790410Z","signature_b64":"P4BxA1fQ28ANTRszRiFTKIyUToswOeE0V+eKCsLihK5D84N4TBrvLPEAH176Ci0FMmVu/5N6d/ueNSMlNtPjBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8d344f6e17fa8dfaae7bdfe3cf071944680cb02565fa1b3cbf19c638fa88d1ad","last_reissued_at":"2026-05-28T01:04:33.789866Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:33.789866Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Atomic Skills are the Prerequisite: When Reinforcement Learning Synthesizes Compositional Reasoning, and When It Only Amplifies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Liangming Pan, Ruiwen Zhou, Sitao Cheng, Victor Zhong, William Yang Wang, Xinyi Wang, Xunjian Yin, Yuxuan Li","submitted_at":"2025-12-01T18:27:25Z","abstract_excerpt":"Does Reinforcement Learning (RL) merely amplify existing skills, or synthesize novel skills? We investigate this question through the lens of Complementary Reasoning: the critical practical capability of integrating internal knowledge with external context, a prerequisite for reliable Continual Learning and Retrieval-Augmented Generation. To avoid pre-training contamination, we construct a controlled semanticsynthetic dataset of biographies and decompose this capability into two atomic skills: Parametric Reasoning (retrieving facts encoded in model weights) and Contextual Reasoning (processing"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.01970","kind":"arxiv","version":3},"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/2512.01970/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":"2512.01970","created_at":"2026-05-28T01:04:33.789937+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.01970v3","created_at":"2026-05-28T01:04:33.789937+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.01970","created_at":"2026-05-28T01:04:33.789937+00:00"},{"alias_kind":"pith_short_12","alias_value":"RU2E63QX7KG7","created_at":"2026-05-28T01:04:33.789937+00:00"},{"alias_kind":"pith_short_16","alias_value":"RU2E63QX7KG7VLT3","created_at":"2026-05-28T01:04:33.789937+00:00"},{"alias_kind":"pith_short_8","alias_value":"RU2E63QX","created_at":"2026-05-28T01:04:33.789937+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.08472","citing_title":"Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models","ref_index":5,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR","json":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR.json","graph_json":"https://pith.science/api/pith-number/RU2E63QX7KG7VLT337R46BYZIR/graph.json","events_json":"https://pith.science/api/pith-number/RU2E63QX7KG7VLT337R46BYZIR/events.json","paper":"https://pith.science/paper/RU2E63QX"},"agent_actions":{"view_html":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR","download_json":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR.json","view_paper":"https://pith.science/paper/RU2E63QX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.01970&json=true","fetch_graph":"https://pith.science/api/pith-number/RU2E63QX7KG7VLT337R46BYZIR/graph.json","fetch_events":"https://pith.science/api/pith-number/RU2E63QX7KG7VLT337R46BYZIR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR/action/storage_attestation","attest_author":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR/action/author_attestation","sign_citation":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR/action/citation_signature","submit_replication":"https://pith.science/pith/RU2E63QX7KG7VLT337R46BYZIR/action/replication_record"}},"created_at":"2026-05-28T01:04:33.789937+00:00","updated_at":"2026-05-28T01:04:33.789937+00:00"}