{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XXKSVFJ2J7UCFHST2TXIEVX7YA","short_pith_number":"pith:XXKSVFJ2","schema_version":"1.0","canonical_sha256":"bdd52a953a4fe8229e53d4ee8256ffc034fc9912761564d257b34266626bd091","source":{"kind":"arxiv","id":"2605.24202","version":1},"attestation_state":"computed","paper":{"title":"When Does Multi-Agent RL Improve LLM Workflows? Workflow, Scale, and Policy-Sharing Tradeoffs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Huazheng Wang, Kun Wan, Qingyun Wu, Wentian Zhao, Yaolun Zhang, Yifan Zeng, Yiran Wu","submitted_at":"2026-05-22T20:43:30Z","abstract_excerpt":"Multi-agent LLM workflows route inference through specialized roles to lift end-task accuracy, but jointly training those roles with reinforcement learning is unstable in ways that are poorly understood. We study when end-to-end RL training of multi-agent LLM workflows improves over their base models, comparing Shared-Policy training, where all roles update one policy, with Isolated-Policy training, where each role has its own parameters. Our experimental matrix spans Eval-Opt, Voting, and Orch-Workers workflows, math and code tasks, and three model scales (0.6B, 1.7B, 4B). We find that multi-"},"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":"2605.24202","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-22T20:43:30Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"86c05f403a05cb6001c3dc555d0c585830b0a1be68f8f9e2adcce882935f269c","abstract_canon_sha256":"a4e598bf07f2dbb30cc185555e60976e151d497eb49acca96cabc8b20ce8432e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:02:52.201996Z","signature_b64":"Khwb9QPYUuaFuUGPm03TCMma+L8VTOdEXmAP7FBbxUfWf8NJK8Of6JeFBVAKjASx8nuiAtOQStoln5S9rzJeCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bdd52a953a4fe8229e53d4ee8256ffc034fc9912761564d257b34266626bd091","last_reissued_at":"2026-05-26T01:02:52.201166Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:02:52.201166Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Does Multi-Agent RL Improve LLM Workflows? Workflow, Scale, and Policy-Sharing Tradeoffs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Huazheng Wang, Kun Wan, Qingyun Wu, Wentian Zhao, Yaolun Zhang, Yifan Zeng, Yiran Wu","submitted_at":"2026-05-22T20:43:30Z","abstract_excerpt":"Multi-agent LLM workflows route inference through specialized roles to lift end-task accuracy, but jointly training those roles with reinforcement learning is unstable in ways that are poorly understood. We study when end-to-end RL training of multi-agent LLM workflows improves over their base models, comparing Shared-Policy training, where all roles update one policy, with Isolated-Policy training, where each role has its own parameters. Our experimental matrix spans Eval-Opt, Voting, and Orch-Workers workflows, math and code tasks, and three model scales (0.6B, 1.7B, 4B). We find that multi-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.24202","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/2605.24202/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":"2605.24202","created_at":"2026-05-26T01:02:52.201307+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.24202v1","created_at":"2026-05-26T01:02:52.201307+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.24202","created_at":"2026-05-26T01:02:52.201307+00:00"},{"alias_kind":"pith_short_12","alias_value":"XXKSVFJ2J7UC","created_at":"2026-05-26T01:02:52.201307+00:00"},{"alias_kind":"pith_short_16","alias_value":"XXKSVFJ2J7UCFHST","created_at":"2026-05-26T01:02:52.201307+00:00"},{"alias_kind":"pith_short_8","alias_value":"XXKSVFJ2","created_at":"2026-05-26T01:02:52.201307+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/XXKSVFJ2J7UCFHST2TXIEVX7YA","json":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA.json","graph_json":"https://pith.science/api/pith-number/XXKSVFJ2J7UCFHST2TXIEVX7YA/graph.json","events_json":"https://pith.science/api/pith-number/XXKSVFJ2J7UCFHST2TXIEVX7YA/events.json","paper":"https://pith.science/paper/XXKSVFJ2"},"agent_actions":{"view_html":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA","download_json":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA.json","view_paper":"https://pith.science/paper/XXKSVFJ2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.24202&json=true","fetch_graph":"https://pith.science/api/pith-number/XXKSVFJ2J7UCFHST2TXIEVX7YA/graph.json","fetch_events":"https://pith.science/api/pith-number/XXKSVFJ2J7UCFHST2TXIEVX7YA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA/action/storage_attestation","attest_author":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA/action/author_attestation","sign_citation":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA/action/citation_signature","submit_replication":"https://pith.science/pith/XXKSVFJ2J7UCFHST2TXIEVX7YA/action/replication_record"}},"created_at":"2026-05-26T01:02:52.201307+00:00","updated_at":"2026-05-26T01:02:52.201307+00:00"}