{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:J276MSRKZYZD4E7PELII3T7AEU","short_pith_number":"pith:J276MSRK","schema_version":"1.0","canonical_sha256":"4ebfe64a2ace323e13ef22d08dcfe0253f2625ccb9655806abfe5d41eedbda61","source":{"kind":"arxiv","id":"2605.15204","version":1},"attestation_state":"computed","paper":{"title":"SDOF: Taming the Alignment Tax in Multi-Agent Orchestration with State-Constrained Dispatch","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"SDOF models multi-agent orchestration as a constrained state machine to let a 7B router beat zero-shot GPT-4o on adversarial routing while blocking all illegal operations.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Zhantao Wang","submitted_at":"2026-04-20T12:51:39Z","abstract_excerpt":"Multi-agent orchestration frameworks such as LangChain, LangGraph, and CrewAI route tasks through graph-based pipelines but do not enforce the stage constraints that govern real business processes. We present SDOF, a framework that treats multi-agent execution as a constrained state machine. SDOF operates through two primary defensive layers, implemented by three components: (1) an Online-RLHF Specialized Intent Router trained via Generative Reward Modeling (GRPO) and (2) a StateAwareDispatcher with GoalStage finite-automaton checks and precondition/postcondition SkillRegistry validation for a"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15204","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.AI","submitted_at":"2026-04-20T12:51:39Z","cross_cats_sorted":[],"title_canon_sha256":"2682461e28eb2548ba4359aad730a8d194755abb6fe559087bb241894b3a8491","abstract_canon_sha256":"98f1b1c82e3214113f2157c03416f5a2de529c2313ac5a3a7cfab68915a87ca1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:45.901271Z","signature_b64":"cNzsizSoI0+181FU04xL19C8RDMNeWGkr+Mlfic3PLYrMuHPjgJPHSKkLv+aRKq6OSdQfQBdV8gj9VKc0tmnCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4ebfe64a2ace323e13ef22d08dcfe0253f2625ccb9655806abfe5d41eedbda61","last_reissued_at":"2026-05-20T00:00:45.900309Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:45.900309Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SDOF: Taming the Alignment Tax in Multi-Agent Orchestration with State-Constrained Dispatch","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"SDOF models multi-agent orchestration as a constrained state machine to let a 7B router beat zero-shot GPT-4o on adversarial routing while blocking all illegal operations.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Zhantao Wang","submitted_at":"2026-04-20T12:51:39Z","abstract_excerpt":"Multi-agent orchestration frameworks such as LangChain, LangGraph, and CrewAI route tasks through graph-based pipelines but do not enforce the stage constraints that govern real business processes. We present SDOF, a framework that treats multi-agent execution as a constrained state machine. SDOF operates through two primary defensive layers, implemented by three components: (1) an Online-RLHF Specialized Intent Router trained via Generative Reward Modeling (GRPO) and (2) a StateAwareDispatcher with GoalStage finite-automaton checks and precondition/postcondition SkillRegistry validation for a"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our GSPO-aligned 7B Intent Router achieves higher joint accuracy than zero-shot GPT-4o on this FSM-constrained adversarial routing benchmark (80.9% versus 48.9%). In end-to-end execution, SDOF reaches 86.5% task completion and blocks all 22 operations in the injection, illegal HR subset.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The expert-curated 185 scenarios and the Beisen iTalent platform data accurately represent general multi-agent orchestration challenges and that the GoalStage finite-automaton mapping faithfully captures real business process constraints without missing edge cases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SDOF combines an RLHF-trained intent router with a state-aware dispatcher using finite automata to constrain multi-agent orchestration, reporting 80.9% routing accuracy and 86.5% task completion on a recruitment platform while blocking unsafe actions.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SDOF models multi-agent orchestration as a constrained state machine to let a 7B router beat zero-shot GPT-4o on adversarial routing while blocking all illegal operations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6f16b5d1760ea970fad60e1592e8eec4e8c7ca5c6a4ac62a99657f2f18d84bf0"},"source":{"id":"2605.15204","kind":"arxiv","version":1},"verdict":{"id":"f53d7ca7-77c7-4b67-bec4-967edc16a39e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T18:01:50.686244Z","strongest_claim":"Our GSPO-aligned 7B Intent Router achieves higher joint accuracy than zero-shot GPT-4o on this FSM-constrained adversarial routing benchmark (80.9% versus 48.9%). In end-to-end execution, SDOF reaches 86.5% task completion and blocks all 22 operations in the injection, illegal HR subset.","one_line_summary":"SDOF combines an RLHF-trained intent router with a state-aware dispatcher using finite automata to constrain multi-agent orchestration, reporting 80.9% routing accuracy and 86.5% task completion on a recruitment platform while blocking unsafe actions.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The expert-curated 185 scenarios and the Beisen iTalent platform data accurately represent general multi-agent orchestration challenges and that the GoalStage finite-automaton mapping faithfully captures real business process constraints without missing edge cases.","pith_extraction_headline":"SDOF models multi-agent orchestration as a constrained state machine to let a 7B router beat zero-shot GPT-4o on adversarial routing while blocking all illegal operations."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15204/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":25,"sample":[{"doi":"","year":2025,"title":"AgentAuditor: Safety and security evaluation for large language model agents","work_id":"e527547e-ca92-4fe0-84cb-83fa78cac2f3","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Langchain: Building applications with LLMs through composability.https://github","work_id":"542b2884-59d3-4552-80b0-d5d52720aa7c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"AgentVerse: Facilitating multi-agent collaboration and exploring emergent be- haviors","work_id":"a1aba051-38f4-4c13-a2d0-a2adf31bdf01","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Cooperative AI: machines must learn to find common ground","work_id":"fed3351c-df58-41d9-ac89-7b6480095e3b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Agentscope: A flexible yet robust multi-agent platform","work_id":"41f8fc1e-1e31-4adb-81fa-414f73b79d41","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":25,"snapshot_sha256":"edbdc84b7204a8e01f1134a9bc4c0af25b977323831bbec05cf8fb30f5bfaec1","internal_anchors":4},"formal_canon":{"evidence_count":2,"snapshot_sha256":"b759ef96b48494f90fc79883a1a71f2604110f2f4f552ba54e2056079b36fbc2"},"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":"2605.15204","created_at":"2026-05-20T00:00:45.900486+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15204v1","created_at":"2026-05-20T00:00:45.900486+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15204","created_at":"2026-05-20T00:00:45.900486+00:00"},{"alias_kind":"pith_short_12","alias_value":"J276MSRKZYZD","created_at":"2026-05-20T00:00:45.900486+00:00"},{"alias_kind":"pith_short_16","alias_value":"J276MSRKZYZD4E7P","created_at":"2026-05-20T00:00:45.900486+00:00"},{"alias_kind":"pith_short_8","alias_value":"J276MSRK","created_at":"2026-05-20T00:00:45.900486+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU","json":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU.json","graph_json":"https://pith.science/api/pith-number/J276MSRKZYZD4E7PELII3T7AEU/graph.json","events_json":"https://pith.science/api/pith-number/J276MSRKZYZD4E7PELII3T7AEU/events.json","paper":"https://pith.science/paper/J276MSRK"},"agent_actions":{"view_html":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU","download_json":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU.json","view_paper":"https://pith.science/paper/J276MSRK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15204&json=true","fetch_graph":"https://pith.science/api/pith-number/J276MSRKZYZD4E7PELII3T7AEU/graph.json","fetch_events":"https://pith.science/api/pith-number/J276MSRKZYZD4E7PELII3T7AEU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU/action/storage_attestation","attest_author":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU/action/author_attestation","sign_citation":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU/action/citation_signature","submit_replication":"https://pith.science/pith/J276MSRKZYZD4E7PELII3T7AEU/action/replication_record"}},"created_at":"2026-05-20T00:00:45.900486+00:00","updated_at":"2026-05-20T00:00:45.900486+00:00"}