{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:MHY4IJOM5AWQZDUXMEKYT2CVDX","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"52e8853a78ec8c87734ac98a6b25bb71dc31ecc07f94480c1c465ac50bb3f52c","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T21:27:21Z","title_canon_sha256":"567d44002f2e6213549591e3fea47891316a6c7be41cba3fa01d3f383819bb43"},"schema_version":"1.0","source":{"id":"2605.14126","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14126","created_at":"2026-05-17T23:39:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14126v1","created_at":"2026-05-17T23:39:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14126","created_at":"2026-05-17T23:39:11Z"},{"alias_kind":"pith_short_12","alias_value":"MHY4IJOM5AWQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"MHY4IJOM5AWQZDUX","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"MHY4IJOM","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:5f642abdb4260a358560bbda2dca7b3265fbf16f58e4d67bb835ae45c9f1cda2","target":"graph","created_at":"2026-05-17T23:39:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Empirically, our approach improves answer correctness from 50% (o4-mini) to 77% on FHIR-AgentBench using a smaller and cheaper Qwen3-8B model."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That an LLM Judge can supply reliable execution-grounded rewards without bias or systematic errors in judging multi-step FHIR traversals."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"RL post-training lifts answer correctness on FHIR-AgentBench from 50% (o4-mini) to 77% with a cheaper Qwen3-8B CodeAct agent."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Reinforcement learning post-training raises a Qwen3-8B model to 77 percent correctness on FHIR clinical queries, surpassing larger closed models."}],"snapshot_sha256":"2095daef2331626e1d156c9ec4667e7e43672ac467fb9794f2adbff6b352a9c9"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Fast Healthcare Interoperability Resources (FHIR) is the dominant standard for interoperable exchange of healthcare data. In FHIR, electronic health records form a directed graph of resources. Answering clinically meaningful questions over FHIR requires agents to perform multi-step reasoning, filtering, and aggregation across multiple resource types. Prior work shows that even tool-augmented LLM agents (retrieval, code execution, multi-turn planning) often select the wrong resources or violate traversal constraints. We study this problem in the context of FHIR-AgentBench, a benchmark for reali","authors_text":"Jan P. Bremer, Marius S. Knorr, Nils Schweingruber, Robert M\\\"uller","cross_cats":["cs.AI"],"headline":"Reinforcement learning post-training raises a Qwen3-8B model to 77 percent correctness on FHIR clinical queries, surpassing larger closed models.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T21:27:21Z","title":"Reinforcement Learning for Tool-Calling Agents in Fast Healthcare Interoperability Resources (FHIR)"},"references":{"count":66,"internal_anchors":8,"resolved_work":66,"sample":[{"cited_arxiv_id":"","doi":"10.1093/jamia/ocx080","is_internal_anchor":false,"ref_index":1,"title":"Journal of the American Medical Informatics Association , author =","work_id":"c24070b8-a397-4077-97ed-5d4542edbc17","year":2017},{"cited_arxiv_id":"","doi":"10.5009/gnl230272","is_internal_anchor":false,"ref_index":2,"title":"Gut and Liver , author =","work_id":"eabf3f84-9d5d-4b21-80a0-54e9e513f7b6","year":2024},{"cited_arxiv_id":"","doi":"10.1001/journalofethics.2017.19.3.stas1-1703","is_internal_anchor":false,"ref_index":3,"title":"AMA journal of ethics , author =","work_id":"c8fd5c90-8a64-448f-b22f-64ef902ae287","year":2017},{"cited_arxiv_id":"","doi":"10.1002/wics.1549","is_internal_anchor":false,"ref_index":4,"title":"WIREs Computational Statistics , author =","work_id":"2dfbcab8-e428-4f15-aedd-905ef9aa0e76","year":2021},{"cited_arxiv_id":"","doi":"10.1016/j.isci.2024.109713","is_internal_anchor":false,"ref_index":5,"title":"iScience , author =","work_id":"acf0f2f5-0d21-4841-94a0-9e376fa33c40","year":2024}],"snapshot_sha256":"febf422a970a46c5f57efe1720c3f691045ab70d9445ee05ce14580ddab0f610"},"source":{"id":"2605.14126","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T04:53:35.494733Z","id":"2f1d9550-d38b-4e8b-8e77-f703c37219b7","model_set":{"reader":"grok-4.3"},"one_line_summary":"RL post-training lifts answer correctness on FHIR-AgentBench from 50% (o4-mini) to 77% with a cheaper Qwen3-8B CodeAct agent.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Reinforcement learning post-training raises a Qwen3-8B model to 77 percent correctness on FHIR clinical queries, surpassing larger closed models.","strongest_claim":"Empirically, our approach improves answer correctness from 50% (o4-mini) to 77% on FHIR-AgentBench using a smaller and cheaper Qwen3-8B model.","weakest_assumption":"That an LLM Judge can supply reliable execution-grounded rewards without bias or systematic errors in judging multi-step FHIR traversals."}},"verdict_id":"2f1d9550-d38b-4e8b-8e77-f703c37219b7"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b2f31ed8c879a69c15de4298a199b5522c44314f1d72cca45822039168f2aec5","target":"record","created_at":"2026-05-17T23:39:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"52e8853a78ec8c87734ac98a6b25bb71dc31ecc07f94480c1c465ac50bb3f52c","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T21:27:21Z","title_canon_sha256":"567d44002f2e6213549591e3fea47891316a6c7be41cba3fa01d3f383819bb43"},"schema_version":"1.0","source":{"id":"2605.14126","kind":"arxiv","version":1}},"canonical_sha256":"61f1c425cce82d0c8e97611589e8551de182451e0ffd6adfbd7199d79f0f8a12","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"61f1c425cce82d0c8e97611589e8551de182451e0ffd6adfbd7199d79f0f8a12","first_computed_at":"2026-05-17T23:39:11.844744Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:11.844744Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"N+z7dji17VA84s713s2VIuavtNqVIbFO1ixN/PAV00tvWrEZ4J+C8GT4Z7L+eM0V2xbArOJBda9s3tb+sQ8KAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:11.845231Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14126","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b2f31ed8c879a69c15de4298a199b5522c44314f1d72cca45822039168f2aec5","sha256:5f642abdb4260a358560bbda2dca7b3265fbf16f58e4d67bb835ae45c9f1cda2"],"state_sha256":"19693e11b950252dd8a36e7f28148c2c7703e648bb2657511a4eb81c09e66beb"}