{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AG4N5RQFKKIKIXZ3A5M3KMSXZV","short_pith_number":"pith:AG4N5RQF","schema_version":"1.0","canonical_sha256":"01b8dec6055290a45f3b0759b53257cd74d33c4027b49067613f56423c6df3b6","source":{"kind":"arxiv","id":"2606.29054","version":1},"attestation_state":"computed","paper":{"title":"When Can Conformal Risk Control Certify LLM Outputs? Bounds, Impossibility, and Adaptation for Structured Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Varun Kotte","submitted_at":"2026-06-27T19:25:07Z","abstract_excerpt":"Large language models (LLMs) deployed for structured generation (NER, JSON extraction, QA, and classification) lack formal reliability guarantees, and standard heuristic abstention policies miss user-specified risk targets by 7.5--12.5%. We characterize when conformal risk control (CRC) can certify structured LLM outputs and when it provably cannot. First, we prove an impossibility result: when the base risk (\\mu > \\alpha), any distribution-free method must abstain on at least ((\\mu-\\alpha)/(1-\\alpha)) examples, yielding a closed-form feasibility test: one can check whether CRC will work befor"},"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":"2606.29054","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-27T19:25:07Z","cross_cats_sorted":[],"title_canon_sha256":"b4c9f7f8662e91b266e1efcd81d9b50aacbb31b0ba3853d4e2291931cd51f04a","abstract_canon_sha256":"537fe47b534496f7fcc66f7fc200175934ce0db0a6b930bce631a62188a2cc41"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T01:17:50.960628Z","signature_b64":"HnztGjgt/HSMu6Fgv5e3s6wTKb8iH/Kd2y2uq9dVMb6YPExARmAoftb77wTUzFxXgeLtdwTZ8FadT9EwRkcpBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"01b8dec6055290a45f3b0759b53257cd74d33c4027b49067613f56423c6df3b6","last_reissued_at":"2026-06-30T01:17:50.959970Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T01:17:50.959970Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Can Conformal Risk Control Certify LLM Outputs? Bounds, Impossibility, and Adaptation for Structured Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Varun Kotte","submitted_at":"2026-06-27T19:25:07Z","abstract_excerpt":"Large language models (LLMs) deployed for structured generation (NER, JSON extraction, QA, and classification) lack formal reliability guarantees, and standard heuristic abstention policies miss user-specified risk targets by 7.5--12.5%. We characterize when conformal risk control (CRC) can certify structured LLM outputs and when it provably cannot. First, we prove an impossibility result: when the base risk (\\mu > \\alpha), any distribution-free method must abstain on at least ((\\mu-\\alpha)/(1-\\alpha)) examples, yielding a closed-form feasibility test: one can check whether CRC will work befor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.29054","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/2606.29054/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":"2606.29054","created_at":"2026-06-30T01:17:50.960070+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.29054v1","created_at":"2026-06-30T01:17:50.960070+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.29054","created_at":"2026-06-30T01:17:50.960070+00:00"},{"alias_kind":"pith_short_12","alias_value":"AG4N5RQFKKIK","created_at":"2026-06-30T01:17:50.960070+00:00"},{"alias_kind":"pith_short_16","alias_value":"AG4N5RQFKKIKIXZ3","created_at":"2026-06-30T01:17:50.960070+00:00"},{"alias_kind":"pith_short_8","alias_value":"AG4N5RQF","created_at":"2026-06-30T01:17:50.960070+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2608.12444","citing_title":"Non-Degenerate Risk Certification for Automated Security Decisions: A Decision-Contract Theory with ATT\\&CK-Aligned Triage as a Worked Instance","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV","json":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV.json","graph_json":"https://pith.science/api/pith-number/AG4N5RQFKKIKIXZ3A5M3KMSXZV/graph.json","events_json":"https://pith.science/api/pith-number/AG4N5RQFKKIKIXZ3A5M3KMSXZV/events.json","paper":"https://pith.science/paper/AG4N5RQF"},"agent_actions":{"view_html":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV","download_json":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV.json","view_paper":"https://pith.science/paper/AG4N5RQF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.29054&json=true","fetch_graph":"https://pith.science/api/pith-number/AG4N5RQFKKIKIXZ3A5M3KMSXZV/graph.json","fetch_events":"https://pith.science/api/pith-number/AG4N5RQFKKIKIXZ3A5M3KMSXZV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV/action/storage_attestation","attest_author":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV/action/author_attestation","sign_citation":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV/action/citation_signature","submit_replication":"https://pith.science/pith/AG4N5RQFKKIKIXZ3A5M3KMSXZV/action/replication_record"}},"created_at":"2026-06-30T01:17:50.960070+00:00","updated_at":"2026-06-30T01:17:50.960070+00:00"}