{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:QFMTYWN7NPNNS5RS46HYTUFL4S","short_pith_number":"pith:QFMTYWN7","schema_version":"1.0","canonical_sha256":"81593c59bf6bdad97632e78f89d0abe48a1864cdc6201d296d62e4ed5e9fc67c","source":{"kind":"arxiv","id":"2606.22840","version":1},"attestation_state":"computed","paper":{"title":"RLM-Cascade: Response-Level Speculative Decoding for Cost-Efficient LLM API Serving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Fangbo Tu, Haifeng Wu, Jian Wan, Junhua Zhao, Srinivasan Manoharan","submitted_at":"2026-06-22T04:27:45Z","abstract_excerpt":"We present RLM-Cascade, a proxy-layer system that applies speculative decoding at the response level to reduce LLM API costs without requiring model architecture access or a shared vocabulary. A fast, inexpensive draft model generates a candidate response; a capable verify model accepts, enhances, or is bypassed entirely depending on a lightweight complexity router. On a real-world agentic coding workload (Claude Code), RLM-Cascade achieves a draft-use rate of 88.8% across 125 production requests, reducing API cost by 45.8% relative to a direct Opus baseline. Counter-intuitively, the proxy als"},"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":"2606.22840","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-22T04:27:45Z","cross_cats_sorted":[],"title_canon_sha256":"e0648e9156edcec9db51b4165254d133bbb3d2b7d602bba90c01294ffb6529d9","abstract_canon_sha256":"62dde882873bf92ff2c194b241da3d4658989fa68751ecb2934bbb0c419a8dea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T02:14:00.826008Z","signature_b64":"dyy7k99IO8XeY95qy7BtXOBonTzx0utg+DarFskcVpO0Czn1XqUQufdUl4aH6r1vMNaQMQ38Dxe77suQaOpIBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"81593c59bf6bdad97632e78f89d0abe48a1864cdc6201d296d62e4ed5e9fc67c","last_reissued_at":"2026-06-23T02:14:00.825622Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T02:14:00.825622Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RLM-Cascade: Response-Level Speculative Decoding for Cost-Efficient LLM API Serving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Fangbo Tu, Haifeng Wu, Jian Wan, Junhua Zhao, Srinivasan Manoharan","submitted_at":"2026-06-22T04:27:45Z","abstract_excerpt":"We present RLM-Cascade, a proxy-layer system that applies speculative decoding at the response level to reduce LLM API costs without requiring model architecture access or a shared vocabulary. A fast, inexpensive draft model generates a candidate response; a capable verify model accepts, enhances, or is bypassed entirely depending on a lightweight complexity router. On a real-world agentic coding workload (Claude Code), RLM-Cascade achieves a draft-use rate of 88.8% across 125 production requests, reducing API cost by 45.8% relative to a direct Opus baseline. Counter-intuitively, the proxy als"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.22840","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/2606.22840/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":"2606.22840","created_at":"2026-06-23T02:14:00.825688+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.22840v1","created_at":"2026-06-23T02:14:00.825688+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.22840","created_at":"2026-06-23T02:14:00.825688+00:00"},{"alias_kind":"pith_short_12","alias_value":"QFMTYWN7NPNN","created_at":"2026-06-23T02:14:00.825688+00:00"},{"alias_kind":"pith_short_16","alias_value":"QFMTYWN7NPNNS5RS","created_at":"2026-06-23T02:14:00.825688+00:00"},{"alias_kind":"pith_short_8","alias_value":"QFMTYWN7","created_at":"2026-06-23T02:14:00.825688+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/QFMTYWN7NPNNS5RS46HYTUFL4S","json":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S.json","graph_json":"https://pith.science/api/pith-number/QFMTYWN7NPNNS5RS46HYTUFL4S/graph.json","events_json":"https://pith.science/api/pith-number/QFMTYWN7NPNNS5RS46HYTUFL4S/events.json","paper":"https://pith.science/paper/QFMTYWN7"},"agent_actions":{"view_html":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S","download_json":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S.json","view_paper":"https://pith.science/paper/QFMTYWN7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.22840&json=true","fetch_graph":"https://pith.science/api/pith-number/QFMTYWN7NPNNS5RS46HYTUFL4S/graph.json","fetch_events":"https://pith.science/api/pith-number/QFMTYWN7NPNNS5RS46HYTUFL4S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S/action/storage_attestation","attest_author":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S/action/author_attestation","sign_citation":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S/action/citation_signature","submit_replication":"https://pith.science/pith/QFMTYWN7NPNNS5RS46HYTUFL4S/action/replication_record"}},"created_at":"2026-06-23T02:14:00.825688+00:00","updated_at":"2026-06-23T02:14:00.825688+00:00"}