{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:R4VTASEBVZYHL6WWOWAU7XAE52","short_pith_number":"pith:R4VTASEB","canonical_record":{"source":{"id":"2606.09635","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-08T15:33:13Z","cross_cats_sorted":[],"title_canon_sha256":"034b9cde875c54e8dc8d5158890771d5b03ef61c7e530d6c74110c556812c44e","abstract_canon_sha256":"f0c83fce2e668d9c506a530ecb0ea085424a28bdd48bb8525c8c9826185507b7"},"schema_version":"1.0"},"canonical_sha256":"8f2b304881ae7075fad675814fdc04eeb216836609611dff3ffa1a9885664de0","source":{"kind":"arxiv","id":"2606.09635","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.09635","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"arxiv_version","alias_value":"2606.09635v1","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.09635","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"pith_short_12","alias_value":"R4VTASEBVZYH","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"pith_short_16","alias_value":"R4VTASEBVZYHL6WW","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"pith_short_8","alias_value":"R4VTASEB","created_at":"2026-06-09T02:09:00Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:R4VTASEBVZYHL6WWOWAU7XAE52","target":"record","payload":{"canonical_record":{"source":{"id":"2606.09635","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-08T15:33:13Z","cross_cats_sorted":[],"title_canon_sha256":"034b9cde875c54e8dc8d5158890771d5b03ef61c7e530d6c74110c556812c44e","abstract_canon_sha256":"f0c83fce2e668d9c506a530ecb0ea085424a28bdd48bb8525c8c9826185507b7"},"schema_version":"1.0"},"canonical_sha256":"8f2b304881ae7075fad675814fdc04eeb216836609611dff3ffa1a9885664de0","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T02:09:00.302173Z","signature_b64":"WLjiLa+p93uy4safsatO/g+1S+NtNsB3YtNXTev0DjUxUG4u206YWKgGjP0XIrV3OOEvymG3AHs5cSywHlYRCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8f2b304881ae7075fad675814fdc04eeb216836609611dff3ffa1a9885664de0","last_reissued_at":"2026-06-09T02:09:00.301264Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T02:09:00.301264Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2606.09635","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-09T02:09:00Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HcMDYNUQO8nTP7CzXa6VYgNm+s4YWBGXNGZUp7E8zvl9PUrMeYMWSsvr2RpJ/9WupBBFSGBq6+Rt3Ax27vDTCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-03T07:33:09.474660Z"},"content_sha256":"55a4a42c8767ad94dc506f14a82a56c08816ac8de2ac8c88323dfbada2cffe2c","schema_version":"1.0","event_id":"sha256:55a4a42c8767ad94dc506f14a82a56c08816ac8de2ac8c88323dfbada2cffe2c"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:R4VTASEBVZYHL6WWOWAU7XAE52","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Gradient-Guided Reward Optimization for Inference-time Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hankun Lin, Ruqi Zhang","submitted_at":"2026-06-08T15:33:13Z","abstract_excerpt":"Ensuring the reliability of Large Language Models (LLMs) under distribution drift requires inference-time adaptation. While inference-time alignment methods such as Best-of-$N$ and rejection sampling are widely used, they frame the task as a sampling-intensive, reward-guided search, leading to two key limitations: their performance is bounded by the base model's generation quality, and their reliance on imperfect reward models makes them vulnerable to reward hacking. To address these challenges, we introduce Gradient-Guided Reward Optimization (GGRO), a lightweight inference-time method that p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.09635","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/2606.09635/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-09T02:09:00Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Cwb//UPdL4XWAEWHat5op2GYieuwTU1/j+hhOHue7OgMvoywN2Ph0S9+CiOlQtunrreuGVAooyyHiqpiAYwMDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-03T07:33:09.475027Z"},"content_sha256":"bd290d7592e9750611aa589d02091835e75e97daa7236d4a667d0a6ac0936dd8","schema_version":"1.0","event_id":"sha256:bd290d7592e9750611aa589d02091835e75e97daa7236d4a667d0a6ac0936dd8"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/R4VTASEBVZYHL6WWOWAU7XAE52/bundle.json","state_url":"https://pith.science/pith/R4VTASEBVZYHL6WWOWAU7XAE52/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/R4VTASEBVZYHL6WWOWAU7XAE52/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-03T07:33:09Z","links":{"resolver":"https://pith.science/pith/R4VTASEBVZYHL6WWOWAU7XAE52","bundle":"https://pith.science/pith/R4VTASEBVZYHL6WWOWAU7XAE52/bundle.json","state":"https://pith.science/pith/R4VTASEBVZYHL6WWOWAU7XAE52/state.json","well_known_bundle":"https://pith.science/.well-known/pith/R4VTASEBVZYHL6WWOWAU7XAE52/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:R4VTASEBVZYHL6WWOWAU7XAE52","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":"f0c83fce2e668d9c506a530ecb0ea085424a28bdd48bb8525c8c9826185507b7","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-08T15:33:13Z","title_canon_sha256":"034b9cde875c54e8dc8d5158890771d5b03ef61c7e530d6c74110c556812c44e"},"schema_version":"1.0","source":{"id":"2606.09635","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2606.09635","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"arxiv_version","alias_value":"2606.09635v1","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.09635","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"pith_short_12","alias_value":"R4VTASEBVZYH","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"pith_short_16","alias_value":"R4VTASEBVZYHL6WW","created_at":"2026-06-09T02:09:00Z"},{"alias_kind":"pith_short_8","alias_value":"R4VTASEB","created_at":"2026-06-09T02:09:00Z"}],"graph_snapshots":[{"event_id":"sha256:bd290d7592e9750611aa589d02091835e75e97daa7236d4a667d0a6ac0936dd8","target":"graph","created_at":"2026-06-09T02:09:00Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2606.09635/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Ensuring the reliability of Large Language Models (LLMs) under distribution drift requires inference-time adaptation. While inference-time alignment methods such as Best-of-$N$ and rejection sampling are widely used, they frame the task as a sampling-intensive, reward-guided search, leading to two key limitations: their performance is bounded by the base model's generation quality, and their reliance on imperfect reward models makes them vulnerable to reward hacking. To address these challenges, we introduce Gradient-Guided Reward Optimization (GGRO), a lightweight inference-time method that p","authors_text":"Hankun Lin, Ruqi Zhang","cross_cats":[],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-08T15:33:13Z","title":"Gradient-Guided Reward Optimization for Inference-time Alignment"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.09635","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:55a4a42c8767ad94dc506f14a82a56c08816ac8de2ac8c88323dfbada2cffe2c","target":"record","created_at":"2026-06-09T02:09:00Z","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":"f0c83fce2e668d9c506a530ecb0ea085424a28bdd48bb8525c8c9826185507b7","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-08T15:33:13Z","title_canon_sha256":"034b9cde875c54e8dc8d5158890771d5b03ef61c7e530d6c74110c556812c44e"},"schema_version":"1.0","source":{"id":"2606.09635","kind":"arxiv","version":1}},"canonical_sha256":"8f2b304881ae7075fad675814fdc04eeb216836609611dff3ffa1a9885664de0","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"8f2b304881ae7075fad675814fdc04eeb216836609611dff3ffa1a9885664de0","first_computed_at":"2026-06-09T02:09:00.301264Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-09T02:09:00.301264Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"WLjiLa+p93uy4safsatO/g+1S+NtNsB3YtNXTev0DjUxUG4u206YWKgGjP0XIrV3OOEvymG3AHs5cSywHlYRCw==","signature_status":"signed_v1","signed_at":"2026-06-09T02:09:00.302173Z","signed_message":"canonical_sha256_bytes"},"source_id":"2606.09635","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:55a4a42c8767ad94dc506f14a82a56c08816ac8de2ac8c88323dfbada2cffe2c","sha256:bd290d7592e9750611aa589d02091835e75e97daa7236d4a667d0a6ac0936dd8"],"state_sha256":"78bb1b49e73bad4f98f99cadd169c96468e6e2d223cd1a2fc8e6f5768078029b"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"uDHKX2rIFN1bPFVSr270NticAC/upMV5fbATiPEuyEDqwJuDabNAsm+jMu6gPOJFZi2iGwHmAg1E64Mz5ujIDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-03T07:33:09.477031Z","bundle_sha256":"8c2d4c08556eae22501a892d7dbc519bf5e09a11ae69725825d2a1ea3f99c097"}}