{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:ACIPAPZYL7MOMYKA4MVAYLQX3C","short_pith_number":"pith:ACIPAPZY","canonical_record":{"source":{"id":"2602.02427","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T18:27:26Z","cross_cats_sorted":[],"title_canon_sha256":"052935d791537b07cfb9f5d46b6fb642e3c0ab6124d30f7114e9c46e5eb899e9","abstract_canon_sha256":"85e5e3eea633bf39d658a3659a0f1928a9ba19b08e4f5c8bada3f5a6ddf99818"},"schema_version":"1.0"},"canonical_sha256":"0090f03f385fd8e66140e32a0c2e17d8940b1c79c972160dc6fe81dac406c767","source":{"kind":"arxiv","id":"2602.02427","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.02427","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"arxiv_version","alias_value":"2602.02427v2","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.02427","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"pith_short_12","alias_value":"ACIPAPZYL7MO","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"ACIPAPZYL7MOMYKA","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"ACIPAPZY","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:ACIPAPZYL7MOMYKA4MVAYLQX3C","target":"record","payload":{"canonical_record":{"source":{"id":"2602.02427","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T18:27:26Z","cross_cats_sorted":[],"title_canon_sha256":"052935d791537b07cfb9f5d46b6fb642e3c0ab6124d30f7114e9c46e5eb899e9","abstract_canon_sha256":"85e5e3eea633bf39d658a3659a0f1928a9ba19b08e4f5c8bada3f5a6ddf99818"},"schema_version":"1.0"},"canonical_sha256":"0090f03f385fd8e66140e32a0c2e17d8940b1c79c972160dc6fe81dac406c767","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:16.433623Z","signature_b64":"ACeYe+2qvpOAtrSLMQmUVTdy/UEYn08PLNxpbU4965NZ8yg9mWbUUGVxLkN3nZRnqa5QFfZ57OUCxoBwTHwRAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0090f03f385fd8e66140e32a0c2e17d8940b1c79c972160dc6fe81dac406c767","last_reissued_at":"2026-05-17T23:39:16.432927Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:16.432927Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2602.02427","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gE3YRPpMdf6ajq6PD9NKEka5GNcMO474BUzV63hvhq9vXcPz/AEicHQSIOpI2UpoMkRWFnbWqom8KnY1KknVAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T22:09:26.022021Z"},"content_sha256":"8440f75149701995dcee3ec2d61338585a566830bc0e60d734164126be961f83","schema_version":"1.0","event_id":"sha256:8440f75149701995dcee3ec2d61338585a566830bc0e60d734164126be961f83"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:ACIPAPZYL7MOMYKA4MVAYLQX3C","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Embedding Perturbation may Better Reflect Intermediate-Step Uncertainty in LLM Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Embedding perturbations on preceding tokens reveal uncertainty in LLM reasoning steps better than existing methods.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Han Xu, Jiahao Wang, Pengfei He, Qihao Wen, Ravi Tandon, Yang Nan","submitted_at":"2026-02-02T18:27:26Z","abstract_excerpt":"Large language Models (LLMs) have achieved significant breakthroughs across diverse domains; however, they can still produce unreliable or misleading outputs. For responsible LLM application, Uncertainty Quantification (UQ) techniques are used to estimate a model's uncertainty about its outputs, indicating the likelihood that those outputs may be problematic. For LLM reasoning tasks, it is essential to estimate the uncertainty not only for the final answer, but also for the intermediate steps of the reasoning, as this can enable more fine-grained and targeted interventions. In this study, we e"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"an LLM's incorrect reasoning steps tend to contain tokens which are highly sensitive to the perturbations on the preceding token embeddings, indicating the model's uncertainty among multiple competing continuations","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the observed sensitivity to embedding perturbations directly reflects the model's internal uncertainty in reasoning steps rather than being driven by other factors such as token frequency or architecture-specific artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Embedding perturbation sensitivity provides a stronger signal for uncertainty in intermediate LLM reasoning steps than probability, sampling, or Bayesian baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Embedding perturbations on preceding tokens reveal uncertainty in LLM reasoning steps better than existing methods.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"657847b8490c9beef4379e317d71c4752c6c98dc4ff15e70b6e087f1a446c8bf"},"source":{"id":"2602.02427","kind":"arxiv","version":2},"verdict":{"id":"6e3e9bb6-e617-46ba-bb5e-fcfceaaa0c62","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T08:20:00.476682Z","strongest_claim":"an LLM's incorrect reasoning steps tend to contain tokens which are highly sensitive to the perturbations on the preceding token embeddings, indicating the model's uncertainty among multiple competing continuations","one_line_summary":"Embedding perturbation sensitivity provides a stronger signal for uncertainty in intermediate LLM reasoning steps than probability, sampling, or Bayesian baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the observed sensitivity to embedding perturbations directly reflects the model's internal uncertainty in reasoning steps rather than being driven by other factors such as token frequency or architecture-specific artifacts.","pith_extraction_headline":"Embedding perturbations on preceding tokens reveal uncertainty in LLM reasoning steps better than existing methods."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ba82750b5bb0cd3d03d603261f21ab98218c2540d3f42d5557cd2595f22a6c52"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"6e3e9bb6-e617-46ba-bb5e-fcfceaaa0c62"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:16Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"qDkX6Mqw4Rxovd5mUUZUmQZy322/mHKwTPxko9mbcUPG2xyMVFMfF60V1Ym8+8zYqaFb7E6H5HJfX8Q3n4moBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T22:09:26.022902Z"},"content_sha256":"18f32f7c5ae681e761935b651540a90538f63abd33b8e3379bd477303a4652f0","schema_version":"1.0","event_id":"sha256:18f32f7c5ae681e761935b651540a90538f63abd33b8e3379bd477303a4652f0"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C/bundle.json","state_url":"https://pith.science/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C/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":"2026-05-27T22:09:26Z","links":{"resolver":"https://pith.science/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C","bundle":"https://pith.science/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C/bundle.json","state":"https://pith.science/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C/state.json","well_known_bundle":"https://pith.science/.well-known/pith/ACIPAPZYL7MOMYKA4MVAYLQX3C/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:ACIPAPZYL7MOMYKA4MVAYLQX3C","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":"85e5e3eea633bf39d658a3659a0f1928a9ba19b08e4f5c8bada3f5a6ddf99818","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T18:27:26Z","title_canon_sha256":"052935d791537b07cfb9f5d46b6fb642e3c0ab6124d30f7114e9c46e5eb899e9"},"schema_version":"1.0","source":{"id":"2602.02427","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.02427","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"arxiv_version","alias_value":"2602.02427v2","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.02427","created_at":"2026-05-17T23:39:16Z"},{"alias_kind":"pith_short_12","alias_value":"ACIPAPZYL7MO","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"ACIPAPZYL7MOMYKA","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"ACIPAPZY","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:18f32f7c5ae681e761935b651540a90538f63abd33b8e3379bd477303a4652f0","target":"graph","created_at":"2026-05-17T23:39:16Z","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":"an LLM's incorrect reasoning steps tend to contain tokens which are highly sensitive to the perturbations on the preceding token embeddings, indicating the model's uncertainty among multiple competing continuations"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the observed sensitivity to embedding perturbations directly reflects the model's internal uncertainty in reasoning steps rather than being driven by other factors such as token frequency or architecture-specific artifacts."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Embedding perturbation sensitivity provides a stronger signal for uncertainty in intermediate LLM reasoning steps than probability, sampling, or Bayesian baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Embedding perturbations on preceding tokens reveal uncertainty in LLM reasoning steps better than existing methods."}],"snapshot_sha256":"657847b8490c9beef4379e317d71c4752c6c98dc4ff15e70b6e087f1a446c8bf"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ba82750b5bb0cd3d03d603261f21ab98218c2540d3f42d5557cd2595f22a6c52"},"paper":{"abstract_excerpt":"Large language Models (LLMs) have achieved significant breakthroughs across diverse domains; however, they can still produce unreliable or misleading outputs. For responsible LLM application, Uncertainty Quantification (UQ) techniques are used to estimate a model's uncertainty about its outputs, indicating the likelihood that those outputs may be problematic. For LLM reasoning tasks, it is essential to estimate the uncertainty not only for the final answer, but also for the intermediate steps of the reasoning, as this can enable more fine-grained and targeted interventions. In this study, we e","authors_text":"Han Xu, Jiahao Wang, Pengfei He, Qihao Wen, Ravi Tandon, Yang Nan","cross_cats":[],"headline":"Embedding perturbations on preceding tokens reveal uncertainty in LLM reasoning steps better than existing methods.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T18:27:26Z","title":"Embedding Perturbation may Better Reflect Intermediate-Step Uncertainty in LLM Reasoning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.02427","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-16T08:20:00.476682Z","id":"6e3e9bb6-e617-46ba-bb5e-fcfceaaa0c62","model_set":{"reader":"grok-4.3"},"one_line_summary":"Embedding perturbation sensitivity provides a stronger signal for uncertainty in intermediate LLM reasoning steps than probability, sampling, or Bayesian baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Embedding perturbations on preceding tokens reveal uncertainty in LLM reasoning steps better than existing methods.","strongest_claim":"an LLM's incorrect reasoning steps tend to contain tokens which are highly sensitive to the perturbations on the preceding token embeddings, indicating the model's uncertainty among multiple competing continuations","weakest_assumption":"That the observed sensitivity to embedding perturbations directly reflects the model's internal uncertainty in reasoning steps rather than being driven by other factors such as token frequency or architecture-specific artifacts."}},"verdict_id":"6e3e9bb6-e617-46ba-bb5e-fcfceaaa0c62"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:8440f75149701995dcee3ec2d61338585a566830bc0e60d734164126be961f83","target":"record","created_at":"2026-05-17T23:39:16Z","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":"85e5e3eea633bf39d658a3659a0f1928a9ba19b08e4f5c8bada3f5a6ddf99818","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T18:27:26Z","title_canon_sha256":"052935d791537b07cfb9f5d46b6fb642e3c0ab6124d30f7114e9c46e5eb899e9"},"schema_version":"1.0","source":{"id":"2602.02427","kind":"arxiv","version":2}},"canonical_sha256":"0090f03f385fd8e66140e32a0c2e17d8940b1c79c972160dc6fe81dac406c767","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0090f03f385fd8e66140e32a0c2e17d8940b1c79c972160dc6fe81dac406c767","first_computed_at":"2026-05-17T23:39:16.432927Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:16.432927Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ACeYe+2qvpOAtrSLMQmUVTdy/UEYn08PLNxpbU4965NZ8yg9mWbUUGVxLkN3nZRnqa5QFfZ57OUCxoBwTHwRAw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:16.433623Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.02427","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8440f75149701995dcee3ec2d61338585a566830bc0e60d734164126be961f83","sha256:18f32f7c5ae681e761935b651540a90538f63abd33b8e3379bd477303a4652f0"],"state_sha256":"12249f829816a3869d1f505d299ca24788b269719622a30f437db337f244960a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"09g28JKZFjcQ3UaWO8ymQ4AXdcfxM/7XOiPYylWBvoskN7NKsg96zvafiKOLClZtLANQ6aIXBHMoF4TjhXiJDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-27T22:09:26.027125Z","bundle_sha256":"561cc12dcd08d0f634de7b3c5fab03f3b871c47ad00cf766fadb1bac632f4e80"}}