{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:YWL5L3BMRWK7EETKAU55XA7SQG","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":"076a6624ab838da8b08dc2461a7151832c8763030ca32bb3b8b238ddc853a2cc","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-16T05:57:00Z","title_canon_sha256":"2f60d239e717102168fbc9d854150d6736dca42974dc85134ff1492f763d4d03"},"schema_version":"1.0","source":{"id":"2605.16824","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.16824","created_at":"2026-05-20T00:03:24Z"},{"alias_kind":"arxiv_version","alias_value":"2605.16824v1","created_at":"2026-05-20T00:03:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16824","created_at":"2026-05-20T00:03:24Z"},{"alias_kind":"pith_short_12","alias_value":"YWL5L3BMRWK7","created_at":"2026-05-20T00:03:24Z"},{"alias_kind":"pith_short_16","alias_value":"YWL5L3BMRWK7EETK","created_at":"2026-05-20T00:03:24Z"},{"alias_kind":"pith_short_8","alias_value":"YWL5L3BM","created_at":"2026-05-20T00:03:24Z"}],"graph_snapshots":[{"event_id":"sha256:23a765b8d8595ab0ba3fc6ead4d7ea8c962fb6f2bc0b5c5b24210a5bc6e0f46d","target":"graph","created_at":"2026-05-20T00:03:24Z","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":"Using only token-level confidence values, without access to the input question, reasoning text, hidden states, or external verifiers, low-dimensional representations of confidence trajectories separate correct from incorrect reasoning traces."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the observed low-dimensional separation arises specifically from trace-level correctness rather than from other correlated properties of the generation process such as length, token distribution, or model-specific artifacts, and that this separation generalizes beyond the three evaluated benchmarks without content information."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Token-level confidence trajectories in LLMs encode a content-agnostic geometry that separates correct and incorrect reasoning traces and supports a lightweight correctness estimator called NeuralConf."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Token-level confidence trajectories in LLMs form low-dimensional geometries that separate correct from incorrect reasoning traces without using question or text content."}],"snapshot_sha256":"6f8d105ae6415ed8f6673440e7d7bd04a4cf4028724b3ef01bf8f72402750ce3"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T21:01:26.409011Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.265439Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T19:01:56.265002Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.407531Z","status":"skipped","version":"1.0.0"}],"endpoint":"/pith/2605.16824/integrity.json","findings":[],"snapshot_sha256":"655da22412969dd94f9148298f2ca415cf61f0164ae069d3574bf243a6a18302","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) generate not only reasoning text, but also token-level confidence trajectories that record how uncertainty evolves during inference. Whether these trajectories are relevant to reasoning correctness remains unclear. Here we show that confidence trajectories encode a content-agnostic confidence geometry associated with trace-level final-answer correctness. Using only token-level confidence values, without access to the input question, reasoning text, hidden states, or external verifiers, we find that low-dimensional representations of confidence trajectories separate","authors_text":"Ding Liu, Shi-Ju Ran, Shuo Liu","cross_cats":["cs.CL"],"headline":"Token-level confidence trajectories in LLMs form low-dimensional geometries that separate correct from incorrect reasoning traces without using question or text content.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-16T05:57:00Z","title":"Confidence Geometry Reveals Trace-Level Correctness in Large Language Model Reasoning"},"references":{"count":32,"internal_anchors":7,"resolved_work":32,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Chain-of- thought prompting elicits reasoning in large language models,","work_id":"a072c1ae-ca12-411a-8ff7-c501d9d89c6a","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Large language models are zero-shot reasoners,","work_id":"3117d03a-1cd7-4a6d-b581-fc15bce9a095","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Least-to-most prompting enables complex reasoning in large language models,","work_id":"b11eff8b-8a5f-4338-ba00-0eb3459b4f6d","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Self-consistency improves chain of thought reasoning in language models,","work_id":"d3947bab-e558-4cac-90db-f5bfb5eb3e31","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Tree of thoughts: Deliberate problem solving with large language models,","work_id":"c4646d28-69bb-440a-9038-a4bed3a63804","year":2023}],"snapshot_sha256":"c1a6d303708631bc547fbd0cea936b53908e1ca4246bf729e7525f2aa4e9ead4"},"source":{"id":"2605.16824","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T20:50:25.533304Z","id":"8aed55bb-914b-4a39-acab-eaa6d0d3ffdc","model_set":{"reader":"grok-4.3"},"one_line_summary":"Token-level confidence trajectories in LLMs encode a content-agnostic geometry that separates correct and incorrect reasoning traces and supports a lightweight correctness estimator called NeuralConf.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Token-level confidence trajectories in LLMs form low-dimensional geometries that separate correct from incorrect reasoning traces without using question or text content.","strongest_claim":"Using only token-level confidence values, without access to the input question, reasoning text, hidden states, or external verifiers, low-dimensional representations of confidence trajectories separate correct from incorrect reasoning traces.","weakest_assumption":"That the observed low-dimensional separation arises specifically from trace-level correctness rather than from other correlated properties of the generation process such as length, token distribution, or model-specific artifacts, and that this separation generalizes beyond the three evaluated benchmarks without content information."}},"verdict_id":"8aed55bb-914b-4a39-acab-eaa6d0d3ffdc"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:10b3f05654566bd3daf0ed2f25973817b2a119ca6e37e97574f13dc308d48527","target":"record","created_at":"2026-05-20T00:03:24Z","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":"076a6624ab838da8b08dc2461a7151832c8763030ca32bb3b8b238ddc853a2cc","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-16T05:57:00Z","title_canon_sha256":"2f60d239e717102168fbc9d854150d6736dca42974dc85134ff1492f763d4d03"},"schema_version":"1.0","source":{"id":"2605.16824","kind":"arxiv","version":1}},"canonical_sha256":"c597d5ec2c8d95f2126a053bdb83f281ab06f4ba52ee9c216bdeec9c98193bd1","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c597d5ec2c8d95f2126a053bdb83f281ab06f4ba52ee9c216bdeec9c98193bd1","first_computed_at":"2026-05-20T00:03:24.518186Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:03:24.518186Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"MYnf6TXLMMn5GbELkUs+pEBrlL6eHGs+iiV8QrvWWQWL0BNDWJKzZpD1MzKfOPhkf+e3W5Af10GuIcXgvFRtAA==","signature_status":"signed_v1","signed_at":"2026-05-20T00:03:24.519356Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.16824","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:10b3f05654566bd3daf0ed2f25973817b2a119ca6e37e97574f13dc308d48527","sha256:23a765b8d8595ab0ba3fc6ead4d7ea8c962fb6f2bc0b5c5b24210a5bc6e0f46d"],"state_sha256":"1481c70bb8c05068e53f3204472cc378af0419edc1c6313fb00bbb0769b073ad"}