{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HS5J7IJ7DHPZA5INVNU7MESLOG","short_pith_number":"pith:HS5J7IJ7","schema_version":"1.0","canonical_sha256":"3cba9fa13f19df90750dab69f6124b7197b89173425da5258bcebc9e3cc77e5b","source":{"kind":"arxiv","id":"2508.15842","version":1},"attestation_state":"computed","paper":{"title":"Lexical Hints of Accuracy in LLM Reasoning Chains","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Andres Algaba, Arne Vanhoyweghen, Brecht Verbeken, Vincent Ginis","submitted_at":"2025-08-19T18:20:38Z","abstract_excerpt":"Fine-tuning Large Language Models (LLMs) with reinforcement learning to produce an explicit Chain-of-Thought (CoT) before answering produces models that consistently raise overall performance on code, math, and general-knowledge benchmarks. However, on benchmarks where LLMs currently achieve low accuracy, such as Humanity's Last Exam (HLE), they often report high self-confidence, reflecting poor calibration. Here, we test whether measurable properties of the CoT provide reliable signals of an LLM's internal confidence in its answers. We analyze three feature classes: (i) CoT length, (ii) intra"},"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":"2508.15842","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-08-19T18:20:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"8ea77c80f42142f534b37783da74b172cf56e8e5d4cb37f3f9d13bd04b056d24","abstract_canon_sha256":"cfa64176186bddd4008a6714986d19f4842d612bfeee57626e7d613d45243cfa"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:57:21.851470Z","signature_b64":"y/9ZwWT5B8h124GLh8NZE+YF7fjdUOMwaj6ACqVTaMUQtZkZCfu64SZhY10kmQT3zt4yoHF/3MvcOq+0Wt1tDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3cba9fa13f19df90750dab69f6124b7197b89173425da5258bcebc9e3cc77e5b","last_reissued_at":"2026-07-05T11:57:21.851051Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:57:21.851051Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Lexical Hints of Accuracy in LLM Reasoning Chains","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Andres Algaba, Arne Vanhoyweghen, Brecht Verbeken, Vincent Ginis","submitted_at":"2025-08-19T18:20:38Z","abstract_excerpt":"Fine-tuning Large Language Models (LLMs) with reinforcement learning to produce an explicit Chain-of-Thought (CoT) before answering produces models that consistently raise overall performance on code, math, and general-knowledge benchmarks. However, on benchmarks where LLMs currently achieve low accuracy, such as Humanity's Last Exam (HLE), they often report high self-confidence, reflecting poor calibration. Here, we test whether measurable properties of the CoT provide reliable signals of an LLM's internal confidence in its answers. We analyze three feature classes: (i) CoT length, (ii) intra"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.15842","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/2508.15842/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":"2508.15842","created_at":"2026-07-05T11:57:21.851105+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.15842v1","created_at":"2026-07-05T11:57:21.851105+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.15842","created_at":"2026-07-05T11:57:21.851105+00:00"},{"alias_kind":"pith_short_12","alias_value":"HS5J7IJ7DHPZ","created_at":"2026-07-05T11:57:21.851105+00:00"},{"alias_kind":"pith_short_16","alias_value":"HS5J7IJ7DHPZA5IN","created_at":"2026-07-05T11:57:21.851105+00:00"},{"alias_kind":"pith_short_8","alias_value":"HS5J7IJ7","created_at":"2026-07-05T11:57:21.851105+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.08346","citing_title":"Sanity Checks for Long-Form Hallucination Detection","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG","json":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG.json","graph_json":"https://pith.science/api/pith-number/HS5J7IJ7DHPZA5INVNU7MESLOG/graph.json","events_json":"https://pith.science/api/pith-number/HS5J7IJ7DHPZA5INVNU7MESLOG/events.json","paper":"https://pith.science/paper/HS5J7IJ7"},"agent_actions":{"view_html":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG","download_json":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG.json","view_paper":"https://pith.science/paper/HS5J7IJ7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.15842&json=true","fetch_graph":"https://pith.science/api/pith-number/HS5J7IJ7DHPZA5INVNU7MESLOG/graph.json","fetch_events":"https://pith.science/api/pith-number/HS5J7IJ7DHPZA5INVNU7MESLOG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG/action/storage_attestation","attest_author":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG/action/author_attestation","sign_citation":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG/action/citation_signature","submit_replication":"https://pith.science/pith/HS5J7IJ7DHPZA5INVNU7MESLOG/action/replication_record"}},"created_at":"2026-07-05T11:57:21.851105+00:00","updated_at":"2026-07-05T11:57:21.851105+00:00"}