{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:U7YYKK4MBBTTNMWWWZLDBKRI3E","short_pith_number":"pith:U7YYKK4M","schema_version":"1.0","canonical_sha256":"a7f1852b8c086736b2d6b65630aa28d9158d9b0433de736da3e885fd9ba08782","source":{"kind":"arxiv","id":"2605.19228","version":1},"attestation_state":"computed","paper":{"title":"Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.IT","cs.LG","math.IT"],"primary_cat":"cs.CL","authors_text":"Dengjia Zhang, Hua Wei, Lu Cheng, Tiejin Chen, Xiaoou Liu, Yaqing Wang","submitted_at":"2026-05-19T00:57:51Z","abstract_excerpt":"Large Language Models have achieved strong performance on reasoning tasks with objective answers by generating step-by-step solutions, but diagnosing where a multi-step reasoning trace might fail remains difficult. Confidence estimation offers a diagnostic signal, yet existing methods are restricted to final answers or require internal model access. In this paper, we introduce Stepwise Confidence Attribution (SCA), a framework for closed-source LLMs that assigns step-level confidence based only on generated reasoning traces. SCA applies the Information Bottleneck principle: steps aligning with"},"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":"2605.19228","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-19T00:57:51Z","cross_cats_sorted":["cs.AI","cs.IT","cs.LG","math.IT"],"title_canon_sha256":"3fd87d5deb5441d943835b20bfedeb99ac007ae52968b8765859065729483f05","abstract_canon_sha256":"aa190b13477e1c13e8fc5905af4c4da24965768a7d948749c4fb819504b6bf53"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:34.526580Z","signature_b64":"9iuipZ0H6S8RYGxBqX7NsvLR7jHPyKQVO1Hnl7ubuXS4hqv3KdKPyxEK8ZaGdXgAILoSGheaHA8TyAPCSh02BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a7f1852b8c086736b2d6b65630aa28d9158d9b0433de736da3e885fd9ba08782","last_reissued_at":"2026-05-20T01:05:34.525942Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:34.525942Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.IT","cs.LG","math.IT"],"primary_cat":"cs.CL","authors_text":"Dengjia Zhang, Hua Wei, Lu Cheng, Tiejin Chen, Xiaoou Liu, Yaqing Wang","submitted_at":"2026-05-19T00:57:51Z","abstract_excerpt":"Large Language Models have achieved strong performance on reasoning tasks with objective answers by generating step-by-step solutions, but diagnosing where a multi-step reasoning trace might fail remains difficult. Confidence estimation offers a diagnostic signal, yet existing methods are restricted to final answers or require internal model access. In this paper, we introduce Stepwise Confidence Attribution (SCA), a framework for closed-source LLMs that assigns step-level confidence based only on generated reasoning traces. SCA applies the Information Bottleneck principle: steps aligning with"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.19228","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/2605.19228/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":"2605.19228","created_at":"2026-05-20T01:05:34.526024+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.19228v1","created_at":"2026-05-20T01:05:34.526024+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.19228","created_at":"2026-05-20T01:05:34.526024+00:00"},{"alias_kind":"pith_short_12","alias_value":"U7YYKK4MBBTT","created_at":"2026-05-20T01:05:34.526024+00:00"},{"alias_kind":"pith_short_16","alias_value":"U7YYKK4MBBTTNMWW","created_at":"2026-05-20T01:05:34.526024+00:00"},{"alias_kind":"pith_short_8","alias_value":"U7YYKK4M","created_at":"2026-05-20T01:05:34.526024+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/U7YYKK4MBBTTNMWWWZLDBKRI3E","json":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E.json","graph_json":"https://pith.science/api/pith-number/U7YYKK4MBBTTNMWWWZLDBKRI3E/graph.json","events_json":"https://pith.science/api/pith-number/U7YYKK4MBBTTNMWWWZLDBKRI3E/events.json","paper":"https://pith.science/paper/U7YYKK4M"},"agent_actions":{"view_html":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E","download_json":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E.json","view_paper":"https://pith.science/paper/U7YYKK4M","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.19228&json=true","fetch_graph":"https://pith.science/api/pith-number/U7YYKK4MBBTTNMWWWZLDBKRI3E/graph.json","fetch_events":"https://pith.science/api/pith-number/U7YYKK4MBBTTNMWWWZLDBKRI3E/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E/action/storage_attestation","attest_author":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E/action/author_attestation","sign_citation":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E/action/citation_signature","submit_replication":"https://pith.science/pith/U7YYKK4MBBTTNMWWWZLDBKRI3E/action/replication_record"}},"created_at":"2026-05-20T01:05:34.526024+00:00","updated_at":"2026-05-20T01:05:34.526024+00:00"}