{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:S33Y7MIX3PBVGVVQ5NVSIBKEDP","short_pith_number":"pith:S33Y7MIX","schema_version":"1.0","canonical_sha256":"96f78fb117dbc35356b0eb6b2405441bcd68f0b6d29ceb73a9b18f10775bfea1","source":{"kind":"arxiv","id":"2602.11201","version":2},"attestation_state":"computed","paper":{"title":"Mechanistic Evidence for Faithfulness Decay in Chain-of-Thought Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Donald Ye, Jonas Rohweder, Linus Wong, Max Loffgren, Om Kotadia","submitted_at":"2026-02-04T21:55:57Z","abstract_excerpt":"Chain-of-Thought (CoT) explanations are widely used to interpret how language models solve complex problems, yet it remains unclear whether these step-by-step explanations reflect how the model actually reaches its answer, or merely post-hoc justifications. We propose Normalized Logit Difference Decay (NLDD), a metric that measures whether individual reasoning steps are faithful to the model's decision-making process. Our approach corrupts individual reasoning steps from the explanation and measures how much the model's confidence in its answer drops, to determine if a step is truly important."},"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":"2602.11201","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-04T21:55:57Z","cross_cats_sorted":[],"title_canon_sha256":"b17ed728afbee12db9a71a2092962dda67c02e08dc1700228fabc026f6ccb60e","abstract_canon_sha256":"bf13faa4303f4b6441b7ec346a0652df2e3e840e7e2e391bccecd2e2c6a763b2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-08T01:05:07.412394Z","signature_b64":"Cs/YpOIZ3mHF0q3dTZJClHgylo2YaJT5htKHNx0hiciqSAPawBtWwGCVjTFaTs1kq4g6TVyO6CigbjsMwV3gDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"96f78fb117dbc35356b0eb6b2405441bcd68f0b6d29ceb73a9b18f10775bfea1","last_reissued_at":"2026-06-08T01:05:07.411429Z","signature_status":"signed_v1","first_computed_at":"2026-06-08T01:05:07.411429Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mechanistic Evidence for Faithfulness Decay in Chain-of-Thought Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Donald Ye, Jonas Rohweder, Linus Wong, Max Loffgren, Om Kotadia","submitted_at":"2026-02-04T21:55:57Z","abstract_excerpt":"Chain-of-Thought (CoT) explanations are widely used to interpret how language models solve complex problems, yet it remains unclear whether these step-by-step explanations reflect how the model actually reaches its answer, or merely post-hoc justifications. We propose Normalized Logit Difference Decay (NLDD), a metric that measures whether individual reasoning steps are faithful to the model's decision-making process. Our approach corrupts individual reasoning steps from the explanation and measures how much the model's confidence in its answer drops, to determine if a step is truly important."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.11201","kind":"arxiv","version":2},"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/2602.11201/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":"2602.11201","created_at":"2026-06-08T01:05:07.411555+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.11201v2","created_at":"2026-06-08T01:05:07.411555+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.11201","created_at":"2026-06-08T01:05:07.411555+00:00"},{"alias_kind":"pith_short_12","alias_value":"S33Y7MIX3PBV","created_at":"2026-06-08T01:05:07.411555+00:00"},{"alias_kind":"pith_short_16","alias_value":"S33Y7MIX3PBVGVVQ","created_at":"2026-06-08T01:05:07.411555+00:00"},{"alias_kind":"pith_short_8","alias_value":"S33Y7MIX","created_at":"2026-06-08T01:05:07.411555+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2603.22816","citing_title":"Measuring and curing reasoning rigidity: from decorative chain-of-thought to genuine faithfulness","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11746","citing_title":"When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channel","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2604.15726","citing_title":"LLM Reasoning Is Latent, Not the Chain of Thought","ref_index":32,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP","json":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP.json","graph_json":"https://pith.science/api/pith-number/S33Y7MIX3PBVGVVQ5NVSIBKEDP/graph.json","events_json":"https://pith.science/api/pith-number/S33Y7MIX3PBVGVVQ5NVSIBKEDP/events.json","paper":"https://pith.science/paper/S33Y7MIX"},"agent_actions":{"view_html":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP","download_json":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP.json","view_paper":"https://pith.science/paper/S33Y7MIX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.11201&json=true","fetch_graph":"https://pith.science/api/pith-number/S33Y7MIX3PBVGVVQ5NVSIBKEDP/graph.json","fetch_events":"https://pith.science/api/pith-number/S33Y7MIX3PBVGVVQ5NVSIBKEDP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP/action/storage_attestation","attest_author":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP/action/author_attestation","sign_citation":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP/action/citation_signature","submit_replication":"https://pith.science/pith/S33Y7MIX3PBVGVVQ5NVSIBKEDP/action/replication_record"}},"created_at":"2026-06-08T01:05:07.411555+00:00","updated_at":"2026-06-08T01:05:07.411555+00:00"}