{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WMDNK3SSJDUMFITLMQF6NOA4LN","short_pith_number":"pith:WMDNK3SS","schema_version":"1.0","canonical_sha256":"b306d56e5248e8c2a26b640be6b81c5b4a6b5258b4b08b1808ae62cdc370b8be","source":{"kind":"arxiv","id":"2603.03031","version":2},"attestation_state":"computed","paper":{"title":"Step-Level Sparse Autoencoder for Reasoning Process Interpretation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hao Xu, Jiayu Liu, Ning Miao, Xuan Yang, Yuhang Lai, Zhenya Huang","submitted_at":"2026-03-03T14:25:02Z","abstract_excerpt":"Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning. However, their reasoning patterns remain too complicated to analyze. While Sparse Autoencoders (SAEs) have emerged as a powerful tool for interpretability, existing approaches predominantly operate at the token level, creating a granularity mismatch when capturing more critical step-level information, such as reasoning direction and semantic transitions. In this work, we propose step-level sparse autoencoder (SSAE), which serves as an analytical tool to disentangle differe"},"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":"2603.03031","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-03-03T14:25:02Z","cross_cats_sorted":[],"title_canon_sha256":"8b1dbcf7cdf3353b7e553ceff0544b827d23106d0bd96949c1a8930d5cfcdd1c","abstract_canon_sha256":"155c397cfdff5a0d6254fdf6b6bc06af2fa62506dabd5ee5e0b40c81b4581978"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T02:04:51.974650Z","signature_b64":"SMS9YrzY6hxT7Tcy/4r3U54dNbBbA+fwIckrCiprphnOmgZNyO1PpdNu4X1p+SGA9AcP5WLQb0pWoPD0/o8rBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b306d56e5248e8c2a26b640be6b81c5b4a6b5258b4b08b1808ae62cdc370b8be","last_reissued_at":"2026-06-02T02:04:51.974155Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T02:04:51.974155Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Step-Level Sparse Autoencoder for Reasoning Process Interpretation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hao Xu, Jiayu Liu, Ning Miao, Xuan Yang, Yuhang Lai, Zhenya Huang","submitted_at":"2026-03-03T14:25:02Z","abstract_excerpt":"Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning. However, their reasoning patterns remain too complicated to analyze. While Sparse Autoencoders (SAEs) have emerged as a powerful tool for interpretability, existing approaches predominantly operate at the token level, creating a granularity mismatch when capturing more critical step-level information, such as reasoning direction and semantic transitions. In this work, we propose step-level sparse autoencoder (SSAE), which serves as an analytical tool to disentangle differe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.03031","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/2603.03031/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":"2603.03031","created_at":"2026-06-02T02:04:51.974217+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.03031v2","created_at":"2026-06-02T02:04:51.974217+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.03031","created_at":"2026-06-02T02:04:51.974217+00:00"},{"alias_kind":"pith_short_12","alias_value":"WMDNK3SSJDUM","created_at":"2026-06-02T02:04:51.974217+00:00"},{"alias_kind":"pith_short_16","alias_value":"WMDNK3SSJDUMFITL","created_at":"2026-06-02T02:04:51.974217+00:00"},{"alias_kind":"pith_short_8","alias_value":"WMDNK3SS","created_at":"2026-06-02T02:04:51.974217+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.23040","citing_title":"Steered Generation via Gradient-Based Optimization on Sparse Query Features","ref_index":48,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN","json":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN.json","graph_json":"https://pith.science/api/pith-number/WMDNK3SSJDUMFITLMQF6NOA4LN/graph.json","events_json":"https://pith.science/api/pith-number/WMDNK3SSJDUMFITLMQF6NOA4LN/events.json","paper":"https://pith.science/paper/WMDNK3SS"},"agent_actions":{"view_html":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN","download_json":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN.json","view_paper":"https://pith.science/paper/WMDNK3SS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.03031&json=true","fetch_graph":"https://pith.science/api/pith-number/WMDNK3SSJDUMFITLMQF6NOA4LN/graph.json","fetch_events":"https://pith.science/api/pith-number/WMDNK3SSJDUMFITLMQF6NOA4LN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN/action/storage_attestation","attest_author":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN/action/author_attestation","sign_citation":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN/action/citation_signature","submit_replication":"https://pith.science/pith/WMDNK3SSJDUMFITLMQF6NOA4LN/action/replication_record"}},"created_at":"2026-06-02T02:04:51.974217+00:00","updated_at":"2026-06-02T02:04:51.974217+00:00"}