{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:RVQQBIVEQPN2MZGTSG52AHCTNJ","short_pith_number":"pith:RVQQBIVE","schema_version":"1.0","canonical_sha256":"8d6100a2a483dba664d391bba01c536a4cb8edcfb39b5ba1a038e84f4c888eb4","source":{"kind":"arxiv","id":"2606.18383","version":1},"attestation_state":"computed","paper":{"title":"From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Asif Ekbal, Dibyanayan Bandyopadhyay","submitted_at":"2026-06-16T18:28:23Z","abstract_excerpt":"Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction. Our framework derives an upper bound on the base model's expected risk using four measurable quantities: proxy risk, SAE reconstruction gap, concept-pool mismatch, and sparse co"},"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":"2606.18383","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-16T18:28:23Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"f1e6b9738503f4f9bc3fcf99b1901a8637bd7a3d49600c0973728bbbf3dd4adf","abstract_canon_sha256":"eea988d9acc4c29d0e0812158cec10618cffecfeaedcee08b45feaaf8585b854"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:10:59.833968Z","signature_b64":"xkBoj4Wdk4my9Uxup3/embB61FnLkBQ+q1EsYM0RNUVZ5Ooc/tl+kup10lLNCtryvgijrNn1eGDLHxJjnQmDCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8d6100a2a483dba664d391bba01c536a4cb8edcfb39b5ba1a038e84f4c888eb4","last_reissued_at":"2026-06-19T16:10:59.833641Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:10:59.833641Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Asif Ekbal, Dibyanayan Bandyopadhyay","submitted_at":"2026-06-16T18:28:23Z","abstract_excerpt":"Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction. Our framework derives an upper bound on the base model's expected risk using four measurable quantities: proxy risk, SAE reconstruction gap, concept-pool mismatch, and sparse co"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.18383","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/2606.18383/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":"2606.18383","created_at":"2026-06-19T16:10:59.833695+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.18383v1","created_at":"2026-06-19T16:10:59.833695+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.18383","created_at":"2026-06-19T16:10:59.833695+00:00"},{"alias_kind":"pith_short_12","alias_value":"RVQQBIVEQPN2","created_at":"2026-06-19T16:10:59.833695+00:00"},{"alias_kind":"pith_short_16","alias_value":"RVQQBIVEQPN2MZGT","created_at":"2026-06-19T16:10:59.833695+00:00"},{"alias_kind":"pith_short_8","alias_value":"RVQQBIVE","created_at":"2026-06-19T16:10:59.833695+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/RVQQBIVEQPN2MZGTSG52AHCTNJ","json":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ.json","graph_json":"https://pith.science/api/pith-number/RVQQBIVEQPN2MZGTSG52AHCTNJ/graph.json","events_json":"https://pith.science/api/pith-number/RVQQBIVEQPN2MZGTSG52AHCTNJ/events.json","paper":"https://pith.science/paper/RVQQBIVE"},"agent_actions":{"view_html":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ","download_json":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ.json","view_paper":"https://pith.science/paper/RVQQBIVE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.18383&json=true","fetch_graph":"https://pith.science/api/pith-number/RVQQBIVEQPN2MZGTSG52AHCTNJ/graph.json","fetch_events":"https://pith.science/api/pith-number/RVQQBIVEQPN2MZGTSG52AHCTNJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ/action/storage_attestation","attest_author":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ/action/author_attestation","sign_citation":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ/action/citation_signature","submit_replication":"https://pith.science/pith/RVQQBIVEQPN2MZGTSG52AHCTNJ/action/replication_record"}},"created_at":"2026-06-19T16:10:59.833695+00:00","updated_at":"2026-06-19T16:10:59.833695+00:00"}