{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WRHEEGTTVNQBHXMBLPVYOFYNTO","short_pith_number":"pith:WRHEEGTT","schema_version":"1.0","canonical_sha256":"b44e421a73ab6013dd815beb87170d9ba90276c60524683df77fe51f7f8765e1","source":{"kind":"arxiv","id":"2502.20914","version":1},"attestation_state":"computed","paper":{"title":"Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Fran\\c{c}ois Portet, Maxime M\\'eloux, Maxime Peyrard, Silviu Maniu","submitted_at":"2025-02-28T10:13:54Z","abstract_excerpt":"As AI systems are used in high-stakes applications, ensuring interpretability is crucial. Mechanistic Interpretability (MI) aims to reverse-engineer neural networks by extracting human-understandable algorithms to explain their behavior. This work examines a key question: for a given behavior, and under MI's criteria, does a unique explanation exist? Drawing on identifiability in statistics, where parameters are uniquely inferred under specific assumptions, we explore the identifiability of MI explanations.\n  We identify two main MI strategies: (1) \"where-then-what,\" which isolates a circuit r"},"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":"2502.20914","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-28T10:13:54Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"b270e741105cb019167a6029b271360e69c1c710f96efc384acccb10fccb505c","abstract_canon_sha256":"7d912f7ee18a01073219793e6fec7ee74884b49d6ef5a360c454eee73e81bfb6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:14:25.344743Z","signature_b64":"LPiw4U2RCqaSstqNDh56I1IElijEflMv5A4UoKumuAsoutefD73Ul69kxxIXQUFeyWPvQwa3OgtkgVTibYPGDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b44e421a73ab6013dd815beb87170d9ba90276c60524683df77fe51f7f8765e1","last_reissued_at":"2026-06-05T01:14:25.343837Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:14:25.343837Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Fran\\c{c}ois Portet, Maxime M\\'eloux, Maxime Peyrard, Silviu Maniu","submitted_at":"2025-02-28T10:13:54Z","abstract_excerpt":"As AI systems are used in high-stakes applications, ensuring interpretability is crucial. Mechanistic Interpretability (MI) aims to reverse-engineer neural networks by extracting human-understandable algorithms to explain their behavior. This work examines a key question: for a given behavior, and under MI's criteria, does a unique explanation exist? Drawing on identifiability in statistics, where parameters are uniquely inferred under specific assumptions, we explore the identifiability of MI explanations.\n  We identify two main MI strategies: (1) \"where-then-what,\" which isolates a circuit r"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.20914","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/2502.20914/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":"2502.20914","created_at":"2026-06-05T01:14:25.343960+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.20914v1","created_at":"2026-06-05T01:14:25.343960+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.20914","created_at":"2026-06-05T01:14:25.343960+00:00"},{"alias_kind":"pith_short_12","alias_value":"WRHEEGTTVNQB","created_at":"2026-06-05T01:14:25.343960+00:00"},{"alias_kind":"pith_short_16","alias_value":"WRHEEGTTVNQBHXMB","created_at":"2026-06-05T01:14:25.343960+00:00"},{"alias_kind":"pith_short_8","alias_value":"WRHEEGTT","created_at":"2026-06-05T01:14:25.343960+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.08934","citing_title":"From Mechanistic to Compositional Interpretability","ref_index":51,"is_internal_anchor":true},{"citing_arxiv_id":"2605.02234","citing_title":"Bucketing the Good Apples: A Method for Diagnosing and Improving Causal Abstraction","ref_index":15,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO","json":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO.json","graph_json":"https://pith.science/api/pith-number/WRHEEGTTVNQBHXMBLPVYOFYNTO/graph.json","events_json":"https://pith.science/api/pith-number/WRHEEGTTVNQBHXMBLPVYOFYNTO/events.json","paper":"https://pith.science/paper/WRHEEGTT"},"agent_actions":{"view_html":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO","download_json":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO.json","view_paper":"https://pith.science/paper/WRHEEGTT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.20914&json=true","fetch_graph":"https://pith.science/api/pith-number/WRHEEGTTVNQBHXMBLPVYOFYNTO/graph.json","fetch_events":"https://pith.science/api/pith-number/WRHEEGTTVNQBHXMBLPVYOFYNTO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO/action/storage_attestation","attest_author":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO/action/author_attestation","sign_citation":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO/action/citation_signature","submit_replication":"https://pith.science/pith/WRHEEGTTVNQBHXMBLPVYOFYNTO/action/replication_record"}},"created_at":"2026-06-05T01:14:25.343960+00:00","updated_at":"2026-06-05T01:14:25.343960+00:00"}