{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Y7PEQKMHKLOCTMP33G6GWIUHBI","short_pith_number":"pith:Y7PEQKMH","schema_version":"1.0","canonical_sha256":"c7de48298752dc29b1fbd9bc6b22870a2241731ff3b1747595d16029d522d007","source":{"kind":"arxiv","id":"2601.21996","version":2},"attestation_state":"computed","paper":{"title":"Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jianhui Chen, Liangming Pan, Yuzhang Luo","submitted_at":"2026-01-29T17:06:54Z","abstract_excerpt":"While Mechanistic Interpretability has identified interpretable circuits in LLMs, their causal origins in training data remain elusive. We introduce Mechanistic Data Attribution (MDA), a scalable framework that employs Influence Functions to trace interpretable units back to specific training samples. Through extensive experiments on the Pythia family, we causally validate that targeted intervention--removing or augmenting a small fraction of high-influence samples--significantly modulates the emergence of interpretable heads, whereas random interventions show no effect. Our analysis reveals t"},"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":"2601.21996","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-01-29T17:06:54Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"6cd57045873082b291747d2fee2c1e62415bcb4957e503ed982cbb0ef2819a2b","abstract_canon_sha256":"5083f84523e83898d81be11d187eeeea0a18d59f97c94b3094ea835812021a2c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:05:13.442606Z","signature_b64":"XBH9hE+eJT24gGxERyIU087EeVqAGDFxSprbsQ/rSqPGUdZQspAanPrm3tp2C+QYEo9N4w+tfT5i94nH9GakBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c7de48298752dc29b1fbd9bc6b22870a2241731ff3b1747595d16029d522d007","last_reissued_at":"2026-06-09T01:05:13.442071Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:05:13.442071Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jianhui Chen, Liangming Pan, Yuzhang Luo","submitted_at":"2026-01-29T17:06:54Z","abstract_excerpt":"While Mechanistic Interpretability has identified interpretable circuits in LLMs, their causal origins in training data remain elusive. We introduce Mechanistic Data Attribution (MDA), a scalable framework that employs Influence Functions to trace interpretable units back to specific training samples. Through extensive experiments on the Pythia family, we causally validate that targeted intervention--removing or augmenting a small fraction of high-influence samples--significantly modulates the emergence of interpretable heads, whereas random interventions show no effect. Our analysis reveals t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.21996","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/2601.21996/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":"2601.21996","created_at":"2026-06-09T01:05:13.442133+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.21996v2","created_at":"2026-06-09T01:05:13.442133+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.21996","created_at":"2026-06-09T01:05:13.442133+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y7PEQKMHKLOC","created_at":"2026-06-09T01:05:13.442133+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y7PEQKMHKLOCTMP3","created_at":"2026-06-09T01:05:13.442133+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y7PEQKMH","created_at":"2026-06-09T01:05:13.442133+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/Y7PEQKMHKLOCTMP33G6GWIUHBI","json":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI.json","graph_json":"https://pith.science/api/pith-number/Y7PEQKMHKLOCTMP33G6GWIUHBI/graph.json","events_json":"https://pith.science/api/pith-number/Y7PEQKMHKLOCTMP33G6GWIUHBI/events.json","paper":"https://pith.science/paper/Y7PEQKMH"},"agent_actions":{"view_html":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI","download_json":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI.json","view_paper":"https://pith.science/paper/Y7PEQKMH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.21996&json=true","fetch_graph":"https://pith.science/api/pith-number/Y7PEQKMHKLOCTMP33G6GWIUHBI/graph.json","fetch_events":"https://pith.science/api/pith-number/Y7PEQKMHKLOCTMP33G6GWIUHBI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI/action/storage_attestation","attest_author":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI/action/author_attestation","sign_citation":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI/action/citation_signature","submit_replication":"https://pith.science/pith/Y7PEQKMHKLOCTMP33G6GWIUHBI/action/replication_record"}},"created_at":"2026-06-09T01:05:13.442133+00:00","updated_at":"2026-06-09T01:05:13.442133+00:00"}