{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YCMUDDJB3ZALOGGCKZY3WMO74D","short_pith_number":"pith:YCMUDDJB","schema_version":"1.0","canonical_sha256":"c099418d21de40b718c25671bb31dfe0da8d8f8496bbb4c4e71478a0f6408dbc","source":{"kind":"arxiv","id":"2607.19386","version":1},"attestation_state":"computed","paper":{"title":"Building Fast, Evaluating Slow: Pipeline Choices Dominate Autointerpretability Score Variance","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Anna Hedstr\\\"om, Mennatallah El-Assady, Neele Roch, Sinie van der Ben","submitted_at":"2026-07-01T13:18:12Z","abstract_excerpt":"Cross-paper comparison of sparse autoencoder (SAE) interpretability often relies on autointerpretability scores. In this evaluation pipeline, a language model (LM) explains each feature, and another LM scores the explanation. For these comparisons to be meaningful, scores must reflect stable properties of the features rather than confounding aspects of the evaluation pipeline. Through systematic experiments across four metrics (simulation, detection, fuzzing, purity), two models (Pythia-160M, Apertus-8B), and four axes of methodological variation, we show that this assumption does not hold. Sp"},"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":"2607.19386","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-01T13:18:12Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d41c79b322387f95649b1716fec070d3c135496fb8151eeb5afb5448de228f87","abstract_canon_sha256":"b8771b8bf8f475d4b51e695a62881eec254851304ef1e10a5a64b77b5fc2f7d1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-23T00:23:47.434334Z","signature_b64":"4sXxM4xkzI0LzjpbEamxMYZBBaWAaxO8OwspLBlhUY7OCZSAygv5K8nN6CTeAhkKoMtjAfDAOApSVGdol9wzDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c099418d21de40b718c25671bb31dfe0da8d8f8496bbb4c4e71478a0f6408dbc","last_reissued_at":"2026-07-23T00:23:47.433444Z","signature_status":"signed_v1","first_computed_at":"2026-07-23T00:23:47.433444Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Building Fast, Evaluating Slow: Pipeline Choices Dominate Autointerpretability Score Variance","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Anna Hedstr\\\"om, Mennatallah El-Assady, Neele Roch, Sinie van der Ben","submitted_at":"2026-07-01T13:18:12Z","abstract_excerpt":"Cross-paper comparison of sparse autoencoder (SAE) interpretability often relies on autointerpretability scores. In this evaluation pipeline, a language model (LM) explains each feature, and another LM scores the explanation. For these comparisons to be meaningful, scores must reflect stable properties of the features rather than confounding aspects of the evaluation pipeline. Through systematic experiments across four metrics (simulation, detection, fuzzing, purity), two models (Pythia-160M, Apertus-8B), and four axes of methodological variation, we show that this assumption does not hold. Sp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.19386","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/2607.19386/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":"2607.19386","created_at":"2026-07-23T00:23:47.433903+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.19386v1","created_at":"2026-07-23T00:23:47.433903+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.19386","created_at":"2026-07-23T00:23:47.433903+00:00"},{"alias_kind":"pith_short_12","alias_value":"YCMUDDJB3ZAL","created_at":"2026-07-23T00:23:47.433903+00:00"},{"alias_kind":"pith_short_16","alias_value":"YCMUDDJB3ZALOGGC","created_at":"2026-07-23T00:23:47.433903+00:00"},{"alias_kind":"pith_short_8","alias_value":"YCMUDDJB","created_at":"2026-07-23T00:23:47.433903+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/YCMUDDJB3ZALOGGCKZY3WMO74D","json":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D.json","graph_json":"https://pith.science/api/pith-number/YCMUDDJB3ZALOGGCKZY3WMO74D/graph.json","events_json":"https://pith.science/api/pith-number/YCMUDDJB3ZALOGGCKZY3WMO74D/events.json","paper":"https://pith.science/paper/YCMUDDJB"},"agent_actions":{"view_html":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D","download_json":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D.json","view_paper":"https://pith.science/paper/YCMUDDJB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.19386&json=true","fetch_graph":"https://pith.science/api/pith-number/YCMUDDJB3ZALOGGCKZY3WMO74D/graph.json","fetch_events":"https://pith.science/api/pith-number/YCMUDDJB3ZALOGGCKZY3WMO74D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D/action/storage_attestation","attest_author":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D/action/author_attestation","sign_citation":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D/action/citation_signature","submit_replication":"https://pith.science/pith/YCMUDDJB3ZALOGGCKZY3WMO74D/action/replication_record"}},"created_at":"2026-07-23T00:23:47.433903+00:00","updated_at":"2026-07-23T00:23:47.433903+00:00"}