{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:B3YY2JBMPGGY62OTQOJ2EEOQHU","short_pith_number":"pith:B3YY2JBM","schema_version":"1.0","canonical_sha256":"0ef18d242c798d8f69d38393a211d03d22108554a22ac5690617ba71958d4c34","source":{"kind":"arxiv","id":"2606.24952","version":1},"attestation_state":"computed","paper":{"title":"Perfect Detection, Failed Control: The Geometry of Knowing vs. Steering in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Anna Ettorre, Cosimo Galeone, Daniele Ligorio, Giuseppe Ettorre, Minsu Park","submitted_at":"2026-06-23T08:07:51Z","abstract_excerpt":"A central aspiration of mechanistic interpretability is controllability: if we know where a behavior is represented in a model's activations, we should be able to modify it. This rests on a hidden premise -- that the direction which detects a behavior and the direction which controls it are the same, or close. We test this geometrically: what is the angle between the direction that best detects a behavior and the one that best causes it? If detection implies control the cosine is near 1; otherwise it quantifies a detection-intervention gap. On Gemma 2-2B-it, output format (clean JSON vs markdo"},"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":"2606.24952","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-06-23T08:07:51Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"5e4c364d0090230b3349ea51fc04f968069cc65ab28a5de1ea73dfd38e8ad30c","abstract_canon_sha256":"a921a78cc43c40dbe8df6087cff22e67fa4186c3be047a1f327e31ba1bf8650e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-25T00:17:46.744219Z","signature_b64":"524qhuTCj3NxcCA8sPQCCUu6h51XabugU1oDXnVLhmuTWGpiQmVX1cAf5JTGSCJevv56u5xKWRKGaFqtcFBICA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0ef18d242c798d8f69d38393a211d03d22108554a22ac5690617ba71958d4c34","last_reissued_at":"2026-06-25T00:17:46.743875Z","signature_status":"signed_v1","first_computed_at":"2026-06-25T00:17:46.743875Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Perfect Detection, Failed Control: The Geometry of Knowing vs. Steering in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Anna Ettorre, Cosimo Galeone, Daniele Ligorio, Giuseppe Ettorre, Minsu Park","submitted_at":"2026-06-23T08:07:51Z","abstract_excerpt":"A central aspiration of mechanistic interpretability is controllability: if we know where a behavior is represented in a model's activations, we should be able to modify it. This rests on a hidden premise -- that the direction which detects a behavior and the direction which controls it are the same, or close. We test this geometrically: what is the angle between the direction that best detects a behavior and the one that best causes it? If detection implies control the cosine is near 1; otherwise it quantifies a detection-intervention gap. On Gemma 2-2B-it, output format (clean JSON vs markdo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.24952","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/2606.24952/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":"2606.24952","created_at":"2026-06-25T00:17:46.743937+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.24952v1","created_at":"2026-06-25T00:17:46.743937+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.24952","created_at":"2026-06-25T00:17:46.743937+00:00"},{"alias_kind":"pith_short_12","alias_value":"B3YY2JBMPGGY","created_at":"2026-06-25T00:17:46.743937+00:00"},{"alias_kind":"pith_short_16","alias_value":"B3YY2JBMPGGY62OT","created_at":"2026-06-25T00:17:46.743937+00:00"},{"alias_kind":"pith_short_8","alias_value":"B3YY2JBM","created_at":"2026-06-25T00:17:46.743937+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/B3YY2JBMPGGY62OTQOJ2EEOQHU","json":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU.json","graph_json":"https://pith.science/api/pith-number/B3YY2JBMPGGY62OTQOJ2EEOQHU/graph.json","events_json":"https://pith.science/api/pith-number/B3YY2JBMPGGY62OTQOJ2EEOQHU/events.json","paper":"https://pith.science/paper/B3YY2JBM"},"agent_actions":{"view_html":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU","download_json":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU.json","view_paper":"https://pith.science/paper/B3YY2JBM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.24952&json=true","fetch_graph":"https://pith.science/api/pith-number/B3YY2JBMPGGY62OTQOJ2EEOQHU/graph.json","fetch_events":"https://pith.science/api/pith-number/B3YY2JBMPGGY62OTQOJ2EEOQHU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU/action/storage_attestation","attest_author":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU/action/author_attestation","sign_citation":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU/action/citation_signature","submit_replication":"https://pith.science/pith/B3YY2JBMPGGY62OTQOJ2EEOQHU/action/replication_record"}},"created_at":"2026-06-25T00:17:46.743937+00:00","updated_at":"2026-06-25T00:17:46.743937+00:00"}