{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:J2K3B3F3SQNZSZFVAOYIMOJPJA","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"9e8ab8154d64ce00033f8d706e7890b9ee54f12ffc2ec05dde6aa524e46dc405","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-19T18:40:51Z","title_canon_sha256":"259cc95fc978abbdd3662a1cb4970ffc0263b69bf1ef2e0533ccbe93296a0b31"},"schema_version":"1.0","source":{"id":"2504.14379","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.14379","created_at":"2026-07-05T11:01:35Z"},{"alias_kind":"arxiv_version","alias_value":"2504.14379v2","created_at":"2026-07-05T11:01:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.14379","created_at":"2026-07-05T11:01:35Z"},{"alias_kind":"pith_short_12","alias_value":"J2K3B3F3SQNZ","created_at":"2026-07-05T11:01:35Z"},{"alias_kind":"pith_short_16","alias_value":"J2K3B3F3SQNZSZFV","created_at":"2026-07-05T11:01:35Z"},{"alias_kind":"pith_short_8","alias_value":"J2K3B3F3","created_at":"2026-07-05T11:01:35Z"}],"graph_snapshots":[{"event_id":"sha256:34e842f81dcac996627f0e5fab3acb7530f5860534705975d2235bd15a595242","target":"graph","created_at":"2026-07-05T11:01:35Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2504.14379/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"How do reasoning models verify their own answers? We study this question by training a model using DeepSeek R1's recipe on the CountDown task. We leverage the fact that preference tuning leads to mode collapse, yielding a model that always produces highly structured chain-of-thought sequences. With this setup, we do top-down and bottom-up analyses to reverse-engineer how the model verifies its outputs. Top-down, we find Gated Linear Unit (GLU) weights encoding verification-related tokens, such as ``success'' or ``incorrect''. Bottom-up, we find that ``previous-token heads'' are mainly responsi","authors_text":"Andrew Lee, Chris Wendler, Fernanda Vi\\'egas, Lihao Sun, Martin Wattenberg","cross_cats":["cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-19T18:40:51Z","title":"The Geometry of Self-Verification in a Task-Specific Reasoning Model"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.14379","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:63a31158c13ce6169a9c4cd5fb5c982654ab45364e658e200dfc09262afcb40a","target":"record","created_at":"2026-07-05T11:01:35Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"9e8ab8154d64ce00033f8d706e7890b9ee54f12ffc2ec05dde6aa524e46dc405","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-19T18:40:51Z","title_canon_sha256":"259cc95fc978abbdd3662a1cb4970ffc0263b69bf1ef2e0533ccbe93296a0b31"},"schema_version":"1.0","source":{"id":"2504.14379","kind":"arxiv","version":2}},"canonical_sha256":"4e95b0ecbb941b9964b503b086392f48276027e448df46ffeba3e7ce3e242e05","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"4e95b0ecbb941b9964b503b086392f48276027e448df46ffeba3e7ce3e242e05","first_computed_at":"2026-07-05T11:01:35.206445Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:01:35.206445Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"2OGvHgnZu3WLcGCA9mfvJEOyjnj8PzGhxnxJHVwfe5u8SmSv+4VxVp5eE4KSesfTS/lcUx3zAEwhH+QR+HB4Cw==","signature_status":"signed_v1","signed_at":"2026-07-05T11:01:35.207859Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.14379","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:63a31158c13ce6169a9c4cd5fb5c982654ab45364e658e200dfc09262afcb40a","sha256:34e842f81dcac996627f0e5fab3acb7530f5860534705975d2235bd15a595242"],"state_sha256":"eabe70b98bf9dfff4d4cbcc3be94a8c030a97fd610a7a7635a34f7dabeaf245a"}