{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:TQGFFC7FBMDWN7FR7UPF6PYNPS","short_pith_number":"pith:TQGFFC7F","schema_version":"1.0","canonical_sha256":"9c0c528be50b0766fcb1fd1e5f3f0d7cb7f9c230ca1d584dbf5b1dc93ed15fff","source":{"kind":"arxiv","id":"2607.27836","version":1},"attestation_state":"computed","paper":{"title":"Crossing the Margin Cliff: Toward Relearn-Robust LLM Unlearning via Margin Calibration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Chih-Hong Cheng, Jiaxu Liu, Xiangyu Yin, Zhen Chen","submitted_at":"2026-07-30T08:15:58Z","abstract_excerpt":"Large language model unlearning is consistently fragile under relearn attacks. On TOFU, fine-tuning on twenty forget examples substantially recovers held-out forget-set ROUGE for every method we evaluate, and we trace this fragility to optimization geometry. The per-token answer margin of fourteen post-hoc methods spanning gradient, preference, and distillation families converges into a narrow band above the retain reference in 41 of 42 method--size cells, a regularity we call the margin cliff. We prove that this cliff follows whenever the retain coupling holds the diagnostic log-odds of forge"},"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":"2607.27836","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-07-30T08:15:58Z","cross_cats_sorted":[],"title_canon_sha256":"658895d417f1e7cb1f3a29e2fa40716fbe9604b4d53275734e8f5f94a5e205a2","abstract_canon_sha256":"4d0334eed805c6689e86eb8cb728f9e428b151207413b42029f1c724950a8e83"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9c0c528be50b0766fcb1fd1e5f3f0d7cb7f9c230ca1d584dbf5b1dc93ed15fff","last_reissued_at":"2026-07-31T01:34:26.480253Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-31T01:34:26.480253Z"},"graph_snapshot":{"paper":{"title":"Crossing the Margin Cliff: Toward Relearn-Robust LLM Unlearning via Margin Calibration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Chih-Hong Cheng, Jiaxu Liu, Xiangyu Yin, Zhen Chen","submitted_at":"2026-07-30T08:15:58Z","abstract_excerpt":"Large language model unlearning is consistently fragile under relearn attacks. On TOFU, fine-tuning on twenty forget examples substantially recovers held-out forget-set ROUGE for every method we evaluate, and we trace this fragility to optimization geometry. The per-token answer margin of fourteen post-hoc methods spanning gradient, preference, and distillation families converges into a narrow band above the retain reference in 41 of 42 method--size cells, a regularity we call the margin cliff. We prove that this cliff follows whenever the retain coupling holds the diagnostic log-odds of forge"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.27836","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/2607.27836/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":"2607.27836","created_at":"2026-07-31T01:34:26.483326+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.27836v1","created_at":"2026-07-31T01:34:26.483326+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.27836","created_at":"2026-07-31T01:34:26.483326+00:00"},{"alias_kind":"pith_short_12","alias_value":"TQGFFC7FBMDW","created_at":"2026-07-31T01:34:26.483326+00:00"},{"alias_kind":"pith_short_16","alias_value":"TQGFFC7FBMDWN7FR","created_at":"2026-07-31T01:34:26.483326+00:00"},{"alias_kind":"pith_short_8","alias_value":"TQGFFC7F","created_at":"2026-07-31T01:34:26.483326+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/TQGFFC7FBMDWN7FR7UPF6PYNPS","json":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS.json","graph_json":"https://pith.science/api/pith-number/TQGFFC7FBMDWN7FR7UPF6PYNPS/graph.json","events_json":"https://pith.science/api/pith-number/TQGFFC7FBMDWN7FR7UPF6PYNPS/events.json","paper":"https://pith.science/paper/TQGFFC7F"},"agent_actions":{"view_html":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS","download_json":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS.json","view_paper":"https://pith.science/paper/TQGFFC7F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.27836&json=true","fetch_graph":"https://pith.science/api/pith-number/TQGFFC7FBMDWN7FR7UPF6PYNPS/graph.json","fetch_events":"https://pith.science/api/pith-number/TQGFFC7FBMDWN7FR7UPF6PYNPS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS/action/storage_attestation","attest_author":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS/action/author_attestation","sign_citation":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS/action/citation_signature","submit_replication":"https://pith.science/pith/TQGFFC7FBMDWN7FR7UPF6PYNPS/action/replication_record"}},"created_at":"2026-07-31T01:34:26.483326+00:00","updated_at":"2026-07-31T01:34:26.483326+00:00"}