{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IBP2FU7PMQQLLYZ53QLDTWK576","short_pith_number":"pith:IBP2FU7P","schema_version":"1.0","canonical_sha256":"405fa2d3ef6420b5e33ddc1639d95dffb7240f6db0efd5c1efa8de177ead74e3","source":{"kind":"arxiv","id":"2607.04640","version":1},"attestation_state":"computed","paper":{"title":"Wrong Before Right: Late Rescue and Interface Failure in Aligned Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jiaqi Deng","submitted_at":"2026-07-06T03:51:24Z","abstract_excerpt":"We study how correctness is assembled inside aligned language models, not only whether the final answer is right. Using layer-wise difference-in-differences (DiD) trajectories over polarity-controlled minimal pairs, we identify the wrong-dip: in mid layers (25-90% depth), internal preference transiently commits to the incorrect answer and is rescued only by late-layer correction. We verify this causally with patchscope-style activation transplantation across 17 models, three families, and 64x scale (0.5B-32B). Four findings follow. (1) Alignment amplification of the causal wrong-dip is recipe-"},"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":"2607.04640","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-07-06T03:51:24Z","cross_cats_sorted":[],"title_canon_sha256":"1e05dd9d6850bd2fefc6dbd6c9e658b71fbe964f25be8e557452f8037cb41906","abstract_canon_sha256":"baa9097be5e017aa2b4dfb923e44ccb3cd246a986c662d9f7069a84a942393d9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-07T02:19:29.046246Z","signature_b64":"z9nuUrdCIUAOFgSxfe+TlLPQS/dngKSwE29L0nUn8D/aTEy9iacunIALR2NcbtW4/qooYNEkAqRdDxvUnZnmAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"405fa2d3ef6420b5e33ddc1639d95dffb7240f6db0efd5c1efa8de177ead74e3","last_reissued_at":"2026-07-07T02:19:29.045572Z","signature_status":"signed_v1","first_computed_at":"2026-07-07T02:19:29.045572Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Wrong Before Right: Late Rescue and Interface Failure in Aligned Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jiaqi Deng","submitted_at":"2026-07-06T03:51:24Z","abstract_excerpt":"We study how correctness is assembled inside aligned language models, not only whether the final answer is right. Using layer-wise difference-in-differences (DiD) trajectories over polarity-controlled minimal pairs, we identify the wrong-dip: in mid layers (25-90% depth), internal preference transiently commits to the incorrect answer and is rescued only by late-layer correction. We verify this causally with patchscope-style activation transplantation across 17 models, three families, and 64x scale (0.5B-32B). Four findings follow. (1) Alignment amplification of the causal wrong-dip is recipe-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.04640","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/2607.04640/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":"2607.04640","created_at":"2026-07-07T02:19:29.045678+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.04640v1","created_at":"2026-07-07T02:19:29.045678+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.04640","created_at":"2026-07-07T02:19:29.045678+00:00"},{"alias_kind":"pith_short_12","alias_value":"IBP2FU7PMQQL","created_at":"2026-07-07T02:19:29.045678+00:00"},{"alias_kind":"pith_short_16","alias_value":"IBP2FU7PMQQLLYZ5","created_at":"2026-07-07T02:19:29.045678+00:00"},{"alias_kind":"pith_short_8","alias_value":"IBP2FU7P","created_at":"2026-07-07T02:19:29.045678+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/IBP2FU7PMQQLLYZ53QLDTWK576","json":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576.json","graph_json":"https://pith.science/api/pith-number/IBP2FU7PMQQLLYZ53QLDTWK576/graph.json","events_json":"https://pith.science/api/pith-number/IBP2FU7PMQQLLYZ53QLDTWK576/events.json","paper":"https://pith.science/paper/IBP2FU7P"},"agent_actions":{"view_html":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576","download_json":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576.json","view_paper":"https://pith.science/paper/IBP2FU7P","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.04640&json=true","fetch_graph":"https://pith.science/api/pith-number/IBP2FU7PMQQLLYZ53QLDTWK576/graph.json","fetch_events":"https://pith.science/api/pith-number/IBP2FU7PMQQLLYZ53QLDTWK576/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576/action/storage_attestation","attest_author":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576/action/author_attestation","sign_citation":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576/action/citation_signature","submit_replication":"https://pith.science/pith/IBP2FU7PMQQLLYZ53QLDTWK576/action/replication_record"}},"created_at":"2026-07-07T02:19:29.045678+00:00","updated_at":"2026-07-07T02:19:29.045678+00:00"}