{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZXPMSMOIGCP5NLFGWQQB53FJTU","short_pith_number":"pith:ZXPMSMOI","schema_version":"1.0","canonical_sha256":"cddec931c8309fd6aca6b4201eeca99d28fecdc4b7e3d30cc7b1bad8453202eb","source":{"kind":"arxiv","id":"2605.21127","version":1},"attestation_state":"computed","paper":{"title":"Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Helen Yannakoudakis, Jie M. Zhang, Lukas Twist","submitted_at":"2026-05-20T12:58:01Z","abstract_excerpt":"Explicit reasoning models are trained to produce intermediate reasoning traces before final answers, but downstream fine-tuning is often performed on ordinary instruction-response data that contains no such traces. We show that this mismatch can induce reasoning-trace collapse: a fine-tuned model continues to produce plausible final answers while losing the structurally valid explicit reasoning traces that made it a reasoning model in the first place. We introduce a structural evaluation framework that separates answer correctness from reasoning-trace validity, measuring valid, empty, missing,"},"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":"2605.21127","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T12:58:01Z","cross_cats_sorted":[],"title_canon_sha256":"6b8b507880695659dc70b987bbb10e272b97b2957da5355f609031a4cc07db30","abstract_canon_sha256":"a773d3922875b7c538c17516eef4df447341c21ffe3bb57f099bb552e6450abd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:05:38.851531Z","signature_b64":"WJMnkynjF7A7MiA/Wl4ywQMKOw8Hh7fYn+4o6F92cdv7qVr3703fW4Ovk6H4FiFWAJY3LqxKX0Yl7g2hju93DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cddec931c8309fd6aca6b4201eeca99d28fecdc4b7e3d30cc7b1bad8453202eb","last_reissued_at":"2026-05-21T01:05:38.850790Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:05:38.850790Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Helen Yannakoudakis, Jie M. Zhang, Lukas Twist","submitted_at":"2026-05-20T12:58:01Z","abstract_excerpt":"Explicit reasoning models are trained to produce intermediate reasoning traces before final answers, but downstream fine-tuning is often performed on ordinary instruction-response data that contains no such traces. We show that this mismatch can induce reasoning-trace collapse: a fine-tuned model continues to produce plausible final answers while losing the structurally valid explicit reasoning traces that made it a reasoning model in the first place. We introduce a structural evaluation framework that separates answer correctness from reasoning-trace validity, measuring valid, empty, missing,"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.21127","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/2605.21127/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":"2605.21127","created_at":"2026-05-21T01:05:38.850913+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.21127v1","created_at":"2026-05-21T01:05:38.850913+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.21127","created_at":"2026-05-21T01:05:38.850913+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZXPMSMOIGCP5","created_at":"2026-05-21T01:05:38.850913+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZXPMSMOIGCP5NLFG","created_at":"2026-05-21T01:05:38.850913+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZXPMSMOI","created_at":"2026-05-21T01:05:38.850913+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/ZXPMSMOIGCP5NLFGWQQB53FJTU","json":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU.json","graph_json":"https://pith.science/api/pith-number/ZXPMSMOIGCP5NLFGWQQB53FJTU/graph.json","events_json":"https://pith.science/api/pith-number/ZXPMSMOIGCP5NLFGWQQB53FJTU/events.json","paper":"https://pith.science/paper/ZXPMSMOI"},"agent_actions":{"view_html":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU","download_json":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU.json","view_paper":"https://pith.science/paper/ZXPMSMOI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.21127&json=true","fetch_graph":"https://pith.science/api/pith-number/ZXPMSMOIGCP5NLFGWQQB53FJTU/graph.json","fetch_events":"https://pith.science/api/pith-number/ZXPMSMOIGCP5NLFGWQQB53FJTU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU/action/storage_attestation","attest_author":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU/action/author_attestation","sign_citation":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU/action/citation_signature","submit_replication":"https://pith.science/pith/ZXPMSMOIGCP5NLFGWQQB53FJTU/action/replication_record"}},"created_at":"2026-05-21T01:05:38.850913+00:00","updated_at":"2026-05-21T01:05:38.850913+00:00"}