{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:3RES4GZVLQLXB7VYV6PPRKXDTJ","short_pith_number":"pith:3RES4GZV","canonical_record":{"source":{"id":"2605.16874","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-16T08:33:31Z","cross_cats_sorted":[],"title_canon_sha256":"c0004652b2b530242bee41cf3e561ad9c0833ae09b31c751047edb1048d36ad9","abstract_canon_sha256":"3eb0e59a986b0dd21b782a2e73c6c5252b1f6a3e0f75243f097c982a4f0a5045"},"schema_version":"1.0"},"canonical_sha256":"dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272","source":{"kind":"arxiv","id":"2605.16874","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.16874","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"arxiv_version","alias_value":"2605.16874v1","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16874","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"pith_short_12","alias_value":"3RES4GZVLQLX","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"pith_short_16","alias_value":"3RES4GZVLQLXB7VY","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"pith_short_8","alias_value":"3RES4GZV","created_at":"2026-05-20T00:03:27Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:3RES4GZVLQLXB7VYV6PPRKXDTJ","target":"record","payload":{"canonical_record":{"source":{"id":"2605.16874","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-16T08:33:31Z","cross_cats_sorted":[],"title_canon_sha256":"c0004652b2b530242bee41cf3e561ad9c0833ae09b31c751047edb1048d36ad9","abstract_canon_sha256":"3eb0e59a986b0dd21b782a2e73c6c5252b1f6a3e0f75243f097c982a4f0a5045"},"schema_version":"1.0"},"canonical_sha256":"dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:27.622951Z","signature_b64":"kBM2BEA66weVHZg6QlzOOrb61Aq9Ioq3SuGLM0YiQuZFqv72DbbJ1nlQ1Pf5BMHhXrONZ6dMuOicugEy5OgMDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272","last_reissued_at":"2026-05-20T00:03:27.622281Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:27.622281Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.16874","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:03:27Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"BTFkTecOqpraKTqhbeK0cMUbsuF1vwS2N+7M62kq7UIab1vCNj49rla/rB4rgzq7UUkTESjNvK2xzx9dZyWuDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-20T14:11:31.895430Z"},"content_sha256":"0e1c5c0038891e136a20f2b74b0ac8b29346843b943110f251cd3bbf236da41a","schema_version":"1.0","event_id":"sha256:0e1c5c0038891e136a20f2b74b0ac8b29346843b943110f251cd3bbf236da41a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:3RES4GZVLQLXB7VYV6PPRKXDTJ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Reasoning Can Be Restored by Correcting a Few Decision Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"An Zhang, Changshuo Shen, Leheng Sheng, Xiang Wang, Yuxin Chen","submitted_at":"2026-05-16T08:33:31Z","abstract_excerpt":"Large reasoning models (LRMs) substantially outperform their base LLM counterparts on challenging reasoning benchmarks, yet it remains poorly understood where base models go wrong during token-by-token generation and how to narrow this gap efficiently. We study the base-reasoning gap through quantifying token-level distributional disagreement between a base model and a stronger reasoning model using likelihood-based divergences. Across benchmarks, we find that the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Q"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across benchmarks, the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Qwen3-0.6B, only ~8% of generated tokens account for the salient disagreement, and these tokens concentrate early in the response, are strongly enriched in planning-related decisions (17x), and coincide with high base-model uncertainty.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the positions of high likelihood-based distributional disagreement between base and reasoning models are precisely the causal points where the base model’s early planning errors derail the subsequent reasoning trajectory, rather than merely correlated symptoms.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Reasoning gaps between base LLMs and LRMs concentrate on ~8% of early planning tokens; intervening with the reasoning model only at high-disagreement positions recovers performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"f781aae1e3143eebab73738df5644d515e07334a7b61102f7bf2c0d2c02506dc"},"source":{"id":"2605.16874","kind":"arxiv","version":1},"verdict":{"id":"0fc56498-5930-4705-bf68-ead1ca1ba92b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:52:50.412735Z","strongest_claim":"Across benchmarks, the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Qwen3-0.6B, only ~8% of generated tokens account for the salient disagreement, and these tokens concentrate early in the response, are strongly enriched in planning-related decisions (17x), and coincide with high base-model uncertainty.","one_line_summary":"Reasoning gaps between base LLMs and LRMs concentrate on ~8% of early planning tokens; intervening with the reasoning model only at high-disagreement positions recovers performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the positions of high likelihood-based distributional disagreement between base and reasoning models are precisely the causal points where the base model’s early planning errors derail the subsequent reasoning trajectory, rather than merely correlated symptoms.","pith_extraction_headline":"Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention."},"integrity":{"clean":false,"summary":{"advisory":1,"critical":0,"by_detector":{"doi_compliance":{"total":1,"advisory":1,"critical":0,"informational":0}},"informational":0},"endpoint":"/pith/2605.16874/integrity.json","findings":[{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1109/TIT.2009.2027527.URL) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":10,"audited_at":"2026-05-19T21:01:24.900423Z","detected_doi":"10.1109/TIT.2009.2027527.URL","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T21:01:24.900423Z","status":"completed","version":"1.0.0","findings_count":1},{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.213886Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.295789Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.372472Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"001554fc071bcef2dbad9ab2d4cbe1380ebca1053576159ee2b79183bc7bdd8a"},"references":{"count":33,"sample":[{"doi":"","year":2005,"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","ref_index":1,"cited_arxiv_id":"2005.14165","is_internal_anchor":true},{"doi":"","year":null,"title":"arXiv preprint arXiv:2510.00553 , year=","work_id":"a6fda285-7688-48ed-9537-cbaa641673ef","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Cascade speculative drafting for even faster llm inference,","work_id":"a42682ab-46cc-4a73-823d-d74a2fe116e9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","ref_index":4,"cited_arxiv_id":"2110.14168","is_internal_anchor":true},{"doi":"","year":null,"title":"Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs","work_id":"f65a84bf-c5b4-4491-a618-18bb263c60e5","ref_index":5,"cited_arxiv_id":"2503.01307","is_internal_anchor":true}],"resolved_work":33,"snapshot_sha256":"19f1cf9b040ccd4069233bab69759a120d96a88c0291a629b1ea91f008d01313","internal_anchors":15},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e5de90e08a366df0720a68a411b87a1a69cdf948a4230f53c5f6224a4599c976"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"0fc56498-5930-4705-bf68-ead1ca1ba92b"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:03:27Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"LqQp0WHtpCSeDfN+r1y0T5NUV22vIqV1r+TF1YD+cdWk/BAXHYx1Wb6iAHGonj14ueOPimyX9IH+po9Sm9B9Bg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-20T14:11:31.896242Z"},"content_sha256":"9b05fe1ab56ce8707df507f1eadba26d04df72a04d2cb88ab723acec0022225b","schema_version":"1.0","event_id":"sha256:9b05fe1ab56ce8707df507f1eadba26d04df72a04d2cb88ab723acec0022225b"},{"event_type":"integrity_finding","subject_pith_number":"pith:2026:3RES4GZVLQLXB7VYV6PPRKXDTJ","target":"integrity","payload":{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1109/TIT.2009.2027527.URL) was visible in the surrounding text but could not be confirmed against doi.org as printed.","snippet":"ISSN 0018-9448. doi: 10.1109/TIT.2009. 2027527. URL https://doi.org/10.1109/TIT. 2009.2027527. Jaech, A., Kalai, A., Lerer, A., Richardson, A., El-Kishky, A., Low, A., Helyar, A., Madry, A., Beutel, A., Carney, A., Iftimie, A., Karpenko, A.","arxiv_id":"2605.16874","detector":"doi_compliance","evidence":{"ref_index":10,"verdict_class":"incontrovertible","resolved_title":null,"printed_excerpt":"ISSN 0018-9448. doi: 10.1109/TIT.2009. 2027527. URL https://doi.org/10.1109/TIT. 2009.2027527. Jaech, A., Kalai, A., Lerer, A., Richardson, A., El-Kishky, A., Low, A., Helyar, A., Madry, A., Beutel, A., Carney, A., Iftimie, A., Karpenko, A.","reconstructed_doi":"10.1109/TIT.2009.2027527.URL"},"severity":"advisory","ref_index":10,"audited_at":"2026-05-19T21:01:24.900423Z","event_type":"pith.integrity.v1","detected_doi":"10.1109/TIT.2009.2027527.URL","detector_url":"https://pith.science/pith-integrity-protocol#doi_compliance","external_url":null,"finding_type":"recoverable_identifier","evidence_hash":"281de689c03af764685ed6381650f26def0e937d2078e7312a9ee2277eeabcaa","paper_version":1,"verdict_class":"incontrovertible","resolved_title":null,"detector_version":"1.0.0","detected_arxiv_id":null,"integrity_event_id":3032,"payload_sha256":"de134dd177eef407ef382f213c8af3578d696fe5d3bb20bee45e0b4a5bbc6fba","signature_b64":"YMwzRuvu2OJs8d5vStjFVCb3uXZ48ESmEHGlBOq1/wPCCczOW17oDLncKqXp/HYIMgWoTPvEm2KHK82nk0hTCw==","signing_key_id":"pith-v1-2026-05"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-19T21:02:22Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"7V276hWn5GEHBKh2SAd7i3leVinxBEXed4TcGHYjDHHi9HeL1KMpgVTUwzPLdbI+qnZvx/VZSss+yMue0qjmDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-20T14:11:31.897105Z"},"content_sha256":"49a76f015c4f33e09454a82cf9a27c02a1bab1408e4d679ee1682ff877c11041","schema_version":"1.0","event_id":"sha256:49a76f015c4f33e09454a82cf9a27c02a1bab1408e4d679ee1682ff877c11041"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ/bundle.json","state_url":"https://pith.science/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-20T14:11:31Z","links":{"resolver":"https://pith.science/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ","bundle":"https://pith.science/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ/bundle.json","state":"https://pith.science/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/3RES4GZVLQLXB7VYV6PPRKXDTJ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:3RES4GZVLQLXB7VYV6PPRKXDTJ","merge_version":"pith-open-graph-merge-v1","event_count":3,"valid_event_count":3,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"3eb0e59a986b0dd21b782a2e73c6c5252b1f6a3e0f75243f097c982a4f0a5045","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-16T08:33:31Z","title_canon_sha256":"c0004652b2b530242bee41cf3e561ad9c0833ae09b31c751047edb1048d36ad9"},"schema_version":"1.0","source":{"id":"2605.16874","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.16874","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"arxiv_version","alias_value":"2605.16874v1","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16874","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"pith_short_12","alias_value":"3RES4GZVLQLX","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"pith_short_16","alias_value":"3RES4GZVLQLXB7VY","created_at":"2026-05-20T00:03:27Z"},{"alias_kind":"pith_short_8","alias_value":"3RES4GZV","created_at":"2026-05-20T00:03:27Z"}],"graph_snapshots":[{"event_id":"sha256:9b05fe1ab56ce8707df507f1eadba26d04df72a04d2cb88ab723acec0022225b","target":"graph","created_at":"2026-05-20T00:03:27Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Across benchmarks, the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Qwen3-0.6B, only ~8% of generated tokens account for the salient disagreement, and these tokens concentrate early in the response, are strongly enriched in planning-related decisions (17x), and coincide with high base-model uncertainty."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the positions of high likelihood-based distributional disagreement between base and reasoning models are precisely the causal points where the base model’s early planning errors derail the subsequent reasoning trajectory, rather than merely correlated symptoms."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Reasoning gaps between base LLMs and LRMs concentrate on ~8% of early planning tokens; intervening with the reasoning model only at high-disagreement positions recovers performance."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention."}],"snapshot_sha256":"f781aae1e3143eebab73738df5644d515e07334a7b61102f7bf2c0d2c02506dc"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e5de90e08a366df0720a68a411b87a1a69cdf948a4230f53c5f6224a4599c976"},"integrity":{"available":true,"clean":false,"detectors_run":[{"findings_count":1,"name":"doi_compliance","ran_at":"2026-05-19T21:01:24.900423Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.213886Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.295789Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.372472Z","status":"skipped","version":"1.0.0"}],"endpoint":"/pith/2605.16874/integrity.json","findings":[{"audited_at":"2026-05-19T21:01:24.900423Z","detected_arxiv_id":null,"detected_doi":"10.1109/TIT.2009.2027527.URL","detector":"doi_compliance","finding_type":"recoverable_identifier","note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1109/TIT.2009.2027527.URL) was visible in the surrounding text but could not be confirmed against doi.org as printed.","ref_index":10,"severity":"advisory","verdict_class":"incontrovertible"}],"snapshot_sha256":"001554fc071bcef2dbad9ab2d4cbe1380ebca1053576159ee2b79183bc7bdd8a","summary":{"advisory":1,"by_detector":{"doi_compliance":{"advisory":1,"critical":0,"informational":0,"total":1}},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large reasoning models (LRMs) substantially outperform their base LLM counterparts on challenging reasoning benchmarks, yet it remains poorly understood where base models go wrong during token-by-token generation and how to narrow this gap efficiently. We study the base-reasoning gap through quantifying token-level distributional disagreement between a base model and a stronger reasoning model using likelihood-based divergences. Across benchmarks, we find that the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Q","authors_text":"An Zhang, Changshuo Shen, Leheng Sheng, Xiang Wang, Yuxin Chen","cross_cats":[],"headline":"Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-16T08:33:31Z","title":"Reasoning Can Be Restored by Correcting a Few Decision Tokens"},"references":{"count":33,"internal_anchors":15,"resolved_work":33,"sample":[{"cited_arxiv_id":"2005.14165","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Language Models are Few-Shot Learners","work_id":"214732c0-2edd-44a0-af9e-28184a2b8279","year":2005},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"arXiv preprint arXiv:2510.00553 , year=","work_id":"a6fda285-7688-48ed-9537-cbaa641673ef","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Cascade speculative drafting for even faster llm inference,","work_id":"a42682ab-46cc-4a73-823d-d74a2fe116e9","year":null},{"cited_arxiv_id":"2110.14168","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","year":null},{"cited_arxiv_id":"2503.01307","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs","work_id":"f65a84bf-c5b4-4491-a618-18bb263c60e5","year":null}],"snapshot_sha256":"19f1cf9b040ccd4069233bab69759a120d96a88c0291a629b1ea91f008d01313"},"source":{"id":"2605.16874","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T20:52:50.412735Z","id":"0fc56498-5930-4705-bf68-ead1ca1ba92b","model_set":{"reader":"grok-4.3"},"one_line_summary":"Reasoning gaps between base LLMs and LRMs concentrate on ~8% of early planning tokens; intervening with the reasoning model only at high-disagreement positions recovers performance.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention.","strongest_claim":"Across benchmarks, the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Qwen3-0.6B, only ~8% of generated tokens account for the salient disagreement, and these tokens concentrate early in the response, are strongly enriched in planning-related decisions (17x), and coincide with high base-model uncertainty.","weakest_assumption":"That the positions of high likelihood-based distributional disagreement between base and reasoning models are precisely the causal points where the base model’s early planning errors derail the subsequent reasoning trajectory, rather than merely correlated symptoms."}},"verdict_id":"0fc56498-5930-4705-bf68-ead1ca1ba92b"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:0e1c5c0038891e136a20f2b74b0ac8b29346843b943110f251cd3bbf236da41a","target":"record","created_at":"2026-05-20T00:03:27Z","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":"3eb0e59a986b0dd21b782a2e73c6c5252b1f6a3e0f75243f097c982a4f0a5045","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-16T08:33:31Z","title_canon_sha256":"c0004652b2b530242bee41cf3e561ad9c0833ae09b31c751047edb1048d36ad9"},"schema_version":"1.0","source":{"id":"2605.16874","kind":"arxiv","version":1}},"canonical_sha256":"dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272","first_computed_at":"2026-05-20T00:03:27.622281Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:03:27.622281Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"kBM2BEA66weVHZg6QlzOOrb61Aq9Ioq3SuGLM0YiQuZFqv72DbbJ1nlQ1Pf5BMHhXrONZ6dMuOicugEy5OgMDQ==","signature_status":"signed_v1","signed_at":"2026-05-20T00:03:27.622951Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.16874","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:49a76f015c4f33e09454a82cf9a27c02a1bab1408e4d679ee1682ff877c11041","sha256:0e1c5c0038891e136a20f2b74b0ac8b29346843b943110f251cd3bbf236da41a","sha256:9b05fe1ab56ce8707df507f1eadba26d04df72a04d2cb88ab723acec0022225b"],"state_sha256":"4406d9b0fcf9021eeb656f6b404d0f4152602773725989c665ac35953b073f12"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"PAqKirTZzwskTMTeVSxyfN8F5ASUDkungzaZmMBpxC6LmTm8HqdDAYCY/ecW8Axbxi5jK5rfzS/wq6hWYHj8AA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-20T14:11:31.899836Z","bundle_sha256":"0b1ecf1a3c93afafaa9c34d005ea0753daae78b3fb873f4694b5ed55db342642"}}