{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:AI3Q6DGUEKWZTGOULIK5TBNYPV","short_pith_number":"pith:AI3Q6DGU","canonical_record":{"source":{"id":"2604.26940","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-29T17:51:39Z","cross_cats_sorted":[],"title_canon_sha256":"850d5e03e09599d22303606d0b455f575ce2eebb46eb0e5b2d77712776b2acf8","abstract_canon_sha256":"ed0358e7d0ccb92fd89780428de36be4d864d083f542cbb5c4c94f1484c717a6"},"schema_version":"1.0"},"canonical_sha256":"02370f0cd422ad9999d45a15d985b87d5988dd172ea81c759169e5e584777d9e","source":{"kind":"arxiv","id":"2604.26940","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.26940","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"arxiv_version","alias_value":"2604.26940v2","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.26940","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"pith_short_12","alias_value":"AI3Q6DGUEKWZ","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"pith_short_16","alias_value":"AI3Q6DGUEKWZTGOU","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"pith_short_8","alias_value":"AI3Q6DGU","created_at":"2026-06-12T01:09:28Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:AI3Q6DGUEKWZTGOULIK5TBNYPV","target":"record","payload":{"canonical_record":{"source":{"id":"2604.26940","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-29T17:51:39Z","cross_cats_sorted":[],"title_canon_sha256":"850d5e03e09599d22303606d0b455f575ce2eebb46eb0e5b2d77712776b2acf8","abstract_canon_sha256":"ed0358e7d0ccb92fd89780428de36be4d864d083f542cbb5c4c94f1484c717a6"},"schema_version":"1.0"},"canonical_sha256":"02370f0cd422ad9999d45a15d985b87d5988dd172ea81c759169e5e584777d9e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-12T01:09:28.179652Z","signature_b64":"50EbXUQ37xqnPnaTmRUbeNpMccD7sJS1uoT5UiUbS2ebR5eCOKjqpcwIV4cBQQjX8HzmBSpabdaKVngiWRFCBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"02370f0cd422ad9999d45a15d985b87d5988dd172ea81c759169e5e584777d9e","last_reissued_at":"2026-06-12T01:09:28.179226Z","signature_status":"signed_v1","first_computed_at":"2026-06-12T01:09:28.179226Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.26940","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-12T01:09:28Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"eLvhwXwDVfcc1oSCDgBSp8QzwRz72GxKg/Egnb8sMRt3fMJP6w7R9LQIfevjnHlww+DTgUj32Vb3ZMDweBvYBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-21T18:53:49.324031Z"},"content_sha256":"07031ec3d4f67ccb3048af355d6721a723020ba7cf1e066b2887422689a33a3a","schema_version":"1.0","event_id":"sha256:07031ec3d4f67ccb3048af355d6721a723020ba7cf1e066b2887422689a33a3a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:AI3Q6DGUEKWZTGOULIK5TBNYPV","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Select to Think: Unlocking SLM Potential with Local Sufficiency","license":"http://creativecommons.org/licenses/by/4.0/","headline":"The top-8 predictions of a 1.5B SLM contain the 32B LLM's preferred token 95 percent of the time at reasoning divergence points.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Georg Carle, Wenxuan Ye, Xueli An, Yangyang Zhang, Yunpu Ma","submitted_at":"2026-04-29T17:51:39Z","abstract_excerpt":"Small language models (SLMs) offer efficient deployment, yet they often lag behind their larger counterparts (LLMs) in reasoning. Existing remedies either invoke an LLM at points of reasoning divergence, incurring substantial latency and cost, or rely on standard distillation, which is limited by the SLM's capacity to accurately mimic the LLM's complex generative distribution. We address this dilemma by identifying local sufficiency: at divergence points, the LLM's preferred token often resides within the SLM's top-K next-token predictions, even when failing to emerge as the SLM top-1 choice. "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"A 1.5B SLM's top-8 candidates capture the 32B LLM's choice with 95% hit rate. S2T-LOCAL improves greedy decoding by 24.1% on average across benchmarks, matching 8-path self-consistency with single-trajectory efficiency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the local sufficiency property holds consistently across diverse tasks and model scales, allowing the distilled selection logic to generalize without degrading performance on unseen data.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Small language models can achieve near large-model reasoning performance by learning to re-rank their own top-K token predictions after distilling selection from the large model.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"The top-8 predictions of a 1.5B SLM contain the 32B LLM's preferred token 95 percent of the time at reasoning divergence points.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"78004dc603f47f96854eb3effa17710285e4544d28b0ce5f35d52a10fab958ff"},"source":{"id":"2604.26940","kind":"arxiv","version":2},"verdict":{"id":"a95ff53d-fa93-4d0a-9e81-b7c7d4fb938f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T08:37:23.877641Z","strongest_claim":"A 1.5B SLM's top-8 candidates capture the 32B LLM's choice with 95% hit rate. S2T-LOCAL improves greedy decoding by 24.1% on average across benchmarks, matching 8-path self-consistency with single-trajectory efficiency.","one_line_summary":"Small language models can achieve near large-model reasoning performance by learning to re-rank their own top-K token predictions after distilling selection from the large model.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the local sufficiency property holds consistently across diverse tasks and model scales, allowing the distilled selection logic to generalize without degrading performance on unseen data.","pith_extraction_headline":"The top-8 predictions of a 1.5B SLM contain the 32B LLM's preferred token 95 percent of the time at reasoning divergence points."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.26940/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T23:36:32.069161Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T19:41:42.741244Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"7b5dcb49942ae7bbee0d83f11ce515ed8697801fb32a90b91cbf4787b6fb1c19"},"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"},"verdict_id":"a95ff53d-fa93-4d0a-9e81-b7c7d4fb938f"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-12T01:09:28Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"R8eS2lJlTJu0dZLknoKeArSEf+H2zZ0m0FgRTkvXfs5WWrR1WnlpqYAbGqZxMhTP3KFRLaJtW3FJrXqRVl0YBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-21T18:53:49.324729Z"},"content_sha256":"57cd8daf32b6f0fc9da0adfce86853ffa5bd98f67ba89f49319153f2c4a1c5d4","schema_version":"1.0","event_id":"sha256:57cd8daf32b6f0fc9da0adfce86853ffa5bd98f67ba89f49319153f2c4a1c5d4"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV/bundle.json","state_url":"https://pith.science/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV/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":"2026-08-21T18:53:49Z","links":{"resolver":"https://pith.science/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV","bundle":"https://pith.science/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV/bundle.json","state":"https://pith.science/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV/state.json","well_known_bundle":"https://pith.science/.well-known/pith/AI3Q6DGUEKWZTGOULIK5TBNYPV/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:AI3Q6DGUEKWZTGOULIK5TBNYPV","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":"ed0358e7d0ccb92fd89780428de36be4d864d083f542cbb5c4c94f1484c717a6","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-29T17:51:39Z","title_canon_sha256":"850d5e03e09599d22303606d0b455f575ce2eebb46eb0e5b2d77712776b2acf8"},"schema_version":"1.0","source":{"id":"2604.26940","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.26940","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"arxiv_version","alias_value":"2604.26940v2","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.26940","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"pith_short_12","alias_value":"AI3Q6DGUEKWZ","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"pith_short_16","alias_value":"AI3Q6DGUEKWZTGOU","created_at":"2026-06-12T01:09:28Z"},{"alias_kind":"pith_short_8","alias_value":"AI3Q6DGU","created_at":"2026-06-12T01:09:28Z"}],"graph_snapshots":[{"event_id":"sha256:57cd8daf32b6f0fc9da0adfce86853ffa5bd98f67ba89f49319153f2c4a1c5d4","target":"graph","created_at":"2026-06-12T01:09:28Z","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":"A 1.5B SLM's top-8 candidates capture the 32B LLM's choice with 95% hit rate. S2T-LOCAL improves greedy decoding by 24.1% on average across benchmarks, matching 8-path self-consistency with single-trajectory efficiency."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the local sufficiency property holds consistently across diverse tasks and model scales, allowing the distilled selection logic to generalize without degrading performance on unseen data."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Small language models can achieve near large-model reasoning performance by learning to re-rank their own top-K token predictions after distilling selection from the large model."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"The top-8 predictions of a 1.5B SLM contain the 32B LLM's preferred token 95 percent of the time at reasoning divergence points."}],"snapshot_sha256":"78004dc603f47f96854eb3effa17710285e4544d28b0ce5f35d52a10fab958ff"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T23:36:32.069161Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T19:41:42.741244Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2604.26940/integrity.json","findings":[],"snapshot_sha256":"7b5dcb49942ae7bbee0d83f11ce515ed8697801fb32a90b91cbf4787b6fb1c19","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Small language models (SLMs) offer efficient deployment, yet they often lag behind their larger counterparts (LLMs) in reasoning. Existing remedies either invoke an LLM at points of reasoning divergence, incurring substantial latency and cost, or rely on standard distillation, which is limited by the SLM's capacity to accurately mimic the LLM's complex generative distribution. We address this dilemma by identifying local sufficiency: at divergence points, the LLM's preferred token often resides within the SLM's top-K next-token predictions, even when failing to emerge as the SLM top-1 choice. ","authors_text":"Georg Carle, Wenxuan Ye, Xueli An, Yangyang Zhang, Yunpu Ma","cross_cats":[],"headline":"The top-8 predictions of a 1.5B SLM contain the 32B LLM's preferred token 95 percent of the time at reasoning divergence points.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-29T17:51:39Z","title":"Select to Think: Unlocking SLM Potential with Local Sufficiency"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.26940","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-07T08:37:23.877641Z","id":"a95ff53d-fa93-4d0a-9e81-b7c7d4fb938f","model_set":{"reader":"grok-4.3"},"one_line_summary":"Small language models can achieve near large-model reasoning performance by learning to re-rank their own top-K token predictions after distilling selection from the large model.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"The top-8 predictions of a 1.5B SLM contain the 32B LLM's preferred token 95 percent of the time at reasoning divergence points.","strongest_claim":"A 1.5B SLM's top-8 candidates capture the 32B LLM's choice with 95% hit rate. S2T-LOCAL improves greedy decoding by 24.1% on average across benchmarks, matching 8-path self-consistency with single-trajectory efficiency.","weakest_assumption":"That the local sufficiency property holds consistently across diverse tasks and model scales, allowing the distilled selection logic to generalize without degrading performance on unseen data."}},"verdict_id":"a95ff53d-fa93-4d0a-9e81-b7c7d4fb938f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:07031ec3d4f67ccb3048af355d6721a723020ba7cf1e066b2887422689a33a3a","target":"record","created_at":"2026-06-12T01:09:28Z","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":"ed0358e7d0ccb92fd89780428de36be4d864d083f542cbb5c4c94f1484c717a6","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-29T17:51:39Z","title_canon_sha256":"850d5e03e09599d22303606d0b455f575ce2eebb46eb0e5b2d77712776b2acf8"},"schema_version":"1.0","source":{"id":"2604.26940","kind":"arxiv","version":2}},"canonical_sha256":"02370f0cd422ad9999d45a15d985b87d5988dd172ea81c759169e5e584777d9e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"02370f0cd422ad9999d45a15d985b87d5988dd172ea81c759169e5e584777d9e","first_computed_at":"2026-06-12T01:09:28.179226Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-12T01:09:28.179226Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"50EbXUQ37xqnPnaTmRUbeNpMccD7sJS1uoT5UiUbS2ebR5eCOKjqpcwIV4cBQQjX8HzmBSpabdaKVngiWRFCBA==","signature_status":"signed_v1","signed_at":"2026-06-12T01:09:28.179652Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.26940","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:07031ec3d4f67ccb3048af355d6721a723020ba7cf1e066b2887422689a33a3a","sha256:57cd8daf32b6f0fc9da0adfce86853ffa5bd98f67ba89f49319153f2c4a1c5d4"],"state_sha256":"2801457c437f84b43a62e291bc41885e4bb58362c2247f82dad9d2d9f2d64db0"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"q6HbBkwCCV17VXrrt6hhypo9ZAEnWqDUtCT0lFeKqlWfTNlWyjPKDm8e1tfV13ou+SrhtK7c7ZEbx07HpV4WBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-21T18:53:49.329170Z","bundle_sha256":"5b8c4a3d60095b01609ed4c823e062bacef142573abf53329fec3af56cd01fb2"}}