{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:EYH5ATMUN7OHZEYIBJM54EFL5N","short_pith_number":"pith:EYH5ATMU","canonical_record":{"source":{"id":"2605.15177","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T17:57:40Z","cross_cats_sorted":[],"title_canon_sha256":"fd1fb2f837b2166a439c7d7f9aa14e2259b8257a916a780b57a0860e6fba1d3d","abstract_canon_sha256":"ef02dfcef9273526a0b591c2f1ab4343c6f3163c6bf9372f2a743131cef8e1a4"},"schema_version":"1.0"},"canonical_sha256":"260fd04d946fdc7c93080a59de10abeb658e80d1fc3f1730b8558f75b529ec5a","source":{"kind":"arxiv","id":"2605.15177","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.15177","created_at":"2026-05-17T21:18:32Z"},{"alias_kind":"arxiv_version","alias_value":"2605.15177v1","created_at":"2026-05-17T21:18:32Z"},{"alias_kind":"pith_short_12","alias_value":"EYH5ATMUN7OH","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"EYH5ATMUN7OHZEYI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"EYH5ATMU","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:EYH5ATMUN7OHZEYIBJM54EFL5N","target":"record","payload":{"canonical_record":{"source":{"id":"2605.15177","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T17:57:40Z","cross_cats_sorted":[],"title_canon_sha256":"fd1fb2f837b2166a439c7d7f9aa14e2259b8257a916a780b57a0860e6fba1d3d","abstract_canon_sha256":"ef02dfcef9273526a0b591c2f1ab4343c6f3163c6bf9372f2a743131cef8e1a4"},"schema_version":"1.0"},"canonical_sha256":"260fd04d946fdc7c93080a59de10abeb658e80d1fc3f1730b8558f75b529ec5a","receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.2","canonical_sha256":"260fd04d946fdc7c93080a59de10abeb658e80d1fc3f1730b8558f75b529ec5a","last_reissued_at":"2026-05-17T21:57:18.566592Z","signature_status":"unsigned_v0","first_computed_at":"2026-05-17T21:40:25.215359Z"},"source_kind":"arxiv","source_id":"2605.15177","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":"2026-05-17T21:18:32Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JaLKSefRRzsQzc17u7r4POM6Xw3Cfw+asz+GInATfCa9rIHR1LHuEZHinJbqSD+gf3GOwq4AFYtPTx3dogSgDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T10:45:43.401403Z"},"content_sha256":"11a5f0f4797f2c7acb0daf890f614ed020ea7f52aad4cf60489690ad22565749","schema_version":"1.0","event_id":"sha256:11a5f0f4797f2c7acb0daf890f614ed020ea7f52aad4cf60489690ad22565749"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:EYH5ATMUN7OHZEYIBJM54EFL5N","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"OpenDeepThink: Parallel Reasoning via Bradley--Terry Aggregation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Bradley-Terry aggregation of LLM pairwise judgments improves reasoning by 405 Elo points.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Huanzhi Mao, Jingbo Shang, Kaiyuan Liu, Qiuyang Mang, Shang Zhou, Wenhao Chai","submitted_at":"2026-05-14T17:57:40Z","abstract_excerpt":"Test-time compute scaling is a primary axis for improving LLM reasoning. Existing methods primarily scale depth by extending a single reasoning trace. Scaling breadth by sampling multiple candidates in parallel is straightforward, but introduces a selection bottleneck: choosing the best candidate without a ground-truth verifier, since pointwise LLM judging is noisy and biased. To address this, we introduce OpenDeepThink, a population-based test-time compute framework that selects via pairwise Bradley-Terry comparison. Each generation, the LLM judges random pairs of candidates and aggregates vo"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405 points in eight sequential LLM-call rounds (~27 minutes wall-clock).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That LLM pairwise judgments are sufficiently consistent and unbiased across random pairs to yield a global ranking via Bradley-Terry that reliably identifies higher-quality reasoning traces.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"OpenDeepThink improves LLM reasoning by ranking parallel candidate traces via Bradley-Terry aggregation of LLM pairwise judgments, achieving a +405 Codeforces Elo gain on Gemini 3.1 Pro after eight rounds.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Bradley-Terry aggregation of LLM pairwise judgments improves reasoning by 405 Elo points.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b17c4585aaac34d81f1696c81a6b1d091252a68afe6e8ce3d90a750ef57fb223"},"source":{"id":"2605.15177","kind":"arxiv","version":1},"verdict":{"id":"96c0c9f2-8d59-4201-b0ed-b01494f8bcab","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:02:21.710282Z","strongest_claim":"OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405 points in eight sequential LLM-call rounds (~27 minutes wall-clock).","one_line_summary":"OpenDeepThink improves LLM reasoning by ranking parallel candidate traces via Bradley-Terry aggregation of LLM pairwise judgments, achieving a +405 Codeforces Elo gain on Gemini 3.1 Pro after eight rounds.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That LLM pairwise judgments are sufficiently consistent and unbiased across random pairs to yield a global ranking via Bradley-Terry that reliably identifies higher-quality reasoning traces.","pith_extraction_headline":"Bradley-Terry aggregation of LLM pairwise judgments improves reasoning by 405 Elo points."},"references":{"count":24,"sample":[{"doi":"","year":null,"title":"Large Language Monkeys: Scaling Inference Compute with Repeated Sampling","work_id":"b124064d-5a56-42ad-86f5-3cc349b86a3a","ref_index":1,"cited_arxiv_id":"2407.21787","is_internal_anchor":true},{"doi":"","year":null,"title":"Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference","work_id":"a1eb83da-5727-4b63-977f-81c6fdd33936","ref_index":2,"cited_arxiv_id":"2403.04132","is_internal_anchor":true},{"doi":"","year":null,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","ref_index":3,"cited_arxiv_id":"2110.14168","is_internal_anchor":true},{"doi":"","year":null,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","work_id":"008df105-2fdd-45d8-857a-8e35868aecb6","ref_index":4,"cited_arxiv_id":"2507.06261","is_internal_anchor":true},{"doi":"","year":null,"title":"Escaping the cognitive well: Efficient competition math with off-the-shelf models.arXiv preprint arXiv:2602.16793,","work_id":"122f4150-56a0-4e98-a0aa-0fc5efc6c27e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":24,"snapshot_sha256":"05e1196e466d7ef893b1135b6ba8b2e97248da698b9eeeeeb8f6e5d534c74c56","internal_anchors":12},"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":"96c0c9f2-8d59-4201-b0ed-b01494f8bcab"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T21:57:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RDcSvcmo5RRh5CCpmlg7wnlOJbiBHBx+1Aul679Q5hNQd0v7sZYISVrNlCHQ2wMSw+XvRG4jVN9OEHhaaWRTDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T10:45:43.402398Z"},"content_sha256":"aef79cd4ff41c917fbf2b7752824a4d41c6ab9b34acb9f877ae67b79e125f3af","schema_version":"1.0","event_id":"sha256:aef79cd4ff41c917fbf2b7752824a4d41c6ab9b34acb9f877ae67b79e125f3af"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/EYH5ATMUN7OHZEYIBJM54EFL5N/bundle.json","state_url":"https://pith.science/pith/EYH5ATMUN7OHZEYIBJM54EFL5N/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/EYH5ATMUN7OHZEYIBJM54EFL5N/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":"2026-06-10T10:45:43Z","links":{"resolver":"https://pith.science/pith/EYH5ATMUN7OHZEYIBJM54EFL5N","bundle":"https://pith.science/pith/EYH5ATMUN7OHZEYIBJM54EFL5N/bundle.json","state":"https://pith.science/pith/EYH5ATMUN7OHZEYIBJM54EFL5N/state.json","well_known_bundle":"https://pith.science/.well-known/pith/EYH5ATMUN7OHZEYIBJM54EFL5N/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:EYH5ATMUN7OHZEYIBJM54EFL5N","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":"ef02dfcef9273526a0b591c2f1ab4343c6f3163c6bf9372f2a743131cef8e1a4","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T17:57:40Z","title_canon_sha256":"fd1fb2f837b2166a439c7d7f9aa14e2259b8257a916a780b57a0860e6fba1d3d"},"schema_version":"1.0","source":{"id":"2605.15177","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.15177","created_at":"2026-05-17T21:18:32Z"},{"alias_kind":"arxiv_version","alias_value":"2605.15177v1","created_at":"2026-05-17T21:18:32Z"},{"alias_kind":"pith_short_12","alias_value":"EYH5ATMUN7OH","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"EYH5ATMUN7OHZEYI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"EYH5ATMU","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:aef79cd4ff41c917fbf2b7752824a4d41c6ab9b34acb9f877ae67b79e125f3af","target":"graph","created_at":"2026-05-17T21:57:18Z","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":"OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405 points in eight sequential LLM-call rounds (~27 minutes wall-clock)."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That LLM pairwise judgments are sufficiently consistent and unbiased across random pairs to yield a global ranking via Bradley-Terry that reliably identifies higher-quality reasoning traces."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"OpenDeepThink improves LLM reasoning by ranking parallel candidate traces via Bradley-Terry aggregation of LLM pairwise judgments, achieving a +405 Codeforces Elo gain on Gemini 3.1 Pro after eight rounds."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Bradley-Terry aggregation of LLM pairwise judgments improves reasoning by 405 Elo points."}],"snapshot_sha256":"b17c4585aaac34d81f1696c81a6b1d091252a68afe6e8ce3d90a750ef57fb223"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Test-time compute scaling is a primary axis for improving LLM reasoning. Existing methods primarily scale depth by extending a single reasoning trace. Scaling breadth by sampling multiple candidates in parallel is straightforward, but introduces a selection bottleneck: choosing the best candidate without a ground-truth verifier, since pointwise LLM judging is noisy and biased. To address this, we introduce OpenDeepThink, a population-based test-time compute framework that selects via pairwise Bradley-Terry comparison. Each generation, the LLM judges random pairs of candidates and aggregates vo","authors_text":"Huanzhi Mao, Jingbo Shang, Kaiyuan Liu, Qiuyang Mang, Shang Zhou, Wenhao Chai","cross_cats":[],"headline":"Bradley-Terry aggregation of LLM pairwise judgments improves reasoning by 405 Elo points.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T17:57:40Z","title":"OpenDeepThink: Parallel Reasoning via Bradley--Terry Aggregation"},"references":{"count":24,"internal_anchors":12,"resolved_work":24,"sample":[{"cited_arxiv_id":"2407.21787","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Large Language Monkeys: Scaling Inference Compute with Repeated Sampling","work_id":"b124064d-5a56-42ad-86f5-3cc349b86a3a","year":null},{"cited_arxiv_id":"2403.04132","doi":"","is_internal_anchor":true,"ref_index":2,"title":"Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference","work_id":"a1eb83da-5727-4b63-977f-81c6fdd33936","year":null},{"cited_arxiv_id":"2110.14168","doi":"","is_internal_anchor":true,"ref_index":3,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","year":null},{"cited_arxiv_id":"2507.06261","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","work_id":"008df105-2fdd-45d8-857a-8e35868aecb6","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Escaping the cognitive well: Efficient competition math with off-the-shelf models.arXiv preprint arXiv:2602.16793,","work_id":"122f4150-56a0-4e98-a0aa-0fc5efc6c27e","year":null}],"snapshot_sha256":"05e1196e466d7ef893b1135b6ba8b2e97248da698b9eeeeeb8f6e5d534c74c56"},"source":{"id":"2605.15177","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T03:02:21.710282Z","id":"96c0c9f2-8d59-4201-b0ed-b01494f8bcab","model_set":{"reader":"grok-4.3"},"one_line_summary":"OpenDeepThink improves LLM reasoning by ranking parallel candidate traces via Bradley-Terry aggregation of LLM pairwise judgments, achieving a +405 Codeforces Elo gain on Gemini 3.1 Pro after eight rounds.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Bradley-Terry aggregation of LLM pairwise judgments improves reasoning by 405 Elo points.","strongest_claim":"OpenDeepThink raises Gemini 3.1 Pro's effective Codeforces Elo by +405 points in eight sequential LLM-call rounds (~27 minutes wall-clock).","weakest_assumption":"That LLM pairwise judgments are sufficiently consistent and unbiased across random pairs to yield a global ranking via Bradley-Terry that reliably identifies higher-quality reasoning traces."}},"verdict_id":"96c0c9f2-8d59-4201-b0ed-b01494f8bcab"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:11a5f0f4797f2c7acb0daf890f614ed020ea7f52aad4cf60489690ad22565749","target":"record","created_at":"2026-05-17T21:18:32Z","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":"ef02dfcef9273526a0b591c2f1ab4343c6f3163c6bf9372f2a743131cef8e1a4","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-14T17:57:40Z","title_canon_sha256":"fd1fb2f837b2166a439c7d7f9aa14e2259b8257a916a780b57a0860e6fba1d3d"},"schema_version":"1.0","source":{"id":"2605.15177","kind":"arxiv","version":1}},"canonical_sha256":"260fd04d946fdc7c93080a59de10abeb658e80d1fc3f1730b8558f75b529ec5a","receipt":{"builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"260fd04d946fdc7c93080a59de10abeb658e80d1fc3f1730b8558f75b529ec5a","first_computed_at":"2026-05-17T21:40:25.215359Z","kind":"pith_receipt","last_reissued_at":"2026-05-17T21:57:18.566592Z","receipt_version":"0.2","signature_status":"unsigned_v0"},"source_id":"2605.15177","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:11a5f0f4797f2c7acb0daf890f614ed020ea7f52aad4cf60489690ad22565749","sha256:aef79cd4ff41c917fbf2b7752824a4d41c6ab9b34acb9f877ae67b79e125f3af"],"state_sha256":"828aee1f060ec56d38129572c1e64664ebb3302fa260da00fd54b07410bcec60"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"32uUF4z5intyDTO9AEtel8BkInfiM0pskx4KjiV5ZV5L1qIKBZq4nrN4TK7khQ6F6ErJGxAkxFYGRG8f1ntCDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-10T10:45:43.407413Z","bundle_sha256":"aa0b84112cad38050255f4506df509adf2bfa3d20110e263c0702d34d87b24d7"}}