{
  "schema_version": 1,
  "id": "evt-2023-baldur",
  "slug": "baldur-whole-proof-generation-repair",
  "title": {
    "en": "Baldur generates and repairs whole formal proofs with language models",
    "zh-CN": "Baldur 使用语言模型生成并修复完整形式证明"
  },
  "dates": {
    "event": "2023-03-08",
    "first_public_claim": "2023-03-08",
    "paper_release": "2023-03-08",
    "artifact_release": null,
    "independently_verified": null,
    "added_to_chronicle": "2026-09-11",
    "last_updated": "2026-09-11"
  },
  "event_types": [
    "proof",
    "system"
  ],
  "significance": {
    "tier": "H2",
    "rationale": {
      "en": "Baldur established whole-proof generation plus checker-guided proof repair as a distinct alternative to step-by-step formal proof search, showing that a transformer could emit complete Isabelle/HOL proofs and use failed attempts and error messages as repair context.",
      "zh-CN": "Baldur 将“完整证明生成 + 检查器反馈修复”确立为逐步形式证明搜索之外的一条独立路线，展示 Transformer 可以一次生成完整 Isabelle/HOL 证明，并利用失败证明与错误信息进行修复。"
    }
  },
  "mathematical_novelty": {
    "type": "tooling_only"
  },
  "interfaces": [
    "formal",
    "symbolic",
    "hybrid"
  ],
  "claim": {
    "en": "On a benchmark of 6,336 Isabelle/HOL theorems, Baldur generated whole proofs and repaired failed generations, proving an additional 8.7% of theorems beyond Thor; the paper reported that Baldur and Thor together proved 65.7% fully automatically.",
    "zh-CN": "在包含 6,336 个 Isabelle/HOL 定理的基准上，Baldur 生成完整证明并修复失败结果；论文报告它相较 Thor 额外证明了 8.7% 的定理，而 Baldur 与 Thor 合计可全自动证明 65.7%。"
  },
  "summary": {
    "en": "Instead of predicting one proof step and searching over continuations, Baldur fine-tuned language models to generate an entire proof in one pass. A separate repair model then received a failed proof attempt together with proof-assistant error feedback and proposed a corrected proof.",
    "zh-CN": "与逐步预测证明动作并搜索后续路径不同，Baldur 微调语言模型一次生成完整证明；随后由独立的修复模型读取失败证明及证明助手的错误反馈，再提出修正后的证明。"
  },
  "why_it_matters": {
    "en": "The work made complete-proof synthesis and feedback-driven repair a credible formal-reasoning pattern, complementing search-heavy theorem provers and anticipating later generate-check-revise workflows.",
    "zh-CN": "该工作使完整证明合成与反馈驱动修复成为可信的形式推理模式，补充了依赖搜索的定理证明路线，并预示了后来常见的“生成—检查—修订”工作流。"
  },
  "ai_role": {
    "level": "ai_primary_human_verified",
    "description": {
      "en": "Fine-tuned language models generated complete candidate proofs and repaired failed attempts using checker feedback; Isabelle/HOL mechanically accepted or rejected the resulting proof text.",
      "zh-CN": "微调后的语言模型生成完整候选证明，并根据检查器反馈修复失败尝试；Isabelle/HOL 对最终证明文本进行机械接受或拒绝。"
    }
  },
  "human_contribution": {
    "description": {
      "en": "Researchers designed the whole-proof and repair training setup, assembled the evaluation protocol, trained the models and analyzed performance against prior theorem-proving systems.",
      "zh-CN": "研究者设计完整证明与修复训练方案、构建评测流程、训练模型，并分析其相对既有定理证明系统的表现。"
    }
  },
  "systems": [
    "baldur",
    "isabelle-hol"
  ],
  "organizations": [],
  "people": [
    "emily-first",
    "markus-n-rabe",
    "talia-ringer",
    "yuriy-brun"
  ],
  "problems": [
    "automated-theorem-proving"
  ],
  "methods": [
    "whole-proof-generation",
    "proof-repair",
    "language-models",
    "proof-assistant-feedback"
  ],
  "verification": {
    "evidence_level": "E2",
    "status": "paper_released",
    "formal_assurance": "none",
    "history": [
      {
        "date": "2023-03-08",
        "status": "paper_released",
        "evidence_level": "E2",
        "note": {
          "en": "The public research record reports the whole-proof generation, repair method and Isabelle/HOL evaluation. This status records the paper and does not imply independent Chronicle verification or a public replay artifact.",
          "zh-CN": "公开研究记录报告了完整证明生成、修复方法及 Isabelle/HOL 评测。该状态仅记录论文已公开，不表示 Chronicle 已独立验证，也不表示存在可公开重放的完整产物。"
        }
      }
    ]
  },
  "sources": [
    {
      "id": "arxiv-baldur-2303.04910",
      "tier": "S1",
      "type": "paper",
      "title": "Baldur: Whole-Proof Generation and Repair with Large Language Models",
      "url": "https://arxiv.org/abs/2303.04910",
      "primary": true,
      "published_date": "2023-03-08",
      "accessed_at": "2026-09-11"
    },
    {
      "id": "esec-fse-baldur-2023",
      "tier": "S1",
      "type": "other",
      "title": "Baldur: Whole-Proof Generation and Repair with Large Language Models",
      "url": "https://2023.esec-fse.org/details/fse-2023-research-papers/2/Baldur-Whole-Proof-Generation-and-Repair-with-Large-Language-Models",
      "primary": true,
      "published_date": "2023-12-06",
      "accessed_at": "2026-09-11"
    }
  ],
  "artifacts": [],
  "relationships": {
    "predecessors": [
      "evt-2022-thor"
    ],
    "successors": [],
    "related": [
      "evt-2023-leandojo",
      "evt-2023-copra"
    ]
  },
  "timeline": {
    "featured": true,
    "significance": "primary"
  },
  "tags": [
    "theorem-proving",
    "isabelle",
    "whole-proof-generation",
    "proof-repair",
    "language-model"
  ],
  "editorial": {
    "created_at": "2026-09-11",
    "last_reviewed": "2026-09-11",
    "reviewers": [
      "Charlie-Wang-03"
    ]
  },
  "corrections": []
}