{
  "title": "把知识留给后来的人：可以纠错的知识目录",
  "version": "1.0",
  "published_on": "2026-11-12",
  "scope": "Six selected public examples. Not a complete review of all personal records or an independent replication.",
  "records": [
    {
      "id": "K01",
      "title": "把 AI 的错误变成测试",
      "title_en": "Turn AI errors into tests",
      "status": "合成测试工具，尚无模型实测成绩",
      "status_en": "Synthetic testing tool; no reported model runs",
      "claim": "可公开保存的是新编题面、失败条件和验收标准，让同一种错误能在新情境里再被检查。",
      "claim_en": "New test prompts, failure conditions and acceptance criteria let others check the same failure mode in another setting.",
      "limit": "10 道题并非 10 起独立事件，也不是模型能力排行榜。标准本身仍需检验。",
      "limit_en": "Ten cases are not ten independent incidents or a model leaderboard. The rubric itself needs validation.",
      "next": "固定模型版本与设置，保留完整输出，盲评并记录评审分歧；另留未公开题目检查泛化。",
      "next_en": "Fix model versions and settings, retain full outputs, record blinded judgments and disagreements, and reserve unseen cases.",
      "path": "/blog/ai-correction-probes/",
      "files": [
        "/blog/ai-correction-probes/cases.zh.json",
        "/blog/ai-correction-probes/cases.en.json"
      ],
      "source_url": "https://weijia-lab.weijiaxian.chatgpt.site/blog/ai-correction-probes/",
      "reviewed_on": "2026-11-12",
      "independently_reproduced_for_this_index": false,
      "download_urls": [
        "https://weijia-lab.weijiaxian.chatgpt.site/blog/ai-correction-probes/cases.zh.json",
        "https://weijia-lab.weijiaxian.chatgpt.site/blog/ai-correction-probes/cases.en.json"
      ]
    },
    {
      "id": "K02",
      "title": "保留没有通过的预测方法",
      "title_en": "Keep methods that did not pass",
      "status": "有边界的历史检验与负面结果",
      "status_en": "Bounded historical tests and negative results",
      "claim": "比特币研究公开了四项具体机制检验的规则、数据与代码，并保留未达到采用门槛的结果。失败记录能减少后来者重复同一种误判。",
      "claim_en": "The Bitcoin study preserves rules, data, code and results from four specific mechanism tests that did not meet its adoption criteria. Such records can help others avoid repeating the same mistake.",
      "limit": "四项检验共享数据和市场阶段，不能当作四个独立市场。它们不证明比特币不可预测，也不证明所有相关指标无效。",
      "limit_en": "The tests overlap in data and market regimes. They do not establish that Bitcoin is unpredictable or that all related indicators are useless.",
      "next": "复跑冻结材料，核对时间切分、目标定义与同日基准；新假设先写规则，再看新增数据。",
      "next_en": "Rerun the frozen materials and check time splits, target definitions and matched baselines. Specify new rules before examining new data.",
      "path": "/blog/bitcoin-bottom/",
      "files": [
        "/blog/bitcoin-bottom/files/research.zip"
      ],
      "source_url": "https://weijia-lab.weijiaxian.chatgpt.site/blog/bitcoin-bottom/",
      "reviewed_on": "2026-11-12",
      "independently_reproduced_for_this_index": false,
      "download_urls": [
        "https://weijia-lab.weijiaxian.chatgpt.site/blog/bitcoin-bottom/files/research.zip"
      ]
    },
    {
      "id": "K03",
      "title": "把撤回和修订也写进研究",
      "title_en": "Preserve withdrawals and revisions",
      "status": "历史类比与有日期的修订",
      "status_en": "Historical analogy with dated revisions",
      "claim": "宁德时代文章保留了固定时滞假设不受支持的结果，以及 10 月 27 日原文和 11 月 6 日追加的评价规则。值得传下去的是如何留下判断轨迹。",
      "claim_en": "The CATL article retains unsupported fixed-lag hypotheses, its October 27 text and the evaluation rules appended on November 6. The reusable contribution is the visible path of revision.",
      "limit": "历史可比样本很少。本文不更新其市场判断，不把旧观察窗口重新包装成可靠预测或交易建议。",
      "limit_en": "Comparable historical samples are scarce. This index does not update the market view or promote an old observation window as a reliable forecast or trading advice.",
      "next": "原文与新增证据分开记日期；到期按事先确定的口径评价，改变规则时保留旧版本和原因。",
      "next_en": "Date new evidence separately from the original text. Evaluate at the planned endpoint and preserve old rules with reasons for changes.",
      "path": "/blog/catl-cycle-bottom/",
      "files": [
        "/downloads/catl-backtest-audit.zip",
        "/downloads/catl-expectations-audit.zip"
      ],
      "source_url": "https://weijia-lab.weijiaxian.chatgpt.site/blog/catl-cycle-bottom/",
      "reviewed_on": "2026-11-12",
      "independently_reproduced_for_this_index": false,
      "download_urls": [
        "https://weijia-lab.weijiaxian.chatgpt.site/downloads/catl-backtest-audit.zip",
        "https://weijia-lab.weijiaxian.chatgpt.site/downloads/catl-expectations-audit.zip"
      ]
    },
    {
      "id": "K04",
      "title": "把协作问题交给可比较的试验",
      "title_en": "Make collaboration testable",
      "status": "研究协议，尚未证明协作增益",
      "status_en": "Research protocol; no established collaboration gain",
      "claim": "人机协作文章把“人、AI、人加 AI”放在可比较的条件下，提供了可下载协议。协议能被接着做，即使结果还不存在。",
      "claim_en": "The human–AI collaboration article proposes comparable human-only, AI-only and combined conditions and provides a downloadable protocol. A protocol can be continued before results exist.",
      "limit": "协议是一项提案，不能把网页已完成当作试验已完成，也不能据此声称人机组合一定更好。",
      "limit_en": "The protocol is a proposal. A finished webpage is not a completed experiment, and synergy is not guaranteed.",
      "next": "先固定任务、评价标准与资源条件，再比较三个基线；结果中保留错误、耗时与无增益的情况。",
      "next_en": "Fix tasks, evaluation criteria and resource conditions before comparing the three baselines. Retain errors, time costs and null results.",
      "path": "/blog/human-ai-division-of-work/",
      "files": [
        "/blog/human-ai-division-of-work/protocol.json"
      ],
      "source_url": "https://weijia-lab.weijiaxian.chatgpt.site/blog/human-ai-division-of-work/",
      "reviewed_on": "2026-11-12",
      "independently_reproduced_for_this_index": false,
      "download_urls": [
        "https://weijia-lab.weijiaxian.chatgpt.site/blog/human-ai-division-of-work/protocol.json"
      ]
    },
    {
      "id": "K05",
      "title": "区分“看懂了”与“以后还会”",
      "title_en": "Distinguish understanding now from learning that lasts",
      "status": "公开文献解读与待检验的学习建议",
      "status_en": "Literature synthesis and a learning proposal",
      "claim": "AI 学习文章把即时表现、延迟保持和新情境迁移分开讨论，并把“先尝试、后求助”写成可以被反驳的个人检验。",
      "claim_en": "The AI learning article distinguishes immediate performance, delayed retention and transfer, and frames “try first, ask for help later” as a testable personal strategy.",
      "limit": "本文沿用的是已公开的文献解读，未独立复现所引研究；不同人群和任务的结果不能直接合并成万能结论。",
      "limit_en": "This is a published interpretation of literature, not an independent replication. Different populations and tasks cannot be collapsed into a universal conclusion.",
      "next": "回到原始论文检查对象、对照和测量时间；若要做自己的比较，事先固定延迟测试与评分规则。",
      "next_en": "Check the original studies, controls and timing. For a personal comparison, specify delayed testing and scoring in advance.",
      "path": "/blog/ai-understanding/",
      "files": [],
      "source_url": "https://weijia-lab.weijiaxian.chatgpt.site/blog/ai-understanding/",
      "reviewed_on": "2026-11-12",
      "independently_reproduced_for_this_index": false,
      "download_urls": []
    },
    {
      "id": "K06",
      "title": "十次转述不等于十份证据",
      "title_en": "Ten retellings are not ten sources",
      "status": "公开工程修复记录；合成用例",
      "status_en": "Published engineering repair record; synthetic cases",
      "claim": "fanshi-research 的验证记录记载：给十次 AI 转述换名称并人工确认，曾让删除原始来源后的结论仍保持“受支持”。修订把 AI 与转载作为追溯节点，并按来源链归并。",
      "claim_en": "The fanshi-research validation record documents a flaw: ten renamed, manually confirmed AI retellings kept a conclusion “supported” after the original was removed. The repair treats AI and reposts as trace nodes and groups them by provenance.",
      "limit": "这里核对了公开文档，未独立重跑其测试。合成测试与来源登记不能证明现实材料真实，也不能代替真实市场验证。",
      "limit_en": "The public document was checked for this article; its tests were not independently rerun. Synthetic tests and a registry do not authenticate real-world evidence or validate financial performance.",
      "next": "把每条论断连到最初来源；删除一个来源族后重评。保留“未知”，不要用更多转述填满证据缺口。",
      "next_en": "Trace claims to their origin and reassess after removing an origin group. Preserve “unknown” rather than filling the gap with more retellings.",
      "url": "https://github.com/Ericnotme/fanshi-research/blob/main/docs/validation.md",
      "files": [],
      "source_url": "https://github.com/Ericnotme/fanshi-research/blob/main/docs/validation.md",
      "reviewed_on": "2026-11-12",
      "independently_reproduced_for_this_index": false,
      "download_urls": []
    }
  ],
  "blank_record": {
    "id": "",
    "claim": "",
    "source_url": "",
    "source_date": "",
    "evidence_type": "",
    "assumptions": [],
    "limits": [],
    "counterevidence": [],
    "next_test": "",
    "revision_date": "",
    "revision_reason": "",
    "public_sharing_checked": false
  },
  "revisions": [
    {
      "version": "1.0",
      "date": "2026-11-12",
      "change": "Initial public-source index. No private conversations, private health or financial records, or third-party letters included."
    }
  ]
}
