{
  "scope": "Component inventory plus bounded editorial mathematics review; see mathematical-review.json for source-level scope",
  "entries": [
    {
      "language": "en",
      "path": "applications/bounded-tools/",
      "title": "Tool-using assistants: validate, execute and trace",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "applications/document-evidence/",
      "title": "File Q&A: build an evidence-first document assistant",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "applications/validated-extraction/",
      "title": "Document extraction: JSON shape is not factual correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "frontier/adaptive-agent-evaluation/",
      "title": "Self-improving agents: reward gains are not capability gains",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "frontier/agent-harness-engineering/",
      "title": "From LLM Calls to Agent Harnesses",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "frontier/diffusion-language/",
      "title": "Diffusion language models: masking and iterative decoding",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "frontier/robot-recovery/",
      "title": "Robot recovery: when retrying helps, and when it hides failure",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "frontier/selective-state-spaces/",
      "title": "Selective state spaces: Mamba and sequence memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "frontier/test-time-compute/",
      "title": "Test-time compute: search, verification and budgets",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/agent-1/",
      "title": "Human approval gate",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/agent-2/",
      "title": "Authorized mock write",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/agent-3/",
      "title": "Tool allowlist",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/agent-4/",
      "title": "Budget exhaustion",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/agent-5/",
      "title": "Memory truncation",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/attention-1/",
      "title": "Causal baseline",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/attention-2/",
      "title": "Bidirectional context",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/attention-3/",
      "title": "Sharp attention",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/attention-4/",
      "title": "Flat attention",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/attention-5/",
      "title": "Soft attention",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/cache-1/",
      "title": "Locality baseline",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/cache-2/",
      "title": "Capacity thrashing",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/cache-3/",
      "title": "TTL expiry",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/cache-4/",
      "title": "High reuse workload",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/cache-5/",
      "title": "All unique requests",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/embedding-1/",
      "title": "Cosine nearest neighbors",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/embedding-2/",
      "title": "Dot-product norm bias",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/embedding-3/",
      "title": "Query rotation",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/embedding-4/",
      "title": "Opposite direction",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/embedding-5/",
      "title": "Zero-vector edge case",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/judge-1/",
      "title": "Mixed judge agreement",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/judge-2/",
      "title": "Perfect agreement",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/judge-3/",
      "title": "Always choose first",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/judge-4/",
      "title": "Tie-heavy dataset",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/judge-5/",
      "title": "Consistently wrong",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/kv-1/",
      "title": "GQA baseline",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/kv-2/",
      "title": "Multi-head cache",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/kv-3/",
      "title": "Multi-query cache",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/kv-4/",
      "title": "Long-context batch",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/kv-5/",
      "title": "Page fragmentation",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/neural-1/",
      "title": "Learn XOR quadrants",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/neural-2/",
      "title": "Underpowered hidden layer",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/neural-3/",
      "title": "Slow learning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/neural-4/",
      "title": "Circle classification",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/neural-5/",
      "title": "Seed sensitivity",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/quantization-1/",
      "title": "Four-bit baseline",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/quantization-2/",
      "title": "Eight-bit reconstruction",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/quantization-3/",
      "title": "Two-bit stress test",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/quantization-4/",
      "title": "Outlier dominates scale",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/quantization-5/",
      "title": "All-zero tensor",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/retrieval-1/",
      "title": "Hybrid fusion baseline",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/retrieval-2/",
      "title": "Lexical only",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/retrieval-3/",
      "title": "Dense only",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/retrieval-4/",
      "title": "Oracle reranking ceiling",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/retrieval-5/",
      "title": "Candidate starvation",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/routing-1/",
      "title": "Two-tier routing",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/routing-2/",
      "title": "All strong model",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/routing-3/",
      "title": "Frequent fallback",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/routing-4/",
      "title": "Cheap model outage",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "learn/routing-5/",
      "title": "No primary failures",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/adam/",
      "title": "Adam: understand the update, not just the defaults",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/adamw/",
      "title": "AdamW: where you apply weight decay matters",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/confidence-calibration/",
      "title": "Calibration: being right is different from knowing how often",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/contrastive-predictive-coding/",
      "title": "CPC: learn by identifying the right future",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/deep-rl-reproducibility/",
      "title": "Deep RL that Matters: a score is a distribution",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/deepseek-r1/",
      "title": "DeepSeek-R1: rewards, reasoning and what the evidence establishes",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/dpo/",
      "title": "DPO: preference learning through relative log probabilities",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/flashattention/",
      "title": "FlashAttention: optimize data movement before counting FLOPs",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/llm-as-a-judge/",
      "title": "LLM-as-a-judge: evaluate the evaluator",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/lora/",
      "title": "LoRA: constrain the update, not the whole model",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/ppo/",
      "title": "PPO: inspect the clipped objective one sample at a time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/qlora/",
      "title": "QLoRA: a memory budget is more than weight precision",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/react/",
      "title": "ReAct: close the loop between actions and observations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/rlhf/",
      "title": "InstructGPT: change the training signal to change behavior",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/show-your-work/",
      "title": "Show Your Work: compare search budgets, not just winners",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/simclr/",
      "title": "SimCLR: augmentation defines what similarity means",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "papers/transformer-attention/",
      "title": "Transformer: turn token compatibility into information flow",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/diffusion-policy/",
      "title": "05 · Diffusion Policy and multimodal actions",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/feedback-and-planning/",
      "title": "03 · Feedback control and receding-horizon planning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/flow-matching-pi0/",
      "title": "07 · π₀ and continuous action flow matching",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/frames-and-kinematics/",
      "title": "01 · Frames, transformations and robot kinematics",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/imitation-and-act/",
      "title": "04 · Demonstrations, behavior cloning and ACT",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/perception-and-state/",
      "title": "02 · Perception, state estimation and time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/rl-sim2real-evaluation/",
      "title": "08 · Robot RL, sim-to-real and deployment evidence",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "physical-ai/vla-openvla/",
      "title": "06 · OpenVLA: vision, language and action interfaces",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "projects/attention/",
      "title": "Causal attention & KV-cache correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "projects/control/",
      "title": "Delayed feedback and bounded control",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "projects/retrieval/",
      "title": "Evidence retrieval with a measured baseline",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "research/contrastive-learning/",
      "title": "Contrastive Learning Lab",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "research/experiment-design/",
      "title": "Experiment Design Lab",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "research/gradient-debugger/",
      "title": "Gradient Debugger",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "research/information-theory/",
      "title": "Information Theory Lab",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "research/lora-rank/",
      "title": "LoRA Rank Lab",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "research/paper-reproduction/",
      "title": "Paper Reproduction Workbench",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/agent-rl/",
      "title": "Agent RL: train a policy inside a harness",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/exploration-credit/",
      "title": "Exploration, credit assignment & reward design",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/grpo-verifiable/",
      "title": "GRPO, verifiable rewards & reasoning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/mdp-bellman/",
      "title": "MDPs, returns & Bellman equations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/policy-gradient/",
      "title": "Policy gradients, baselines & actor–critic",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/ppo-updates/",
      "title": "PPO, clipping & on-policy data",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/rl-evaluation/",
      "title": "Does RL actually improve capability?",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "rl/rlhf-dpo/",
      "title": "SFT → RLHF → DPO: different training signals",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "systems/kv-serving/",
      "title": "KV memory, paging and serving deadlines",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "systems/moe-budgets/",
      "title": "MoE: active compute is not total memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "systems/speculative-correctness/",
      "title": "Speculative decoding: correctness before speed",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/eval-compare/",
      "title": "Paired evaluation comparison",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/jsonl-inspector/",
      "title": "JSONL dataset inspector",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/prompt-diff/",
      "title": "Prompt diff & token impact",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/rag-chunker/",
      "title": "RAG chunking lab",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/sampling-lab/",
      "title": "Sampling lab: logits → tokens",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/speculative-decoding/",
      "title": "Speculative decoding lab",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tools/vram-calculator/",
      "title": "GPU memory estimator",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tutorials/base64-ai-payload-inspector/",
      "title": "Inspect an AI payload: JSON, Base64, images, and audio",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tutorials/context-budget-rag/",
      "title": "Debug a context overflow: a practical RAG budget",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tutorials/multilingual-tokenization/",
      "title": "Why languages, emoji, and JSON change token counts",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "en",
      "path": "tutorials/tokens-and-tokenizers/",
      "title": "Tokens and tokenizers: from text to model input",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "applications/bounded-tools/",
      "title": "工具型助手：验证、执行与追踪",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "applications/document-evidence/",
      "title": "文件问答：实现证据优先的文档助手",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "applications/validated-extraction/",
      "title": "文档提取：JSON 格式正确不等于事实正确",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "frontier/adaptive-agent-evaluation/",
      "title": "Self-improving agents: reward gains are not capability gains",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "frontier/agent-harness-engineering/",
      "title": "From LLM Calls to Agent Harnesses",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "frontier/diffusion-language/",
      "title": "扩散语言模型：掩码与迭代解码",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "frontier/robot-recovery/",
      "title": "Robot recovery: when retrying helps, and when it hides failure",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "frontier/selective-state-spaces/",
      "title": "选择性状态空间：Mamba 与序列记忆",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "frontier/test-time-compute/",
      "title": "测试时计算：搜索、验证与预算",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/agent-1/",
      "title": "人工批准关卡",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/agent-2/",
      "title": "已批准的模拟写入",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/agent-3/",
      "title": "工具白名单",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/agent-4/",
      "title": "预算耗尽",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/agent-5/",
      "title": "记忆截断",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/attention-1/",
      "title": "因果注意力基线",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/attention-2/",
      "title": "双向上下文",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/attention-3/",
      "title": "尖锐注意力",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/attention-4/",
      "title": "均匀注意力",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/attention-5/",
      "title": "平滑注意力",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/cache-1/",
      "title": "局部性基线",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/cache-2/",
      "title": "容量抖动",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/cache-3/",
      "title": "TTL 到期",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/cache-4/",
      "title": "高复用负载",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/cache-5/",
      "title": "全部唯一请求",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/embedding-1/",
      "title": "余弦近邻",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/embedding-2/",
      "title": "点积范数偏差",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/embedding-3/",
      "title": "查询旋转",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/embedding-4/",
      "title": "反向向量",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/embedding-5/",
      "title": "零向量边界",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/judge-1/",
      "title": "混合裁判一致性",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/judge-2/",
      "title": "完全一致",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/judge-3/",
      "title": "总选第一个",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/judge-4/",
      "title": "平局占多数",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/judge-5/",
      "title": "稳定但错误",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/kv-1/",
      "title": "GQA 基线",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/kv-2/",
      "title": "多头注意力缓存",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/kv-3/",
      "title": "多查询注意力缓存",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/kv-4/",
      "title": "长上下文批次",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/kv-5/",
      "title": "分页碎片",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/neural-1/",
      "title": "学习 XOR 象限",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/neural-2/",
      "title": "隐藏层容量不足",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/neural-3/",
      "title": "小学习率",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/neural-4/",
      "title": "圆形分类",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/neural-5/",
      "title": "随机种子敏感性",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/quantization-1/",
      "title": "四位量化基线",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/quantization-2/",
      "title": "八位重建",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/quantization-3/",
      "title": "两位压力测试",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/quantization-4/",
      "title": "离群值支配尺度",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/quantization-5/",
      "title": "全零张量",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/retrieval-1/",
      "title": "混合融合基线",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/retrieval-2/",
      "title": "纯词法检索",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/retrieval-3/",
      "title": "纯向量检索",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/retrieval-4/",
      "title": "理想重排上界",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/retrieval-5/",
      "title": "候选集不足",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/routing-1/",
      "title": "两级路由",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/routing-2/",
      "title": "全部强模型",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/routing-3/",
      "title": "频繁回退",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/routing-4/",
      "title": "低成本模型故障",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "learn/routing-5/",
      "title": "主模型无故障",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/adam/",
      "title": "Adam：理解更新过程，而不只会使用默认参数",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/adamw/",
      "title": "AdamW：权重衰减加在哪里，结果就会不同",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/confidence-calibration/",
      "title": "校准：预测正确，与知道自己多大概率正确",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/contrastive-predictive-coding/",
      "title": "CPC：通过辨认正确未来学习表示",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/deep-rl-reproducibility/",
      "title": "Deep RL that Matters：分数背后是分布",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/deepseek-r1/",
      "title": "DeepSeek-R1：奖励、推理与证据边界",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/dpo/",
      "title": "DPO：通过相对对数概率学习偏好",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/flashattention/",
      "title": "FlashAttention：除了 FLOPs，还要算数据搬运",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/llm-as-a-judge/",
      "title": "LLM-as-a-judge：先评估评审者",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/lora/",
      "title": "LoRA：约束的是更新，而非整个模型",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/ppo/",
      "title": "PPO：逐样本看懂裁剪目标",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/qlora/",
      "title": "QLoRA：显存预算不能只看权重位宽",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/react/",
      "title": "ReAct：让行动与观察形成闭环",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/rlhf/",
      "title": "InstructGPT：改变训练信号，改变模型行为",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/show-your-work/",
      "title": "Show Your Work：比较搜索预算，而不只比较赢家",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/simclr/",
      "title": "SimCLR：数据增强定义了“相似”",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "papers/transformer-attention/",
      "title": "Transformer：把 token 匹配程度变成信息流",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/diffusion-policy/",
      "title": "05 · Diffusion Policy 与多模态动作",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/feedback-and-planning/",
      "title": "03 · 反馈控制与滚动时域规划",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/flow-matching-pi0/",
      "title": "07 · π₀ 与连续动作流匹配",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/frames-and-kinematics/",
      "title": "01 · 坐标系、变换与机器人运动学",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/imitation-and-act/",
      "title": "04 · 示教、行为克隆与 ACT",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/perception-and-state/",
      "title": "02 · 感知、状态估计与时间同步",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/rl-sim2real-evaluation/",
      "title": "08 · 机器人 RL、仿真迁移与部署证据",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "physical-ai/vla-openvla/",
      "title": "06 · OpenVLA：视觉、语言与动作接口",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "projects/attention/",
      "title": "因果注意力与 KV cache 一致性",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "projects/control/",
      "title": "延迟反馈与有界控制",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "projects/retrieval/",
      "title": "有基线与测量的证据检索",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "research/contrastive-learning/",
      "title": "对比学习与假负样本",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "research/experiment-design/",
      "title": "实验设计与统计证据",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "research/gradient-debugger/",
      "title": "梯度与训练故障诊断",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "research/information-theory/",
      "title": "信息论与分布差异",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "research/lora-rank/",
      "title": "LoRA 低秩适配实验",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "research/paper-reproduction/",
      "title": "论文复现工作台",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/agent-rl/",
      "title": "Agent RL：在执行框架中训练策略",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/exploration-credit/",
      "title": "探索、信用分配与奖励设计",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/grpo-verifiable/",
      "title": "GRPO、可验证奖励与推理",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/mdp-bellman/",
      "title": "MDP、回报与 Bellman 方程",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/policy-gradient/",
      "title": "策略梯度、基线与 Actor–Critic",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/ppo-updates/",
      "title": "PPO、裁剪与 On-policy 数据",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/rl-evaluation/",
      "title": "RL 是否真正提升了能力？",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "rl/rlhf-dpo/",
      "title": "SFT → RLHF → DPO：不同训练信号",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "systems/kv-serving/",
      "title": "KV 内存、分页与服务时限",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "systems/moe-budgets/",
      "title": "MoE：激活计算量不等于总内存",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "systems/speculative-correctness/",
      "title": "投机解码：先证明正确，再讨论加速",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/eval-compare/",
      "title": "配对评测结果对比",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/jsonl-inspector/",
      "title": "JSONL 数据集检查",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/prompt-diff/",
      "title": "Prompt 差异与 Token 变化",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/rag-chunker/",
      "title": "RAG 分块实验室",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/sampling-lab/",
      "title": "Sampling 实验：从 logits 到 Token",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/speculative-decoding/",
      "title": "Speculative Decoding 实验",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tools/vram-calculator/",
      "title": "GPU 显存估算",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tutorials/base64-ai-payload-inspector/",
      "title": "检查 AI Payload：串起 JSON、Base64、图片和音频",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tutorials/context-budget-rag/",
      "title": "排查上下文超限：一个实用的 RAG 预算",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tutorials/multilingual-tokenization/",
      "title": "为什么语言、Emoji 和 JSON 会改变 Token 数量",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "zh",
      "path": "tutorials/tokens-and-tokenizers/",
      "title": "Token 与 Tokenizer：从文本到模型输入",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "applications/bounded-tools/",
      "title": "Tool-using assistants: validate, execute and trace",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "applications/document-evidence/",
      "title": "File Q&A: build an evidence-first document assistant",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "applications/validated-extraction/",
      "title": "Document extraction: JSON shape is not factual correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "frontier/adaptive-agent-evaluation/",
      "title": "Self-improving agents: reward gains are not capability gains",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "frontier/agent-harness-engineering/",
      "title": "From LLM Calls to Agent Harnesses",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "frontier/diffusion-language/",
      "title": "Diffusion language models: masking and iterative decoding",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "frontier/robot-recovery/",
      "title": "Robot recovery: when retrying helps, and when it hides failure",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "frontier/selective-state-spaces/",
      "title": "Selective state spaces: Mamba and sequence memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "frontier/test-time-compute/",
      "title": "Test-time compute: search, verification and budgets",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/agent-1/",
      "title": "Aprobación humana",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/agent-2/",
      "title": "Escritura simulada autorizada",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/agent-3/",
      "title": "Lista de herramientas permitidas",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/agent-4/",
      "title": "Presupuesto agotado",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/agent-5/",
      "title": "Truncamiento de memoria",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/attention-1/",
      "title": "Atención causal",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/attention-2/",
      "title": "Contexto bidireccional",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/attention-3/",
      "title": "Atención concentrada",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/attention-4/",
      "title": "Atención uniforme",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/attention-5/",
      "title": "Atención suave",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/cache-1/",
      "title": "Localidad de referencia",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/cache-2/",
      "title": "Thrashing por capacidad",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/cache-3/",
      "title": "Caducidad del TTL",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/cache-4/",
      "title": "Carga con alta reutilización",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/cache-5/",
      "title": "Solicitudes únicas",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/embedding-1/",
      "title": "Vecinos por coseno",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/embedding-2/",
      "title": "Sesgo de norma del producto escalar",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/embedding-3/",
      "title": "Rotación de consulta",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/embedding-4/",
      "title": "Dirección opuesta",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/embedding-5/",
      "title": "Vector cero",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/judge-1/",
      "title": "Acuerdo mixto del juez",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/judge-2/",
      "title": "Acuerdo perfecto",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/judge-3/",
      "title": "Elegir siempre el primero",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/judge-4/",
      "title": "Datos con muchos empates",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/judge-5/",
      "title": "Error consistente",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/kv-1/",
      "title": "GQA de referencia",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/kv-2/",
      "title": "Caché multihead",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/kv-3/",
      "title": "Caché multiquery",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/kv-4/",
      "title": "Lote de contexto largo",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/kv-5/",
      "title": "Fragmentación de páginas",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/neural-1/",
      "title": "Aprender cuadrantes XOR",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/neural-2/",
      "title": "Capa oculta insuficiente",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/neural-3/",
      "title": "Aprendizaje lento",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/neural-4/",
      "title": "Clasificación circular",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/neural-5/",
      "title": "Sensibilidad a la semilla",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/quantization-1/",
      "title": "Cuantización de 4 bits",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/quantization-2/",
      "title": "Reconstrucción de 8 bits",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/quantization-3/",
      "title": "Prueba de 2 bits",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/quantization-4/",
      "title": "Escala dominada por un atípico",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/quantization-5/",
      "title": "Tensor de ceros",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/retrieval-1/",
      "title": "Fusión híbrida",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/retrieval-2/",
      "title": "Solo búsqueda léxica",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/retrieval-3/",
      "title": "Solo búsqueda densa",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/retrieval-4/",
      "title": "Límite del reranking oráculo",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/retrieval-5/",
      "title": "Falta de candidatos",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/routing-1/",
      "title": "Routing de dos niveles",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/routing-2/",
      "title": "Solo modelo potente",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/routing-3/",
      "title": "Fallback frecuente",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/routing-4/",
      "title": "Fallo del modelo económico",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "learn/routing-5/",
      "title": "Sin fallos primarios",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/adam/",
      "title": "Adam: understand the update, not just the defaults",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/adamw/",
      "title": "AdamW: where you apply weight decay matters",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/confidence-calibration/",
      "title": "Calibration: being right is different from knowing how often",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/contrastive-predictive-coding/",
      "title": "CPC: learn by identifying the right future",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/deep-rl-reproducibility/",
      "title": "Deep RL that Matters: a score is a distribution",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/deepseek-r1/",
      "title": "DeepSeek-R1: rewards, reasoning and what the evidence establishes",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/dpo/",
      "title": "DPO: preference learning through relative log probabilities",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/flashattention/",
      "title": "FlashAttention: optimize data movement before counting FLOPs",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/llm-as-a-judge/",
      "title": "LLM-as-a-judge: evaluate the evaluator",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/lora/",
      "title": "LoRA: constrain the update, not the whole model",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/ppo/",
      "title": "PPO: inspect the clipped objective one sample at a time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/qlora/",
      "title": "QLoRA: a memory budget is more than weight precision",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/react/",
      "title": "ReAct: close the loop between actions and observations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/rlhf/",
      "title": "InstructGPT: change the training signal to change behavior",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/show-your-work/",
      "title": "Show Your Work: compare search budgets, not just winners",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/simclr/",
      "title": "SimCLR: augmentation defines what similarity means",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "papers/transformer-attention/",
      "title": "Transformer: turn token compatibility into information flow",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/diffusion-policy/",
      "title": "05 · Diffusion Policy and multimodal actions",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/feedback-and-planning/",
      "title": "03 · Feedback control and receding-horizon planning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/flow-matching-pi0/",
      "title": "07 · π₀ and continuous action flow matching",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/frames-and-kinematics/",
      "title": "01 · Frames, transformations and robot kinematics",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/imitation-and-act/",
      "title": "04 · Demonstrations, behavior cloning and ACT",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/perception-and-state/",
      "title": "02 · Perception, state estimation and time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/rl-sim2real-evaluation/",
      "title": "08 · Robot RL, sim-to-real and deployment evidence",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "physical-ai/vla-openvla/",
      "title": "06 · OpenVLA: vision, language and action interfaces",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "projects/attention/",
      "title": "Causal attention & KV-cache correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "projects/control/",
      "title": "Delayed feedback and bounded control",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "projects/retrieval/",
      "title": "Evidence retrieval with a measured baseline",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "research/contrastive-learning/",
      "title": "Aprendizaje contrastivo",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "research/experiment-design/",
      "title": "Diseño de experimentos",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "research/gradient-debugger/",
      "title": "Depurador de gradientes",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "research/information-theory/",
      "title": "Laboratorio de teoría de la información",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "research/lora-rank/",
      "title": "Laboratorio de rango LoRA",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "research/paper-reproduction/",
      "title": "Reproducción de artículos",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/agent-rl/",
      "title": "Agent RL: train a policy inside a harness",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/exploration-credit/",
      "title": "Exploration, credit assignment & reward design",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/grpo-verifiable/",
      "title": "GRPO, verifiable rewards & reasoning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/mdp-bellman/",
      "title": "MDPs, returns & Bellman equations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/policy-gradient/",
      "title": "Policy gradients, baselines & actor–critic",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/ppo-updates/",
      "title": "PPO, clipping & on-policy data",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/rl-evaluation/",
      "title": "Does RL actually improve capability?",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "rl/rlhf-dpo/",
      "title": "SFT → RLHF → DPO: different training signals",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "systems/kv-serving/",
      "title": "KV memory, paging and serving deadlines",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "systems/moe-budgets/",
      "title": "MoE: active compute is not total memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "systems/speculative-correctness/",
      "title": "Speculative decoding: correctness before speed",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/eval-compare/",
      "title": "Comparación de evaluaciones pareadas",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/jsonl-inspector/",
      "title": "Inspector de datasets JSONL",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/prompt-diff/",
      "title": "Diferencias de prompts y tokens",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/rag-chunker/",
      "title": "Laboratorio de fragmentación RAG",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/sampling-lab/",
      "title": "Laboratorio de muestreo: logits → tokens",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/speculative-decoding/",
      "title": "Laboratorio de decodificación especulativa",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tools/vram-calculator/",
      "title": "Estimador de memoria GPU",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tutorials/base64-ai-payload-inspector/",
      "title": "Inspecciona un payload de IA: JSON, Base64, imágenes y audio",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tutorials/context-budget-rag/",
      "title": "Depura un desbordamiento: presupuesto práctico para RAG",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tutorials/multilingual-tokenization/",
      "title": "Por qué los idiomas, emojis y JSON cambian el recuento",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "es",
      "path": "tutorials/tokens-and-tokenizers/",
      "title": "Tokens y tokenizadores: del texto a la entrada del modelo",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "applications/bounded-tools/",
      "title": "Tool-using assistants: validate, execute and trace",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "applications/document-evidence/",
      "title": "File Q&A: build an evidence-first document assistant",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "applications/validated-extraction/",
      "title": "Document extraction: JSON shape is not factual correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "frontier/adaptive-agent-evaluation/",
      "title": "Self-improving agents: reward gains are not capability gains",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "frontier/agent-harness-engineering/",
      "title": "From LLM Calls to Agent Harnesses",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "frontier/diffusion-language/",
      "title": "Diffusion language models: masking and iterative decoding",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "frontier/robot-recovery/",
      "title": "Robot recovery: when retrying helps, and when it hides failure",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "frontier/selective-state-spaces/",
      "title": "Selective state spaces: Mamba and sequence memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "frontier/test-time-compute/",
      "title": "Test-time compute: search, verification and budgets",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/agent-1/",
      "title": "人間の承認ゲート",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/agent-2/",
      "title": "承認済み模擬書き込み",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/agent-3/",
      "title": "ツールの許可リスト",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/agent-4/",
      "title": "予算の枯渇",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/agent-5/",
      "title": "メモリの切り捨て",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/attention-1/",
      "title": "因果注意力の基準",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/attention-2/",
      "title": "双方向コンテキスト",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/attention-3/",
      "title": "集中した注意力",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/attention-4/",
      "title": "均一な注意力",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/attention-5/",
      "title": "滑らかな注意力",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/cache-1/",
      "title": "局所性の基準",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/cache-2/",
      "title": "容量不足のスラッシング",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/cache-3/",
      "title": "TTL の期限切れ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/cache-4/",
      "title": "高い再利用率",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/cache-5/",
      "title": "すべて異なるリクエスト",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/embedding-1/",
      "title": "コサイン近傍",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/embedding-2/",
      "title": "内積のノルム偏り",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/embedding-3/",
      "title": "クエリの回転",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/embedding-4/",
      "title": "反対方向のベクトル",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/embedding-5/",
      "title": "ゼロベクトル",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/judge-1/",
      "title": "評価者の混合一致率",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/judge-2/",
      "title": "完全一致",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/judge-3/",
      "title": "常に最初を選ぶ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/judge-4/",
      "title": "引き分けの多いデータ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/judge-5/",
      "title": "一貫して誤る評価者",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/kv-1/",
      "title": "GQA の基準",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/kv-2/",
      "title": "マルチヘッドキャッシュ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/kv-3/",
      "title": "マルチクエリキャッシュ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/kv-4/",
      "title": "長文コンテキストのバッチ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/kv-5/",
      "title": "ページの断片化",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/neural-1/",
      "title": "XOR 象限を学習",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/neural-2/",
      "title": "隠れ層の容量不足",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/neural-3/",
      "title": "小さな学習率",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/neural-4/",
      "title": "円の分類",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/neural-5/",
      "title": "乱数シードの影響",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/quantization-1/",
      "title": "4 ビット量子化",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/quantization-2/",
      "title": "8 ビット再構成",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/quantization-3/",
      "title": "2 ビットの限界",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/quantization-4/",
      "title": "外れ値によるスケール変化",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/quantization-5/",
      "title": "全ゼロテンソル",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/retrieval-1/",
      "title": "ハイブリッド融合",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/retrieval-2/",
      "title": "語彙検索のみ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/retrieval-3/",
      "title": "ベクトル検索のみ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/retrieval-4/",
      "title": "オラクル再順位付けの上限",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/retrieval-5/",
      "title": "候補の不足",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/routing-1/",
      "title": "二段階ルーティング",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/routing-2/",
      "title": "強力モデルのみ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/routing-3/",
      "title": "頻繁なフォールバック",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/routing-4/",
      "title": "低コストモデルの障害",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "learn/routing-5/",
      "title": "一次モデルの障害なし",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/adam/",
      "title": "Adam: understand the update, not just the defaults",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/adamw/",
      "title": "AdamW: where you apply weight decay matters",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/confidence-calibration/",
      "title": "Calibration: being right is different from knowing how often",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/contrastive-predictive-coding/",
      "title": "CPC: learn by identifying the right future",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/deep-rl-reproducibility/",
      "title": "Deep RL that Matters: a score is a distribution",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/deepseek-r1/",
      "title": "DeepSeek-R1: rewards, reasoning and what the evidence establishes",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/dpo/",
      "title": "DPO: preference learning through relative log probabilities",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/flashattention/",
      "title": "FlashAttention: optimize data movement before counting FLOPs",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/llm-as-a-judge/",
      "title": "LLM-as-a-judge: evaluate the evaluator",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/lora/",
      "title": "LoRA: constrain the update, not the whole model",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/ppo/",
      "title": "PPO: inspect the clipped objective one sample at a time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/qlora/",
      "title": "QLoRA: a memory budget is more than weight precision",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/react/",
      "title": "ReAct: close the loop between actions and observations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/rlhf/",
      "title": "InstructGPT: change the training signal to change behavior",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/show-your-work/",
      "title": "Show Your Work: compare search budgets, not just winners",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/simclr/",
      "title": "SimCLR: augmentation defines what similarity means",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "papers/transformer-attention/",
      "title": "Transformer: turn token compatibility into information flow",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/diffusion-policy/",
      "title": "05 · Diffusion Policy and multimodal actions",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/feedback-and-planning/",
      "title": "03 · Feedback control and receding-horizon planning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/flow-matching-pi0/",
      "title": "07 · π₀ and continuous action flow matching",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/frames-and-kinematics/",
      "title": "01 · Frames, transformations and robot kinematics",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/imitation-and-act/",
      "title": "04 · Demonstrations, behavior cloning and ACT",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/perception-and-state/",
      "title": "02 · Perception, state estimation and time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/rl-sim2real-evaluation/",
      "title": "08 · Robot RL, sim-to-real and deployment evidence",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "physical-ai/vla-openvla/",
      "title": "06 · OpenVLA: vision, language and action interfaces",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "projects/attention/",
      "title": "Causal attention & KV-cache correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "projects/control/",
      "title": "Delayed feedback and bounded control",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "projects/retrieval/",
      "title": "Evidence retrieval with a measured baseline",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "research/contrastive-learning/",
      "title": "対照学習ラボ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "research/experiment-design/",
      "title": "実験設計ラボ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "research/gradient-debugger/",
      "title": "勾配デバッガー",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "research/information-theory/",
      "title": "情報理論ラボ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "research/lora-rank/",
      "title": "LoRA ランク実験",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "research/paper-reproduction/",
      "title": "論文再現ワークベンチ",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/agent-rl/",
      "title": "Agent RL: train a policy inside a harness",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/exploration-credit/",
      "title": "Exploration, credit assignment & reward design",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/grpo-verifiable/",
      "title": "GRPO, verifiable rewards & reasoning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/mdp-bellman/",
      "title": "MDPs, returns & Bellman equations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/policy-gradient/",
      "title": "Policy gradients, baselines & actor–critic",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/ppo-updates/",
      "title": "PPO, clipping & on-policy data",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/rl-evaluation/",
      "title": "Does RL actually improve capability?",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "rl/rlhf-dpo/",
      "title": "SFT → RLHF → DPO: different training signals",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "systems/kv-serving/",
      "title": "KV memory, paging and serving deadlines",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "systems/moe-budgets/",
      "title": "MoE: active compute is not total memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "systems/speculative-correctness/",
      "title": "Speculative decoding: correctness before speed",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/eval-compare/",
      "title": "対応あり評価比較",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/jsonl-inspector/",
      "title": "JSONLデータセット検査",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/prompt-diff/",
      "title": "プロンプト差分とトークン変化",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/rag-chunker/",
      "title": "RAGチャンク実験室",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/sampling-lab/",
      "title": "サンプリング実験：logitsからトークンへ",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/speculative-decoding/",
      "title": "投機的デコーディング実験",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tools/vram-calculator/",
      "title": "GPUメモリー見積もり",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tutorials/base64-ai-payload-inspector/",
      "title": "AIペイロードを検査：JSON・Base64・画像・音声",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tutorials/context-budget-rag/",
      "title": "コンテキスト超過を調べる：実践的なRAG予算",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tutorials/multilingual-tokenization/",
      "title": "言語・絵文字・JSONでトークン数が変わる理由",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ja",
      "path": "tutorials/tokens-and-tokenizers/",
      "title": "トークンとトークナイザー：テキストからモデル入力へ",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "applications/bounded-tools/",
      "title": "Tool-using assistants: validate, execute and trace",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "applications/document-evidence/",
      "title": "File Q&A: build an evidence-first document assistant",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "applications/validated-extraction/",
      "title": "Document extraction: JSON shape is not factual correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "frontier/adaptive-agent-evaluation/",
      "title": "Self-improving agents: reward gains are not capability gains",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "frontier/agent-harness-engineering/",
      "title": "From LLM Calls to Agent Harnesses",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "frontier/diffusion-language/",
      "title": "Diffusion language models: masking and iterative decoding",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "frontier/robot-recovery/",
      "title": "Robot recovery: when retrying helps, and when it hides failure",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "frontier/selective-state-spaces/",
      "title": "Selective state spaces: Mamba and sequence memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "frontier/test-time-compute/",
      "title": "Test-time compute: search, verification and budgets",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/agent-1/",
      "title": "사람의 승인 단계",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/agent-2/",
      "title": "승인된 모의 쓰기",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/agent-3/",
      "title": "도구 허용 목록",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/agent-4/",
      "title": "예산 소진",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/agent-5/",
      "title": "메모리 잘라내기",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/attention-1/",
      "title": "인과 어텐션 기준",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/attention-2/",
      "title": "양방향 문맥",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/attention-3/",
      "title": "집중된 어텐션",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/attention-4/",
      "title": "균일한 어텐션",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/attention-5/",
      "title": "부드러운 어텐션",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/cache-1/",
      "title": "지역성 기준",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/cache-2/",
      "title": "용량 부족 스래싱",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/cache-3/",
      "title": "TTL 만료",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/cache-4/",
      "title": "높은 재사용 부하",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/cache-5/",
      "title": "모두 고유한 요청",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/embedding-1/",
      "title": "코사인 최근접",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/embedding-2/",
      "title": "내적의 크기 편향",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/embedding-3/",
      "title": "쿼리 회전",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/embedding-4/",
      "title": "반대 방향",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/embedding-5/",
      "title": "영벡터 경계",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/judge-1/",
      "title": "심판의 혼합 일치율",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/judge-2/",
      "title": "완벽한 일치",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/judge-3/",
      "title": "항상 첫 번째 선택",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/judge-4/",
      "title": "무승부가 많은 데이터",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/judge-5/",
      "title": "일관되게 틀리는 심판",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/kv-1/",
      "title": "GQA 기준",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/kv-2/",
      "title": "멀티헤드 캐시",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/kv-3/",
      "title": "멀티쿼리 캐시",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/kv-4/",
      "title": "긴 문맥 배치",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/kv-5/",
      "title": "페이지 단편화",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/neural-1/",
      "title": "XOR 사분면 학습",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/neural-2/",
      "title": "은닉층 용량 부족",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/neural-3/",
      "title": "느린 학습",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/neural-4/",
      "title": "원형 분류",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/neural-5/",
      "title": "시드 민감도",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/quantization-1/",
      "title": "4비트 양자화",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/quantization-2/",
      "title": "8비트 복원",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/quantization-3/",
      "title": "2비트 스트레스 테스트",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/quantization-4/",
      "title": "이상치가 지배하는 스케일",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/quantization-5/",
      "title": "모든 값이 0인 텐서",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/retrieval-1/",
      "title": "하이브리드 결합",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/retrieval-2/",
      "title": "어휘 검색만",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/retrieval-3/",
      "title": "벡터 검색만",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/retrieval-4/",
      "title": "오라클 재정렬 상한",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/retrieval-5/",
      "title": "후보 부족",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/routing-1/",
      "title": "2단계 라우팅",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/routing-2/",
      "title": "강력한 모델만",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/routing-3/",
      "title": "빈번한 폴백",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/routing-4/",
      "title": "저비용 모델 장애",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "learn/routing-5/",
      "title": "기본 모델 장애 없음",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Engine formulas, scenario presets and numerical safeguards reviewed. Seeded synthetic inputs are not a production benchmark.",
        "已复核引擎公式、场景预设及数值保护。固定种子的合成输入不是生产基准。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/adam/",
      "title": "Adam: understand the update, not just the defaults",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/adamw/",
      "title": "AdamW: where you apply weight decay matters",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/confidence-calibration/",
      "title": "Calibration: being right is different from knowing how often",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/contrastive-predictive-coding/",
      "title": "CPC: learn by identifying the right future",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/deep-rl-reproducibility/",
      "title": "Deep RL that Matters: a score is a distribution",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/deepseek-r1/",
      "title": "DeepSeek-R1: rewards, reasoning and what the evidence establishes",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/dpo/",
      "title": "DPO: preference learning through relative log probabilities",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/flashattention/",
      "title": "FlashAttention: optimize data movement before counting FLOPs",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/llm-as-a-judge/",
      "title": "LLM-as-a-judge: evaluate the evaluator",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/lora/",
      "title": "LoRA: constrain the update, not the whole model",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/ppo/",
      "title": "PPO: inspect the clipped objective one sample at a time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/qlora/",
      "title": "QLoRA: a memory budget is more than weight precision",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/react/",
      "title": "ReAct: close the loop between actions and observations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/rlhf/",
      "title": "InstructGPT: change the training signal to change behavior",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/show-your-work/",
      "title": "Show Your Work: compare search budgets, not just winners",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/simclr/",
      "title": "SimCLR: augmentation defines what similarity means",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "papers/transformer-attention/",
      "title": "Transformer: turn token compatibility into information flow",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Displayed formula, bilingual explanation and declared slider range. This code reproduces the scalar teaching calculation, not the paper training system.",
        "复核所示公式、中英解释及声明滑块范围。代码复现标量教学计算，不是论文训练系统。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/diffusion-policy/",
      "title": "05 · Diffusion Policy and multimodal actions",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/feedback-and-planning/",
      "title": "03 · Feedback control and receding-horizon planning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/flow-matching-pi0/",
      "title": "07 · π₀ and continuous action flow matching",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/frames-and-kinematics/",
      "title": "01 · Frames, transformations and robot kinematics",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/imitation-and-act/",
      "title": "04 · Demonstrations, behavior cloning and ACT",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/perception-and-state/",
      "title": "02 · Perception, state estimation and time",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/rl-sim2real-evaluation/",
      "title": "08 · Robot RL, sim-to-real and deployment evidence",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "physical-ai/vla-openvla/",
      "title": "06 · OpenVLA: vision, language and action interfaces",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "projects/attention/",
      "title": "Causal attention & KV-cache correctness",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "projects/control/",
      "title": "Delayed feedback and bounded control",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "projects/retrieval/",
      "title": "Evidence retrieval with a measured baseline",
      "components": {
        "diagram": false,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Implementation, invariants and declared evidence boundary reviewed; existing acceptance tests remain part of the suite.",
        "已复核实现、不变性及声明的证据边界；保留原有验收测试。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "research/contrastive-learning/",
      "title": "대조 학습",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "research/experiment-design/",
      "title": "실험 설계",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "research/gradient-debugger/",
      "title": "그래디언트 디버거",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "research/information-theory/",
      "title": "정보 이론 실험실",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "research/lora-rank/",
      "title": "LoRA 랭크 실험",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "research/paper-reproduction/",
      "title": "논문 재현 워크벤치",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Objective, derivatives, worked examples and implementation reviewed. A passing finite-difference or manifest check is not a paper reproduction certificate.",
        "已复核目标、导数、算例及实现。通过有限差分或清单检查不代表完成论文复现。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/agent-rl/",
      "title": "Agent RL: train a policy inside a harness",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/exploration-credit/",
      "title": "Exploration, credit assignment & reward design",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/grpo-verifiable/",
      "title": "GRPO, verifiable rewards & reasoning",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/mdp-bellman/",
      "title": "MDPs, returns & Bellman equations",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/policy-gradient/",
      "title": "Policy gradients, baselines & actor–critic",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/ppo-updates/",
      "title": "PPO, clipping & on-policy data",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/rl-evaluation/",
      "title": "Does RL actually improve capability?",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "rl/rlhf-dpo/",
      "title": "SFT → RLHF → DPO: different training signals",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "systems/kv-serving/",
      "title": "KV memory, paging and serving deadlines",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "systems/moe-budgets/",
      "title": "MoE: active compute is not total memory",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "systems/speculative-correctness/",
      "title": "Speculative decoding: correctness before speed",
      "components": {
        "diagram": true,
        "formula": true,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.",
        "已复核公式条件、单位及算例。实现隔离一个数值机制，不训练完整策略，也不控制真实硬件。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/eval-compare/",
      "title": "대응 평가 결과 비교",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/jsonl-inspector/",
      "title": "JSONL 데이터셋 검사",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/prompt-diff/",
      "title": "프롬프트 차이와 토큰 변화",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/rag-chunker/",
      "title": "RAG 청크 실험실",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/sampling-lab/",
      "title": "샘플링 실험: logits → 토큰",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/speculative-decoding/",
      "title": "추측 디코딩 실험",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": true
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tools/vram-calculator/",
      "title": "GPU 메모리 추정기",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Reviewed counting units, statistical assumptions and numerical examples. Provider accounting and actual GPU allocation still require external measurement.",
        "已复核计数单位、统计假设及数值例子；提供商计费和真实 GPU 分配仍须实测。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tutorials/base64-ai-payload-inspector/",
      "title": "AI 페이로드 검사: JSON, Base64, 이미지와 오디오",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tutorials/context-budget-rag/",
      "title": "컨텍스트 초과 디버깅: 실전 RAG 예산",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tutorials/multilingual-tokenization/",
      "title": "언어, 이모지, JSON에 따라 토큰 수가 달라지는 이유",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    },
    {
      "language": "ko",
      "path": "tutorials/tokens-and-tokenizers/",
      "title": "토큰과 토크나이저: 텍스트에서 모델 입력까지",
      "components": {
        "diagram": true,
        "formula": false,
        "code": true,
        "testDiscussion": true,
        "sources": false
      },
      "semanticReview": "mathematical-editorial-review",
      "reviewScope": [
        "Shared code, English mathematical explanation and key multilingual numeric passages reviewed. This is not a full translation proofread.",
        "已复核共享代码、英文数学解释及多语言关键数值段落；不代表完成逐字翻译校对。"
      ],
      "checked": "2026-10-26"
    }
  ]
}