{
  "schema": "insight.wwdc.priority-claim-evidence.zh.v1",
  "session_id": "299",
  "title": "Create robust evaluations for agentic apps",
  "title_zh": "Agent / 评估：Create robust evaluations for agentic apps",
  "priority": "P1",
  "public_policy": "claim_evidence_safe_micro_quotes_no_full_transcript_or_translation",
  "quote_budget": {
    "max_words_per_session": 24,
    "used_words": 20,
    "max_words_per_claim": 5
  },
  "source_paths": {
    "refined_summary": "data/priority_session_refined_zh/299.json",
    "safe_brief": "data/session_briefs_zh/299.json",
    "workbench": "workbench.html?view=sessions/299/session_299_workbench.html"
  },
  "claim_evidence": [
    {
      "claim_id": "r1",
      "title_zh": "多步评估",
      "claim_zh": "Agentic app 不能只评估单轮输出，还要评估工具选择、步骤顺序和中间状态。",
      "evidence_grade": "A-",
      "grade_reason_zh": "P0/P1 refined claim with transcript-indexed anchors and controlled micro-quote; official wording still needs final human quote check.",
      "official_terms": [
        "Evaluations",
        "Xcode",
        "MLX"
      ],
      "anchor_path": [
        {
          "id": "a1",
          "time": 8,
          "label": "0:08",
          "sourceSegment": 0
        },
        {
          "id": "a2",
          "time": 266,
          "label": "4:26",
          "sourceSegment": 79
        },
        {
          "id": "a3",
          "time": 527,
          "label": "8:47",
          "sourceSegment": 158
        }
      ],
      "micro_quote_en": "watch the video Build agentic",
      "micro_quote_word_count": 5,
      "micro_quote_source_segment": 136,
      "support_rationale_zh": "该判断由 3 个安全锚点支撑，并已对齐术语：Evaluations / Xcode / MLX；发布前仍需人工核对短引文和 Apple 官方上下文。",
      "review_actions_zh": [
        "核对 micro-quote 是否准确表达当前 claim。",
        "补充或替换为更贴近 claim 的官方短引文，保持每个 session 总引文预算不超过 24 个英文词。",
        "确认该 claim 是否应上升为主题级判断，或只保留在 session 级。"
      ]
    },
    {
      "claim_id": "r2",
      "title_zh": "鲁棒性",
      "claim_zh": "鲁棒评估需要覆盖异常输入、工具失败、权限缺失和用户中断。",
      "evidence_grade": "A-",
      "grade_reason_zh": "P0/P1 refined claim with transcript-indexed anchors and controlled micro-quote; official wording still needs final human quote check.",
      "official_terms": [
        "Evaluations",
        "Xcode",
        "MLX"
      ],
      "anchor_path": [
        {
          "id": "a2",
          "time": 266,
          "label": "4:26",
          "sourceSegment": 79
        },
        {
          "id": "a3",
          "time": 527,
          "label": "8:47",
          "sourceSegment": 158
        },
        {
          "id": "a4",
          "time": 780,
          "label": "13:00",
          "sourceSegment": 238
        }
      ],
      "micro_quote_en": "watch the video Build agentic",
      "micro_quote_word_count": 5,
      "micro_quote_source_segment": 136,
      "support_rationale_zh": "该判断由 3 个安全锚点支撑，并已对齐术语：Evaluations / Xcode / MLX；发布前仍需人工核对短引文和 Apple 官方上下文。",
      "review_actions_zh": [
        "核对 micro-quote 是否准确表达当前 claim。",
        "补充或替换为更贴近 claim 的官方短引文，保持每个 session 总引文预算不超过 24 个英文词。",
        "确认该 claim 是否应上升为主题级判断，或只保留在 session 级。"
      ]
    },
    {
      "claim_id": "r3",
      "title_zh": "产品发布",
      "claim_zh": "如果 agent 不能稳定完成关键任务，就应限制能力范围或增加人工确认。",
      "evidence_grade": "A-",
      "grade_reason_zh": "P0/P1 refined claim with transcript-indexed anchors and controlled micro-quote; official wording still needs final human quote check.",
      "official_terms": [
        "Evaluations",
        "Xcode",
        "MLX"
      ],
      "anchor_path": [
        {
          "id": "a3",
          "time": 527,
          "label": "8:47",
          "sourceSegment": 158
        },
        {
          "id": "a4",
          "time": 780,
          "label": "13:00",
          "sourceSegment": 238
        },
        {
          "id": "a5",
          "time": 1016,
          "label": "16:56",
          "sourceSegment": 317
        }
      ],
      "micro_quote_en": "watch the video Build agentic",
      "micro_quote_word_count": 5,
      "micro_quote_source_segment": 136,
      "support_rationale_zh": "该判断由 3 个安全锚点支撑，并已对齐术语：Evaluations / Xcode / MLX；发布前仍需人工核对短引文和 Apple 官方上下文。",
      "review_actions_zh": [
        "核对 micro-quote 是否准确表达当前 claim。",
        "补充或替换为更贴近 claim 的官方短引文，保持每个 session 总引文预算不超过 24 个英文词。",
        "确认该 claim 是否应上升为主题级判断，或只保留在 session 级。"
      ]
    },
    {
      "claim_id": "r4",
      "title_zh": "治理价值",
      "claim_zh": "评估结果应反过来定义权限、提示、工具 schema 和 fallback 策略。",
      "evidence_grade": "A-",
      "grade_reason_zh": "P0/P1 refined claim with transcript-indexed anchors and controlled micro-quote; official wording still needs final human quote check.",
      "official_terms": [
        "Evaluations",
        "Xcode",
        "MLX"
      ],
      "anchor_path": [
        {
          "id": "a4",
          "time": 780,
          "label": "13:00",
          "sourceSegment": 238
        },
        {
          "id": "a5",
          "time": 1016,
          "label": "16:56",
          "sourceSegment": 317
        },
        {
          "id": "a6",
          "time": 1277,
          "label": "21:17",
          "sourceSegment": 396
        }
      ],
      "micro_quote_en": "watch the video Build agentic",
      "micro_quote_word_count": 5,
      "micro_quote_source_segment": 136,
      "support_rationale_zh": "该判断由 3 个安全锚点支撑，并已对齐术语：Evaluations / Xcode / MLX；发布前仍需人工核对短引文和 Apple 官方上下文。",
      "review_actions_zh": [
        "核对 micro-quote 是否准确表达当前 claim。",
        "补充或替换为更贴近 claim 的官方短引文，保持每个 session 总引文预算不超过 24 个英文词。",
        "确认该 claim 是否应上升为主题级判断，或只保留在 session 级。"
      ]
    }
  ],
  "quality": {
    "generation": "claim_evidence_enhancement_v1",
    "requires_human_quote_review": true,
    "requires_official_term_check": true,
    "generated_at": "2026-06-22"
  }
}