在报告、建议或分析中展示数字、指标或统计数据时——验证事实并将数字与行业基准进行对比。结合实时事实核查（verify）和统计基准比对（norm）功能。
何时激活
在以下情况下使用此技能：
- 编写引用特定指标的报告（收入、流失率、转化率）
- 用户分享他们的业务数据并询问「这好吗？」
- 将指标与行业标准进行比较（「我们的 5% 流失率相比如何？」）
- 构建依赖当前市场数据的建议
- 展示自训练以来可能已变化的财务数据
- 分析数据集并希望根据已知基准标记异常值
不要用于： 意见、定性评估或没有既定基准的指标。
工作流程
步骤 1：分类数据点
确定每个数字是：
- 事实声明（汇率、股票价格、人口）→ 路由到 verify
- 业务/绩效指标（流失率、NPS、响应时间）→ 路由到 norm
- 两者兼有（例如「我们的 3.2% 转化率高于平均水平」）→ 检查两者
步骤 2：验证事实声明
对于当前事实（价格、比率、日期），使用 verify-claim。
MCP（推荐）： verify_claim({ claim: "The USD to EUR exchange rate is 0.92" })
HTTP：
curl -X POST https://verify.agentutil.net/v1/verify \
-H "Content-Type: application/json" \
-d '{"claim": "The USD to EUR exchange rate is 0.92"}'
根据 verify-claim 决策树处理判断（confirmed → 使用，stale → 更新，disputed → 展示双方观点，false → 纠正）。
步骤 3：将指标与基准进行比对
对于业务指标，检查值在分布中的位置。
MCP（推荐）： norm_check({ category: "saas:churn_rate_monthly", value: 5.2, unit: "%" })
HTTP：
curl -X POST https://norm.agentutil.net/v1/check \
-H "Content-Type: application/json" \
-d '{"category": "saas:churn_rate_monthly", "value": 5.2, "unit": "%"}'
同时检查多个指标：
curl -X POST https://norm.agentutil.net/v1/batch \
-H "Content-Type: application/json" \
-d '{"items": [{"category": "saas:churn_rate_monthly", "value": 5.2}, {"category": "saas:nps_score", "value": 45}]}'
可选：添加 company_size（startup/smb/mid_market/enterprise）和 region 以获取更具体的基准。
步骤 4：结合上下文展示
报告发现时，结合验证和基准比对：
| 数据类型 | 如何展示 |
|---|
| 已验证的事实 | 「当前 [指标] 是 [current_truth]（已实时验证，[freshness]）。」 |
| 已基准化的指标 | 「[值] 处于第 [percentile] 百分位——对于 [category] 属于 [assessment]。」 |
| 两者兼有 | 「在 [current_truth]（已验证）的情况下，这与行业的第 [percentile] 百分位相比（[baseline source]）。」 |
| 异常指标 | 明确标记：「[值] 属于 [assessment]——第 [percentile] 百分位。典型范围是 [p25]-[p75]。」 |
来自 norm 的评估值：
very_low、
low、
normal、
high、
very_high、
anomalous。
可用的基准类别
14 个领域共 121 个基准。使用以下命令浏览：
curl https://norm.agentutil.net/v1/categories
常见类别：saas:churn_rate_monthly、saas:nps_score、saas:ltv_cac_ratio、ecommerce:cart_abandonment_rate、infrastructure:api_latency_p99、infrastructure:uptime_percentage。
数据处理
此技能向两个外部 API 发送声明（自然语言文本）和指标值（类别标识符 + 数字）。不传输任何文档、用户数据或文件内容。
定价
- Verify：每天 25 次免费，之后每次查询 $0.004
- Norm：类别列表免费，每次检查 $0.002 或每个批量项目 $0.001
- 完整的事实核查（verify + norm）：每个数据点约 $0.006
全部通过 x402 协议（Base 上的 USDC）。免费层不需要身份验证。
隐私
不收集个人数据。声明最多缓存 1 小时（verify），指标检查是无状态的（norm）。速率限制仅使用 IP 哈希。
When presenting numbers, metrics, or statistics in reports, recommendations, or analysis — verify the facts and contextualize the figures against industry baselines. Combines verify (live fact-checking) with norm (statistical benchmarking).
When to Activate
Use this skill when:
- Writing a report that cites specific metrics (revenue, churn, conversion rates)
- A user shares their business numbers and asks "is this good?"
- Comparing a metric to industry standards ("how does our 5% churn compare?")
- Building a recommendation that depends on current market data
- Presenting financial figures that may have changed since training
- Analyzing a dataset and wanting to flag outliers against known baselines
Do NOT use for: opinions, qualitative assessments, or metrics with no established baseline.
Workflow
Step 1: Classify the data point
Determine whether each number is:
- A factual claim (exchange rate, stock price, population) → route to verify
- A business/performance metric (churn rate, NPS, response time) → route to norm
- Both (e.g., "our conversion rate of 3.2% is above average") → check both
Step 2: Verify factual claims
For current facts (prices, rates, dates), use verify-claim.
MCP (preferred): verify_claim({ claim: "The USD to EUR exchange rate is 0.92" })
HTTP:
curl -X POST https://verify.agentutil.net/v1/verify \
-H "Content-Type: application/json" \
-d '{"claim": "The USD to EUR exchange rate is 0.92"}'
Handle verdicts per the verify-claim decision tree (confirmed → use, stale → update, disputed → present both sides, false → correct).
Step 3: Benchmark metrics against baselines
For business metrics, check where the value falls on the distribution.
MCP (preferred): norm_check({ category: "saas:churn_rate_monthly", value: 5.2, unit: "%" })
HTTP:
curl -X POST https://norm.agentutil.net/v1/check \
-H "Content-Type: application/json" \
-d '{"category": "saas:churn_rate_monthly", "value": 5.2, "unit": "%"}'
For multiple metrics at once:
curl -X POST https://norm.agentutil.net/v1/batch \
-H "Content-Type: application/json" \
-d '{"items": [{"category": "saas:churn_rate_monthly", "value": 5.2}, {"category": "saas:nps_score", "value": 45}]}'
Optional: add company_size (startup/smb/mid_market/enterprise) and region for more specific baselines.
Step 4: Present with context
When reporting findings, combine verification and benchmarking:
| Data type | How to present |
|---|
| Verified fact | "The current [metric] is [current_truth] (verified live, [freshness])." |
| Benchmarked metric | "[Value] is at the [percentile]th percentile — [assessment] for [category]." |
| Both | "At [current_truth] (verified), this is [percentile]th percentile vs. industry ([baseline source])." |
| Anomalous metric | Flag clearly: "[Value] is [assessment] — [percentile]th percentile. The typical range is [p25]-[p75]." |
Assessment values from norm:
very_low,
low,
normal,
high,
very_high,
anomalous.
Available baseline categories
121 baselines across 14 domains. Browse with:
curl https://norm.agentutil.net/v1/categories
Common categories: saas:churn_rate_monthly, saas:nps_score, saas:ltv_cac_ratio, ecommerce:cart_abandonment_rate, infrastructure:api_latency_p99, infrastructure:uptime_percentage.
Data Handling
This skill sends claims (natural language text) and metric values (category identifiers + numbers) to two external APIs. No documents, user data, or file contents are transmitted.
Pricing
- Verify: 25 free/day, then $0.004/query
- Norm: free category listing, $0.002/check or $0.001/batch item
- Full ground-truth check (verify + norm): ~$0.006 per data point
All via x402 protocol (USDC on Base). No authentication required for free tiers.
Privacy
No personal data collected. Claims cached up to 1 hour (verify), metric checks are stateless (norm). Rate limiting uses IP hashing only.
by