data-ground-truth

name	data-ground-truth
description	Before presenting numbers in reports or recommendations, verify facts and check values against industry baselines.
version	1.0.0
metadata	{"openclaw":{"emoji":"📊","homepage":"https://agentutil.net","always":false}}

data-ground-truth

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.