Cross-Check

Evaluate the reliability of AI analysis results and explicitly flag the points that a human must verify.

---

WHEN TRIGGERED - EXECUTE IMMEDIATELY

Cross-check the most recent analysis result or the specified content.

---

Workflow

Step 1: Identify the analysis target

Locate the analysis result to be checked. If the preceding conversation contains one, use that.

Step 2: Reliability assessment

For each claim / number, evaluate:

Item	Criterion
Data grounding	Can it be verified directly from the source data?
Reference lookup	Was it compared against any workspace reference documents under references/?
Logical consistency	Is the cause → effect reasoning sound?
Scope coverage	Is the analysis period / range sufficient?
Explicit assumptions	Any hidden assumptions?

Step 3: Output

■ Cross-check Results

✅ Verified (confirmable from data)
- {item}: {evidence}

⚠️ Needs verification (human must confirm on site)
- {item}: {reason} → how to verify: {method}

❌ Caution (insufficient data or assumption-based)
- {item}: {reason}

📊 Confidence: {high / medium / low}

Step 4: Recommendations

For each "needs verification" item, suggest a concrete way to confirm:

• Who to ask
• What additional data to pull
• What to check in the source system or workspace