Spreadsheet Insights Generator

# Task
Analyze the data summary below. Separate what the data confirms from what it merely suggests.

# Inputs
- Data summary: {data_summary}
- Business context: {business_context}
- Goal: {goal}
- Audience: {audience}
- Constraints: {constraints}

# Analysis rules
1. Use ONLY the provided data. Do not invent benchmarks, industry averages, or comparison points.
2. For every finding, classify its confidence:
   - **Confirmed**: Clear pattern with sufficient data points.
   - **Directional**: Trend suggested but sample/timeframe may be too small.
   - **Anomaly**: Unexpected data point that needs investigation.
3. Always ask "compared to what?" — if there's no baseline in the data, say so rather than inventing one.
4. Do not conflate correlation with causation. If a relationship is correlational, label it.

# Output structure
1. **Executive summary** (3-5 sentences): The most important finding, its confidence level, and what to do about it.
2. **Key findings** (5-8 items): Each with:
   - Finding statement
   - Confidence: Confirmed / Directional / Anomaly
   - Supporting data point(s)
   - Business implication (1 sentence)
3. **Trends** (2-3): Patterns over time or across segments. Note direction and rate of change.
4. **Anomalies** (1-3): Unexpected values worth investigating. Suggest possible explanations without asserting one.
5. **Visualization suggestions**: For each key finding, suggest the chart type that would best communicate it (bar, line, scatter, heatmap, etc.) and what goes on each axis.
6. **What's missing**: Data points or context that would strengthen the analysis.
7. **Recommended actions** (2-3): What to do based on the findings. Tie each action to a specific finding.

# Output
Return all 7 sections.

<source>{data_summary}</source>
<context>
  <business_context>{business_context}</business_context>
  <goal>{goal}</goal>
  <audience>{audience}</audience>
  <constraints>{constraints}</constraints>
</context>
<task>Analyze the data above. Separate confirmed patterns from directional signals.</task>
<instructions>
  Only provided data. Classify confidence: Confirmed / Directional / Anomaly. Ask "compared to what?" — don't invent baselines. Label correlations as correlational.
</instructions>
<output>
  Return: executive summary, key findings (5-8 with confidence + data + implication), trends (2-3), anomalies (1-3 with possible explanations), visualization suggestions (chart type per finding), what's missing, recommended actions (2-3 tied to findings).
</output>

Data summary: {data_summary}
Business context: {business_context}
Goal: {goal}
Audience: {audience}
Constraints: {constraints}

Rules: only provided data, classify confidence (Confirmed/Directional/Anomaly), don't invent baselines, label correlations.

Based on the entire content above, return:
1. Executive summary (top finding + action)
2. Key findings (5-8 with confidence, data, implication)
3. Trends (2-3)
4. Anomalies (1-3 with possible explanations)
5. Visualization suggestions (chart type per finding)
6. What's missing
7. Recommended actions (2-3 tied to findings)