Data Analysis

• Always use Python: pandas, numpy, and matplotlib are pre-installed.
• One call = one complete script: Each code_execute call runs in a fresh, isolated process. No variables or data persist between calls. Include ALL imports, data loading, and analysis in a single call. Never split related analysis across multiple calls.
• For output files (charts, CSVs): code_execute can write output files to disk; use read to view them
• Additional packages: Use bash to run pip install <package> if a specialized library is needed

Only use write + bash when the script itself needs to be saved for reuse.

Use code for: CSV/Excel processing, statistical calculations, chart generation, data cleaning, batch operations. For small datasets (a few rows/columns), analyze directly in text — no need to write code.

3. Common analysis methods

Descriptive statistics

• Central tendency: Mean, median, mode
• Dispersion: Standard deviation, range, interquartile range
• Distribution: Min, max, percentiles

Trend analysis

• Time series trends (growth, decline, volatility)
• Year-over-year (YoY) growth rate
• Month-over-month (MoM) change rate
• Moving averages

Comparative analysis

• Absolute value comparison (bar charts)
• Proportion comparison (pie/stacked charts)
• Ranking changes
• Variance calculations

Composition analysis

• Share of each component in the total
• Pareto analysis (80/20 rule)
• Structural change over time

Correlation analysis

• Whether two metrics are related
• Positive / negative / no clear correlation

4. Output format

Report structure

1. Data overview: Source, scope, field descriptions
2. Key findings: 3-5 insights (most important first)
3. Detailed analysis: Broken down by dimension, with calculations shown
4. Visualization recommendations: Suggest appropriate chart types (see below)
5. Conclusions and recommendations: Actionable advice based on data

Data presentation

• Use Markdown tables for key data
• Use reasonable precision (2 decimal places for currency, 1 for percentages)
• Format large numbers for readability (e.g., "1.2M" instead of "1200000")
• Mark changes with +/- signs

Chart recommendations

5. Common pitfalls

• Mean trap: Averages can hide outliers — always check the median too
• Base effect: Small base numbers make percentage changes misleading ("200% growth" might be 1 to 3)
• Correlation is not causation: Two metrics moving together doesn't mean one causes the other
• Cherry-picking: Present the complete picture, not just favorable data
• Time window bias: Different time ranges can lead to different conclusions

6. Quality checklist

• Are calculations correct? (Double-check key numbers)
• Are units consistent throughout?
• Are YoY/MoM comparisons clearly labeled?
• Are conclusions supported by data?
• Are recommendations actionable?