Anomaly Finder

Identify anomalies in the following data using multiple detection methods:

**Dataset context**: {{context}}
**Columns to analyze**: {{columns}}
**Data**:
{{data}}

Apply these methods and compare results:
1. **Statistical**: z-score (threshold > 3), modified z-score (MAD-based), IQR (1.5x and 3x)
2. **Distribution-based**: fit expected distribution, flag low-probability points
3. **Multivariate**: Mahalanobis distance for correlated features
4. **Domain-specific**: flag values that violate known business rules

For each anomaly found:
- Row/observation identifier
- Which methods flagged it (consensus scoring)
- Severity (mild/moderate/extreme)
- Likely explanation (data entry error, genuine outlier, measurement issue)
- Recommended action (investigate, remove, cap, keep)

Output as Python code with a summary table.