Nitzan Pomerantz
0e99dfddd2
Fix: Lower outlier detection threshold from 10 to 5 deals
**Problem:** Outlier detection was skipped for 9-deal datasets due to
threshold=10, allowing extreme outliers (₪920K vs ₪1.75M median) to
skew valuation results.
**Root Cause:** Real-world filtered queries (e.g., 3 rooms, ~60m²)
often yield 5-9 comparables. Small samples NEED outlier filtering more
than large ones (one outlier = 11-20% of dataset).
**Solution:** Lower ANALYSIS_MIN_DEALS_FOR_OUTLIER_DETECTION: 10 → 5
**Impact:**
- IQR & percentage methods valid with 5+ deals
- Real estate valuation uses 5-10 comparables standard
- Verified: 9-deal test now removes 2 outliers correctly
**Changes:**
- nadlan_mcp/config.py: Default threshold 10 → 5
- CLAUDE.md: Updated config documentation
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
2025-11-30 23:54:00 +02:00
..
2025-11-29 14:44:14 +02:00
2025-10-30 22:24:40 +02:00
2025-11-30 23:54:00 +02:00
2025-11-28 22:40:29 +02:00
2025-10-30 22:24:40 +02:00