- Add wanderlust_ml_evaluation.ipynb: full ML workflow (EDA, training, evaluation, weight analysis)
- Add sensitivity_analysis.py: Monte Carlo ±30% perturbation showing budget tier invariance
- Update evaluation_report.md: Wave 4 results (intent acc 93.05%, macro F1 94.00% on 13 classes, NER 0.920)
- Document 4 intents with 0 test samples and their impact on 17-class macro F1 (0.76)
- Include +Tour academic reference (Esper et al., Computer Networks 2025) for itinerary optimization
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
