Evaluation Protocols and Leakage-Safe Benchmarking

"A score I could not trust was worse than no score at all, so I learned to grade myself without cheating."

An Honestly Graded Embodied AI Agent

Placeholder chapter page. Sections are produced by the book-skills pipeline.

Sections

1. 65.1 Task metrics vs operational metrics
2. 65.2 User/site/device/session splits
3. 65.3 Time-aware cross-validation
4. 65.4 Rare-event and imbalanced evaluation
5. 65.5 Robustness to missing and faulty sensors
6. 65.6 Calibration and uncertainty metrics
7. 65.7 Reproducibility and field-validation protocols