---
date: 2026-06-05
phase: 2
headline: First deploys on new station — OK but eval not robust
---

## What we did
- Deployed models on the new station setup
- Results were OK but the evaluation setup wasn't robust
- Models still had a lot of room to go

## Reflection
- Want to test whether **data efficiency** is the actual bottleneck — prompting the weekend data-collection investigation. Collect more, see if more demos closes the gap.

## Next steps
- This weekend: collect more demos and see if data scale is the limiter