---
date: 2026-06-28
phase: 4
headline: More evals + figure making
---

## What we did
- More eval runs continuing the post-CupSpill push.
- Figure-making — iterating on Fig 4 / method diagram / overview figure for the paper.
- Side: refined the BIND abstract opener with a Hamming-style hook ("Modern vision models recognize a cup zero-shot. Why do robot policies built on them still need thousands of demonstrations to pick one up?") — primes CupSpill being the headline result later in the paper.

## Next steps
- Continue figure compile + eval coverage.
- Annie hard-mode kickoff Mon Jun 29.