Last recorded focus | Gnaroshi

Reading at the last update

How Seer and SimVLA expose model inputs, state, and action outputs differently.
Ways to reuse model state while keeping approximation and evaluation boundaries clear.

Building at the last update

A shared experiment layout for running VLA models with separate model, method, environment, and result settings.
Run manifests and environment snapshots that make experiment context easier to revisit.

Questions at the last update

View research questions

Which parts of a VLA experiment can be shared across models?
Which representations can be reused without changing behavior in ways I cannot see?
How much context is enough to understand an old experiment later?