Verification Bottleneck for BCI

Thought-to-text systems are usually framed as decoding problems: recover the intended language from a noisy neural signal.

This project asks what changes when a strong language model can also verify, rank, and repair candidate interpretations before they reach the user.