You pick the direction. These are the problems we think are most neglected relative to how much they matter.

Both count. A benchmark nobody had built and a harness that makes an existing claim testable are both outputs we can route onward.

The list is where we think the gaps are. If you have spent a decade inside a problem we did not think to name, that is a better reason to work on it than anything written here.

Bring it in week one and let the group try to break it.