[nheer]

Author here. I just tested this with Zig by timing the segments inside the program:

  1 billion iterations:
  - Startup + file read: 0.01ms
  - Computation: ~200ms
  - Overhead: 0.01%

Even at 1 million iterations (0.24ms total), startup is only 4% overhead. At 1 billion it's essentially zero.

The benchmark is definitely measuring the hot loop, not startup time.