takA tachometer for code

A half-baked experiment in measuring CLI work with retired instruction counts.

Read the experiment

CLI reference

Deterministic first

Instruction counts vary by roughly 0.02% between runs, making small regressions visible where wall time cannot.

Timing is context, not a gate

Wall-clock measurements are recorded, but contention makes them too noisy for a tight CI threshold.

History stays in git

Measurements are JSON lines in refs/notes/tak, with no database, account, or hosted service.

GO AWAY

tak is a half-baked experiment. Do not use it, depend on it, package it, or expect support, stability, or a roadmap. If you actually need to benchmark a command, use hyperfine.