# Small-Batch Change Discipline

- Ship one concern per change. A feature, a refactor and a formatting pass
  are three changes, never one.
- Keep a change small enough that a reviewer can hold it in their head.
  AI assistance tends to grow changes faster than review capacity grows;
  small batches are the countermeasure.
- Slice large work into a sequence of independently reviewable, individually
  releasable increments. Prefer a stack of small diffs over one large diff.
- State the slicing plan before you start: which increment lands first, what
  each one proves, and what stays out.
- Never batch an unrelated cleanup into a feature change. Park it in the
  deferred-work list or make it its own change.
- Merge to the main line frequently from short-lived branches. Long-lived
  branches accumulate merge risk faster than they accumulate value.
- If a generated change grows past what its description promised, stop and
  split it instead of finishing big.