How would you prevent type instability in a large Julia codebase?

You enforce strict type annotations at API boundaries, avoid global state, and continuously profile inference results.

Type instability arises when a variable can hold multiple types across execution paths. In large systems, this spreads quickly through function composition. Prevention requires enforcing concrete types in performance-critical functions, using parametric types, and validating with @code_warntype during CI. Architecture-level discipline is required, not just local fixes.