type-strengthener

Run a mutation-based audit that applies plausible edits—renaming fields, changing types, adding enum variants—to a temporary copy of your code and reruns your type checker. Consumers that receive no new error reveal gaps where stronger types would help.

Create a small probe file with one valid use and one plausible misuse of the type design, then run the project's exact checker command. Use assert_type or reveal_type when inference matters, and test both branches for narrowing functions.

The audit uses whichever checker the project already configures rather than introducing a new one. The mutation script parses pyright and basedpyright JSON output directly, and supports any other checker through a custom command template.

Use NewType for a low-cost static distinction between same-typed values like UserId and OrderId when no runtime behavior is needed. Choose a frozen dataclass or validated value object when construction must enforce rules at runtime.

Common causes include Any propagation, unchecked casts, excluded paths, nonblocking CI jobs, and untyped libraries erasing type information. Confirm coverage by adding a deliberate invalid assignment in a temporary copy and verifying the checker reports it.

Support must be verified separately for syntax parsing, runtime objects, typing_extensions backports, checker behavior, and runtime annotation consumers. The audit builds a capability profile and probes each feature with the project's exact versions before recommending it.