Functional Domain Modeling in TypeScript

Robust server-side TypeScript design with better-result (Result, TaggedError) and valibot (schemas, branded types).

Invariants

Apply these throughout:

• Model each entity state as its own Readonly<> type; discriminate with kind project-wide.
• Define domain types with type, never interface; write functions in type definitions with function property notation (save: (t: Task) => Promise<void>).
• Express state transitions as pure functions whose argument types constrain valid source states and whose return types name the target state. Invalid transitions must be compile errors.
• Errors are values: domain code returns Result<T, E> and never throws. Define each known failure as a TaggedError class; reserve panic for violated invariants and programmer mistakes.
• Compose fallible steps with Result.gen by default; use a lone combinator only for a single transform. Avoid long combinator chains.
• Validate every external input (API request, DB row, env var, file, queue message) with a valibot schema at the boundary; trust types inside the domain — do not re-validate. When a schema already defines a boundary representation, derive its type with v.InferOutput/v.InferInput instead of restating the shape.
• as is banned except as const and as const satisfies T. When a value's type is unknown, parse it through a schema. Brand IDs with v.brand so no cast is ever needed.

Topics

Read only the reference file(s) relevant to the current task.