dotnet-coding

Model expected failures as data in the return type using Fin<A> for short-circuiting failures or Validation<Error, A> for accumulating independent failures. Declare each error as a sealed record extending Expected with a message and code, and reserve exceptions for developer defects.

Use Map for pure transformations, Bind or LINQ from clauses for dependent steps, and guard on Fin to attach a named error to a failing condition. Convert between result types only at one named boundary using ToFin, ToValidation, or ToOption.

Use IO<A> to describe a side effect that performs nothing until the host runs it, which keeps fallback and retry composable. Adapt task-returning operations through IO.liftAsync, and run the effect once at the host with RunSafe to get a Fin for translation.

Yes, inject the narrowest function that represents the consumer's need, such as Func<Instant> for a clock or T -> IO<Unit> for persistence. For consumers reading many capabilities, use Eff<RT, A> with a runtime record implementing one Has trait per capability.

C# does not optimize tail calls, so unbounded direct recursion can overflow the stack. Use Trampoline<A> for pure unbounded transitions, Monad.recur for effectful loops needing only the final value, or LanguageExt.List.unfold when intermediate states matter.

Use the Thinktecture [Union] attribute on an abstract partial record with one sealed record case per valid state. Every Switch lacking an arm for a new case fails to compile, which keeps pattern matching exhaustive as the domain evolves.