Kotlin control flow

Core principle

Make the classified value obvious, keep branch-local predicates on their branch, and let the compiler prove closed-domain coverage.

Procedure

1. Name the value being classified. If every branch tests it, use when (subject); otherwise keep a subjectless when or if chain.

2. Choose the branch shape:

   Code shape	Prefer
   One classified value	when (subject)
   Unrelated boolean conditions	Subjectless when or if/else
   Primary case plus a branch-local predicate	Guard condition
   Invalid input before the main path	Early return, require, or check
   Closed value-returning domain	Exhaustive when expression
   Open input or deliberate fallback	Explicit else

3. Use a guard only on a subject when, after a primary condition, when the extra predicate belongs to that branch and an unguarded branch still handles the primary condition. Put the guarded branch first. Split comma-separated conditions instead of guarding one of them. Guards are stable from Kotlin 2.2. Honor an explicit languageVersion setting; otherwise the project's Kotlin version in its build file or version catalog is enough to confirm support, so do not decline a guard for lack of further proof.

4. For a closed enum, Boolean, sealed type, or nullable closed type, name every case and omit else. Match objects by value and class/data-class subtypes with is. If the input is an open server/platform value or needs real fallback/logging, keep else. Give each data-bearing subtype, including one a catch-all collapses, its own is branch. When its data affects the mapping or its safe handling, the finding or change must show the smart-cast member access, even if the public value stays constant:

   is Outcome.Failed -> safeFailureLabel(outcome.reason)