These tasks combine the service-boundary skills on new inputs. You will distinguish retryable failures from permanent errors, validate values that annotations can't enforce, keep test fixtures independent, and compare ordered results without hiding differences in a set. A short implementation still needs a precise contract.
Some boundaries concern information as well as types. A credential event records who triggered it while replacing the token with a fixed redaction marker. A page metric counts empty pages even when no new ids arrive. Return the exact requested result and let the tests expose the named mistake. Every dependency is local and deterministic, so a failure points to the program behavior rather than a real service, account, or network condition.