"Most failures are preceded by decisions that felt reasonable at the time.
In an architecture review where two constraints collide, the facts arrive out of order and everyone wants the next step to feel obvious. The fork is real even if nobody names it.
Under pressure, people default to the simplest story. If you do not define the boundary, the simplest story will define it for you. If the exception path is informal, it becomes politics; write the signal that makes an exception legitimate and who can approve it. Name what you are refusing to optimize, so future changes do not quietly destroy the property you thought you were protecting.
Write the failure story forward while the decision is still reversible, then insist on one guardrail. That is what makes execution calmer. Write it so execution is obvious.