10,000 years is a trap. The label, often used for floods, heat, or other threshold events, describes an annual exceedance probability of one in 10,000 under stated assumptions; it does not place a velvet rope around the calendar. Bad luck can cluster. Two qualifying events may arrive almost back-to-back, even when each year's modeled chance remains tiny.
The public instinct is understandable, but wrong. A return period is an inverse probability, not a service guarantee: in a stationary process, the odds are reset with each interval, while a Poisson distribution can still produce unusually tight spacing. Think of coin flips. A rare result does not make the next flip wait politely, and the apparent shock grows when observers confuse a long-run average with a deadline. The arithmetic is merciless. If the underlying system changes through altered exposure, shifting climate conditions, measurement revisions, or a new threshold definition, the original estimate may itself be stale; independence and stationarity are assumptions to test, not laws of nature.
That distinction matters more than its dry mathematics suggests. Risk managers who hear a timetable may postpone defenses precisely when repeated loss is possible; those who hear a distribution must plan for coincidence, uncertainty, and the unnerving fact that the second wave can arrive before the first leaves the window. The calendar stays blank. Probability does not.
