Dev retires at 55 with £1,000,000 and plans to spend £40,000 a year in today's money. At a 5% real return the projection has him reaching 95 with £2,207,998 he never touched. Comfortable does not begin to describe it. The pound signs are a convention, by the way: the currency selector on the tool changes the symbol and nothing else, so the same figures in dollars or euros behave identically.
Now put a 35% fall into that plan, the size of 2008, and let the year it lands be the only thing that changes. Land it at 84 and Dev still finishes with £1,236,304. Land it at 64 and he finishes with £444,547. Land it the week he stops working and the money is gone at 90.
Nothing else moved. Not the pot, not the spending, not the size of the fall. That is sequence of returns risk, and it is the largest single reason plans that pass the average-return test fail in practice.
Sequence of returns risk is the order, not the average
Averages hide it because an average has no order. Take four years from the site's own historical series, an approximation drawn from the long record Robert Shiller publishes: the real US returns for 1973 and 1974 followed by the two rebound years, minus 22%, minus 35%, plus 29%, plus 19%. Add six flat years at 5% to make a decade. The average is 2.1% a year whichever way round you play it.
Leave a £1,000,000 pot alone for those ten years and it ends at £1,042,991 in either order. Compounding does not care whether the crash came first or last.
Now draw £40,000 a year from the same pot. Crash first, the way 1973 actually happened, and ten years later £517,748 is left. Crash last and it is £713,593. Same returns, same average, same withdrawals, and £195,845 between the two endings.
The withdrawals are doing it. Once prices have fallen, every £40,000 costs more units than it did the year before, and the units sold at the bottom are gone for good; they take no part in the recovery. Dev's £40,000 was 4% of £1,000,000 the day he retired. After the crash it is 6.15% of the £650,000 he actually has, and he needs to keep taking it for another forty years. A saver still working through the same fall is buying those cheap units rather than selling them, which is why the identical crash at 45 is closer to a gift than a threat.
Same crash, ten different years
The retirement crash test applies exactly that fall on the day the plan retires, then grows what is left at the assumed return with no rebound. Dev's plan runs out at 90 there. What the tool does not show is the same shock arriving later, so here it is, run through the same annual loop the tool uses.
A fall in his second year runs out at 92, and in his third year at 94. In year five, at 59, the plan survives, just, with £93,490 to spare at 95. From year eight, at 62, it is £314,306. Delay the crash to 64 and it is £444,547; to 69 and £719,610; to 74 and £935,129; to 79 and £1,103,994. By the time the fall lands at 84 the plan barely notices, finishing on £1,236,304.
The first few years of retirement carry the whole distance between broke and not. Which is the useful part of the idea: sequence risk is not a forty-year problem, it is a problem about the first handful of years, and that narrows what you need to defend against.
Why the same plan gets three verdicts
Dev's failure has less to do with the crash than with how long the money has to last afterwards. Give the identical £1,000,000, £40,000 and 35% fall to someone retiring at 60 and it runs out at 95, the last year the tool checks. Give it to a 65-year-old and it survives with £151,709. Nothing changed but the number of withdrawals still to come.
The largest first-year fall each of them can absorb says the same thing. At 55 the plan survives a 31.4% drop and not a 35% one. At 65 the line sits at 38.5%. Ten years of horizon buys seven points of crash tolerance at the same withdrawal rate. An early retiree is not running a riskier copy of a 65-year-old's plan; they are running a different plan with a 4% label on it, and this is one of the places where the 4% rule shows its thirty-year origins.
The cheap fix has a floor
The usual advice for sequence risk is flexibility: trim spending for a few years after a crash, or hold a cash buffer to spend from so nothing is sold at the bottom. Both are right, both cost far less than working longer, and neither rescues Dev.
Cut his spending by 10% for the three years after the crash and the money lasts to 91 instead of 90. Cut 15% for three years, 92. Cut 10% for a full decade and it runs out at 95. The smallest three-year cut that gets the plan past 95 is 33.4%, which means living on £26,647 through the worst of it, and that is a different retirement from the one he planned. A cash buffer behaves the same way. Three years of spending held in cash, £120,000 the crash cannot touch, gets him to 95 and not past it, and in the futures where no crash comes it costs £77,933 of final balance for the growth the cash never earned.
Now run the 60-year-old through the same fixes and every one of them works. A 10% cut for three years leaves £32,684 at 95; two years of cash, £96,026; three years of cash, £143,208. The fix that fits depends entirely on how far short the plan falls. A plan running out at 95 is what the cheap fixes were made for. One running out at 90 is not.
What does work for Dev is the thing the tool reports as extra years. Keep working for one more year, still saving £2,000 a month, and the pot the crash lands on is £1,074,000, which becomes £698,100 after the fall and survives to 95 with £116,786 in hand. A second year leaves £471,818. The other route is permanent: the most he can spend through a year-one crash and still last is £37,881 a year, a 5.3% cut for life, or 3.79% of the original pot. And the plan he could have built instead, £35,000 a year on the same million, takes the crash in its first year and still ends on £348,001.
None of that is an argument for working longer over flexibility. It is an argument for knowing which of the two your plan needs before you retire into it, because the answer depends on the size of the hole, and the hole is what the test tells you. None of the fixes above are controls in the tool, by the way; they are the tool's own annual arithmetic run with a spending schedule or a cash pot added, so you can reproduce them on paper.
What history says about the bounce
The crash test is deliberately harsher than the past. Real crashes were usually followed by sharp recoveries, and the tool strips the recovery out, so passing means the plan has structural margin rather than good luck. The difference shows against the historical backtest, which replays Dev's plan from every start year since 1900 with the recoveries left in. It succeeds in 116 of 124 windows, 93.5%, and the eight that fail begin in 1906, 1907, 1929, 1930, 1966, 1968, 1969 and 2000, each of them walked through in backtesting a retirement plan since 1900.
A 2008 start is not among them. The S&P 500 lost 36.55% that year with dividends included, on Damodaran's series, and a retiree who met it survives in the backtest because the decade that followed was exceptionally kind. Take that bounce away and the same fall breaks the plan at 90.
The failures history does record are mostly the slow kind. A retirement that began in 1966 or 2000 met a decade of poor real returns rather than one bad year, and lasted to 85 and 84 respectively before the money went. A single-shock test cannot see that shape at all. The backtest and the Monte Carlo simulator, which scores Dev's plan at 61%, are the tools for a slow decade, and what an 85% success rate actually means covers why a number like 61% reads worse than it is.
Running it on your own plan
The crash test has its own six sliders under "Your numbers", so type in the figures from your FIRE calculator plan rather than expecting it to read them; the how-to guide walks through the controls. Two numbers matter. The age the money runs out tells you how deep the hole is, and the extra years figure tells you what the most expensive fix costs. If the plan fails by a year or two, the cheap fixes cover it: agree a spending cut with yourself in advance, or hold two or three years in cash, and decide now rather than in the middle of a fall. If it fails by five or more, the honest choice is a lower draw or a later date, and finding that out at 55 with a job is worth a great deal more than finding it out at 68 without one.
And if it passes, remember what passed. A plan that can take a 35% fall with no rebound in its first year has real slack in it. Whether it also has slack for a slow decade is the next tool's question, not this one's.