Will the portfolio outlast the plan?

Every simulated life produces a specific sequence of yearly returns. We run each one twice: once with the returns in their actual order (what the rest of the tool does), and once smoothed — every year replaced by that same path's geometric-mean return, so the total compound growth over the whole plan is identical and only the ordering is removed.

Each failing path is then sorted into one of two causes:

• Sequence failure — fails in the real order but survives when smoothed. The returns were adequate; the bad order (a poor early stretch hitting the portfolio while you withdraw) is what sank it.
• Magnitude failure — fails both ways. The returns were simply too low for the spending, so no reordering could have saved it.

The headline figure is sequence failures ÷ all failures, and the bar above shows the same split across every outcome (survives / magnitude / sequence). A few paths do the reverse — fail smoothed but survive in their real order because the ordering happened to help; those are set aside, not counted as failures.

For each assumption I nudge it to a plausible worse and better value — dollars by ±15%, ages by a few years, allocation by ±15 points, tax/fees/floor by a small amount — and re-run the simulation, holding everything else at your current plan. The bar spans the two resulting success rates (the amber side is worse, the blue side better), split at your current "now" line. The small numbers at each end are the input values that produced them. Bars are sorted by swing, so the top row is what your plan is most sensitive to. Rankings depend on those assumed ± ranges — treat it as a guide to relative leverage, not exact magnitudes.

Every rung fixes your target success probability and solves for one number. A tier anchored on spend asks "at this lifestyle, when is the earliest I could stop and still clear the target?"; one anchored on age asks "if I stop then, what's the most I could spend?" Both are found by bisecting the actual simulation — the same engine as everything else, narrowed to the crossing point rather than read off a grid, so the figure is exact to a year or a few hundred dollars rather than to the width of a heatmap cell.

Each tier carries one dot per scenario, joined by a line: the line's length is the finding — how much those assumptions are worth to that rung of the ladder.

By default the axis spans the answers, not the whole search. Retiring in your thirties and planning through 110 means seventy-eight years are searched to answer within about twelve of them, and an axis drawn to the search bounds would pile every rung against the left edge. The range actually searched is stated in words just under the boxes above, and the Chart axis box there can show that full range instead — a fitted axis magnifies small gaps between rungs, which is the right default but not the only honest view.

A hollow, dashed dot means there was no crossing to find, and the figure beside it says which way. "Already, at 55" — the target is met from the first age searched, so there is nothing to wait for. "Not by 94" — no age in range reaches it. "Over $210k" — the spending search hit its own ceiling, so raise the box above rather than reading it as an answer. "Not at any spend" — the plan misses the target even spending nothing, so the shortfall is somewhere other than lifestyle. Those markers sit at whichever end of the axis they ran off, which is usually past the fitted range — so read the figure beside them, not the position.

The pale band — switched on from Arrival range above — is where that rung's answer actually lands. The dot is today's forecast; the band re-solves the rung at the 10th, 50th and 90th percentile of the balance you might reach its date with, so it spans what the market can still do between now and then. Hover a rung for all three figures, or read the low draw columns in the data table, which give the 10th-percentile answer and what it costs against the dot — +3 yrs, or so many thousand a year less. That cost is the honest content of a failure probability: not ruin, but a mid-course correction you would see coming.

Two things follow from how the question is asked. On an age panel the band is one-sided — you cannot stop earlier than the date you asked about, so a good draw sits on the dot and only a bad one moves it right. And a rung with no crossing gets no band: "already, at 55" has nothing to arrive at, and "not by 94" never arrives. Solving the band roughly doubles the work of drawing the ladder, so it is off until you ask for it — switch Arrival range above to Show. The analysis prompt below carries these figures either way, since nothing there competes for space.

One caveat on the age search. It assumes retiring later doesn't make things worse, which is almost always true but not guaranteed: a retirement-relative income stream slides along with each age being tried, and guardrails reset off the balance on the day you stop. Where the relationship doubles back, the search finds a crossing rather than provably the earliest one. Common random numbers keep it stable — the same plan always gives the same answer — but the sensitivity sweep over retirement age is the place to check the shape if a rung looks wrong.

The dashed line is the balance that would hold this tier at the target rate if you were standing at that age today — solved by bisecting the simulation on the starting balance, the same engine as everything else. Behind it are the percentile bands for where your balance is actually projected to go. Above the line is on track; below it is not.

The reading watches the 10th percentile, not the median. A rung is defined as the point where the plan hits its target, so the median clears its own corridor very nearly by construction — it would report "above" for any rung the ladder could solve, which is no news. The lower band is where the content is: the age at which a bad run stops being something to ride out and starts being something to act on.

The vertical axis is fitted to these two series rather than anchored at zero. The whole reading is the distance between lines that sit close together, and a zero-based axis flattens it to nothing — but a balance chart that starts partway up is also how a small difference gets sold as a large one, so the axis label states where it begins and the data table carries the unscaled figures.

A rung with no crossing gets no corridor: there is no date to march toward. Solving one is a bisection per age on top of the whole ladder above, which is why it is off until you pick a tier.