Why 90% chance of success is the right goal and 100% is a trap

"I want a 100% chance of success" sounds like prudence. On a rigid withdrawal rule it is often a request to spend almost nothing, or a request the engine cannot print at all. A 90% chance of success is a point on a curve. 100% is the end of the curve, or past it. This page is that curve on one frozen retiree.

Specimen: $1,000,000, age 65, already retired, 80/15/5, 30 years, seed 20260622, 1,000 shared paths. Treatment: the rigid initial rate, from 2.0% to 6.0% in 0.5% steps, band 0, adjust 0. Only the rate moves. Success is the share of paths that never hit zero. That is a last-or-not meter, not a happiness meter. Flexible rules that can cut spending to zero can print 100% by construction. This roster cannot. That is why it is the right roster for this question.

Why a 90% chance of success is a point, not a personality

Interpolating the frozen grid, 90% success sits near a 3.95% rigid withdrawal. At 4.0% the print is 89.6%. At 3.5% it is 94.0%. Nobody's temperament changed between those two rows. The spending did. A 90% chance of success is the spending that leaves about one in ten modelled lives hitting zero under this rule and this generator. It is not a moral score.

First-year spend at 3.95% of $1,000,000 is about $39,500. First-year spend at 2% is $20,000. The gap is the price of walking the curve toward the left. Whether that price is worth paying is not a question this engine answers. The engine answers how often the pot lasts at each step.

People treat 90% as the grown-up number because some papers and some software default there. Defaults are not evidence. On this freeze, 90% is simply a height on the y-axis that a rate on the x-axis can be read against. You could have asked for 85% or 95%. The curve is the result. The 90% line is a reference, like the 1-point standard-error line on the sample-size chart.

Why 100% is a trap on this grid

The pre-registered claim was: find the rate that prints 100%, if any. The freeze printed none. The 2% row, the kindest rate on the card, still ruins in 0.6% of worlds (99.4% success). There is no 100% to chase without leaving the grid, and leaving the grid would have been search. We did not add 1% after seeing 2% fail. That is the point of the card.

So the trap is not "100% is cowardly." The trap is thinking 100% is a nearby point on the same curve. It is not nearby. On this specimen it is not on the curve. A household that hears "we can get you to 100%" is being offered either a different rule (cut spending when markets fall), a different generator (gentler crashes), a different horizon, or a spend so small it is a different life. Naming which of those four is the offer is the whole job. A slider that says 100% without naming the mechanism is the trap.

Guardrails and other flexible rules can print 100% survival while spending collapses. That 100% is tautological if survival is the primary metric. We did not use those rules as the primary here. Rigid rates cannot hide in a spending cut. When they print less than 100%, the miss is ruin, not a quieter lifestyle. Read 100% on a flexible rule in the same paragraph as the spending it took, or do not read it.

The rest of the curve

At 5% success is 76.8%. At 6% it is 63.0%. Median terminals fall as well: $1,887,349 at 2%, $1,083,603 at 4%, $307,594 at 6%. The left tail hits zero from 3.5% up (p5 is $0). A 90% chance of success is not the rate that maximises median wealth. It is a survival cut. The companion article on probability of success versus projected balance is the same freeze, a different meter.

Sampling error at n = 1,000 is about one point on an 85% rate. The 2% row at 99.4% is not "almost 100% within error." Six failures in a thousand is a small mass. It is not zero. Calling it 100% after rounding would have been the quiet bug this page exists to refuse.

What would falsify this, and what we actually stressed

Q2 was run. We shaved stock mu by 0.02 and re-ran 4% and the interpolated 90% rate. The 3.95% rate fell to 78.2%. 4% fell to 77.5%. 100% remained empty. The spending gap between a 90% print and a 100% print did not close. The ranking of "100% is not a nearby point" survived the shave.

Q3 was not run. This is not a historical claim. A 1966 start on a 30-year rigid 4% is a different paper. Q4 was not a 0-versus-5-bps stress. The vector engine used 5 bps. Tax is off. Rigid rates have similar turnover, so a cost ranking is not the issue. The issue is the empty 100% cell.

What would falsify: a 100% success rate at a spending level within half a point of the 90% rate, on this grid, or the mu-shave making 90% and 100% coincide. Neither happened.

How a 90% chance of success gets used badly

Software that paints 90% green and 89% yellow is inventing a cliff. On this freeze, 3.5% prints 94.0% and 4.0% prints 89.6%. Both are near 90%. A household that changes its life because a slider ticked from 90.1 to 89.9 is reacting to paint, not to a different spending program. The curve is smooth. The trap is treating a round percentage as a gate.

The other abuse is the reverse: refusing to spend because 100% was not printed. 99.4% at 2% is not a reason to live on $20,000 from a million unless that is the life you want. It is a reason to admit that 100% did not print. Those are different admissions. A 100% chance of success, on this rigid grid, is not a nearby conservative sibling of 90%. It is off the page. Spending collapse would be how a flexible rule fakes it. We did not let the rule fake it.

First-year dollars at 3.5% are $35,000. At 4% they are $40,000. At 4.5% they are $45,000. Five thousand dollars a year is a real lifestyle gap. It is also only half a point on the x-axis. Read the curve in dollars as well as in percents, or you will negotiate with a round number instead of with a paycheck.

See the curve on a calculator

The safe withdrawal rate calculator and the 4% rule calculator are the matching tools. They will not print 100% just because the input looks tidy. If a tool does, ask which of the four mechanisms above it used.

What this page refuses to say

It refuses to say that 90% is the right risk tolerance for a person. It refuses to say that 85% is irresponsible. It refuses to say that a couple should spend 3.95% of a million because a freeze interpolated there. It refuses to round 99.4% to 100%. It refuses to add a 1% row after seeing 2% miss. It refuses to treat a green 90 in software as a different species from a yellow 89. Those refusals are the control mechanism working. A page that ended with "so target 90%" would have turned a curve into advice.

It also refuses to hide the dollar gap. Living on $20,000 from a million is a different retirement than living on $40,000. The curve will not pick for you. It will only say how often each paycheck lasted in this model. If that is not enough to decide, you need a preference, not another thousand paths.

Replication spec

Roster: rigid 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5.0, 5.5, 6.0 percent, each with guardBand 0 and adjustPct 0, plus the buy-and-hold anchor, on B4_RETIREE, seed 20260622, 1,000 shared paths. Success is 1 minus ruinProbability from the engine summary. The 90% rate is linear interpolation between the 3.5% row (94.0%) and the 4.0% row (89.6%), which is why the card reports 3.95% rather than a grid point. The 100% rate is the highest grid point with success of 1. There is not one. The 2% row is 99.4%, which is 6 failures in 1,000, not zero.

Q2 re-ran 4.0% and 3.95% after subtracting 0.02 from stock mu in every regime. Results: 4% at 77.5%, the 90% rate at 78.2%. Median leftover at 4% falls from $1,083,603 to $477,945. The empty 100% cell stays empty. That is the stress the card paid for.

p95 at 2% is $10,378,565; at 6% it is $6,170,740. Even the kindest sequences leave less when you spend more, but they still leave millions. Do not let p95 talk you into 6% on a rigid rule. The success rate at 6% is 63.0%. That is the column that names the trap if you were hoping 100% was one click to the left of 90%.

CSV path: /data/studies/90-percent-success-not-100.csv. If a later agent regenerates the CSV, the interpolation will move. The article must move with it. It must not keep 3.95% as folklore.

Notes. B4_RETIREE, seed 20260622, 1,000 paths, regime_switching, shared paths. Roster: rigid rates 2.0%–6.0% step 0.5%, guardBand 0, adjustPct 0. Success = never hit zero. rate90 interpolated between 3.5% and 4.0%. rate100: none on the grid. Q2: stock mu −0.02 on 4% and on the interpolated 90% rate. Q3 not run. Q4: 5 bps vector default, tax off. CSV: /data/studies/90-percent-success-not-100.csv. Card dated 2026-08-28. This article is educational analysis, not investment advice, and does not recommend any security or strategy.