# Retirement Optimizer: Detailed Change Log
Full write-ups of what changed in each release, newest first.
One entry per BRANCH, written against `main`: what an end user would experience differently, ordered
by how much it affects them. Work in progress keeps editing the entry at the top - refresh its
version stamp if you like, but never add a second entry for the next commit. Leave out how things
used to work and anything about the internals; the commit messages carry that. See the rule in
[CLAUDE.md](CLAUDE.md).
The Documentation tab inside [retirement_optimizer.html](retirement_optimizer.html) carries a short
summary of the most recent releases and links here for the detail. Entries marked **behavior
change** alter the numbers an existing plan produces, so a saved scenario or a shared link can give a
different answer than it did before that release.
For what the tool does and how to use it, see [README.md](README.md).
---
## 11.173b
**Money you convert now grows for the rest of the year it was converted.**
*Behavior change: ending Roth rises in every converting plan.*
A Roth conversion was credited to the account only after the year's growth had already been applied, so the converted amount
earned nothing in the year you converted it. Surplus banked to Cash or Brokerage instead was credited before growth and did
earn it, so the same dollar grew or did not purely according to where it landed.
It now grows from the point of conversion for the remainder of the year, which is the same treatment every other destination
already had. The amount is not given a full year of growth either - it already grew inside the IRA up to the moment it was
withdrawn, and it grows in the Roth for what remains.
Conversions were being undervalued, so everything that weighs one moves: **Break Even arrives sooner**, the **Marginal Heirs
Tax Rate** needed to justify a conversion **falls**, and **Optimize Conversions** suggests **larger** amounts. On one test plan
the break-even heirs rate fell from 65% to 55% and the suggested conversion rose from $100,000 to $150,000 a year.
Spending, taxes and withdrawals are unchanged. A saved plan or shared link will report a different ending Roth and End Wealth
than it did before this release.
**End Wealth now prices the leftover IRA at your late-life tax rate, not at one year's.**
*Behavior change: End Wealth moves for every plan.*
End Wealth has always been an after-tax figure. It subtracts the tax that would be owed on whatever is left in
the IRA, and the capital-gains tax on brokerage growth above its basis. What changed is the rate used for the
IRA half.
It used to be the marginal rate of the plan's very last year. One year is a poor guide to a lifetime: a year
with a large brokerage harvest and a year with a large Roth conversion land in very different brackets. On one
test plan, two otherwise similar plans were priced 7.3 percentage points apart for that reason alone, which was
worth $277,192 on the IRA being valued, and it was enough to change which plan looked better.
It is now the average rate across the plan's final years **that share the same filing status**. A plan that
ends in widowhood is priced at the widow's rate rather than at a blend that includes the married years before
the first death, and the jump between those two is large. A plan with a long widowhood averages more years and
is estimated better, which is where the number matters most.
If you have set a **Marginal Heirs Tax Rate**, nothing here applies to you: that rate is used as given, exactly
as before.
Spending, taxes and withdrawals do not move by a cent. A saved plan or a shared link will report a different
End Wealth than it did before this release, and where *Optimize for* ranks plans by wealth, the order can
change with it.
**The IRA Goal greys out for the strategies that ignore it.**
Three strategies never read the *IRA Goal* at all: Proportional Withdraw, Ordered and
Guyton-Klinger. No value you type there changes their result. The field now greys out and says
"Not used by this strategy" when one of those is selected, and the number you had is kept rather
than cleared, so switching away and back does not lose it.
The three that do use it use it differently, and the tooltip now says which is which. *Reduce IRA
in N Years* treats it as a drawdown target and it moves that plan a great deal. *Fill Fed/IRMAA
Bracket* and *IRA Draw* treat it as a floor the withdrawal will not cross, so it only takes effect
once the balance is already near it, and a goal well below your balance will not change anything.
**The wealth total is now named for what it is.**
The total on the balances chart and in the Annual Details table was labeled `TotalWealth`, which
sounds like a plain sum of your accounts. It is not one. That figure already subtracts the tax that
would be owed on the IRA if it were drawn, and the capital-gains tax on brokerage growth above its
basis; Roth, cash and returned basis are counted at face value. It now reads **`TotalNetWealth`** in
both places.
No number changes, and no plan produces a different answer. If you export the Annual Details or read
the column by name, the name is what moved.
---
## 11.171f
### The required distribution on an inherited IRA is calculated again
For the first year after a spouse's death, no RMD was taken on the IRA the survivor took over. This
broke in 11.168c and is fixed, so a plan with a death inside it distributes and taxes that year
correctly. Spending does not move.
### Earliest Break Even breaks ties on the ending Roth balance
Under *Optimize for* -> *Earliest Break Even*, two plans that break even in the same year are now
ordered by the ending Roth balance, and by End Wealth only when that ties as well, so the row at the
top of that goal can differ and the "Earliest Break Even" marker in the Best table moves with it.
That goal's table also carries **Final Roth**, the balance that separated the two plans, and
**Final IRA**, the pre-tax balance the conversions were drawn from. Every other goal's order is
untouched.
---
## 11.1703
### Cash Reserve now defaults to 0
**Behavior change for new sessions and for shared links that did not carry a Cash Reserve.** The
field used to default to Off, which left every dollar of surplus in Cash at the cash yield. It now
defaults to 0: no cash buffer, all surplus reinvested in your Brokerage. Measured across withdrawal
families and account mixes on the corrected engine, 0 is never worse than Off and usually better,
survival is unchanged, and every dollar of buffer costs a little. Type Off to keep the original
all-cash routing; a positive amount still keeps that much in Cash and reinvests only the overflow.
Saved scenarios keep whatever they carry.
### Annual Details shows every way money enters Brokerage
**BrokerageG** is the market return on the balance plus, when Dividend Reinvestment is on, the
dividends reinvested into it; its tooltip now says exactly that. Two new columns under
**Brokerage Δ** break the rest out: **DRIP**, the reinvested dividends, which are already inside
BrokerageG, and **SurplusBrok**, the surplus the Cash Reserve rule routed into Brokerage, which
BrokerageG never included. **SumBrokIn** is the running total of the two: every dollar that was put
into Brokerage rather than earned there. Last year's Brokerage, minus Brokerage-, plus BrokerageG,
plus SurplusBrok is this year's balance, and a test now holds the tool to that identity.
### The Balances chart has a Scale control
**Linear, log10 or log2**, beside the Both / Mine / Spouse buttons. A log scale spreads out the
small balances that sit in the looks-like-zero band under a large one. log10 puts gridlines at 1k,
10k, 100k and so on; log2 at every doubling. On a log axis a zero balance has no point to draw and
shows as a gap in that line rather than a false floor. The page opens on linear.
### Annual Details shows your Cash Reserve
A **CashReserve** column under **Balances** and **Cash Δ** is the part of each year's Cash balance
that is the reserve: the smaller of the target, grown with inflation, and the Cash actually held.
It shows the buffer being held, topped up, or broken into, and hides itself when there is no
reserve (Cash Reserve Off or 0).
### Proportional and Guyton-Klinger plans were withdrawing, and converting, money they did not need
**Proportional and Guyton-Klinger strategies contained an error that has been corrected. The error
was small, but meaningful.** When a year's spending was funded partly from Cash or Roth, the
tool counted that money as spent but not as available, so it drew the same amount a second time
from the IRA and Brokerage and returned the excess at year end. With *Max Conversion* on, the
excess was converted to Roth instead of returned: a Proportional +0% plan, which asks for no
conversions at all, was converting several thousand dollars a year for as long as its Cash lasted.
**Spending is unchanged to the dollar. Ending wealth moves**, in most plans upward and in some
down, by an amount that depends on how much of each year's spending came from Cash or Roth. Plans
whose first draw comes only from the IRA - Fill Fed Bracket, IRMAA Tier, ACA Cliff, IRA Draw % and
Reduce IRA - and Ordered plans are unchanged. Re-run the Optimizer before treating an old best row
as settled.
---
## 11.16fb
### Annual Details now accounts for every dollar that leaves Cash
When a Roth conversion's tax is paid from Cash, that money left the account with no column to find
it in, so the Cash balance could fall with nothing on screen to account for it. **Cash WD** does not
show it, because that column is the draw that funded spending.
Two new columns fix it. **ttlCashWD**, beside **Cash WD** in the Withdrawals band and in **Cash Δ**,
is every dollar that left Cash in the year: last year's Cash minus ttlCashWD, plus interest and
growth, is this year's Cash. **ConvTaxCash**, in **Cash Δ** and **Opp. Cost**, breaks out the
conversion-tax part. Both stay at zero unless *Fund conversion taxes from cash* is on.
### An account with a zero balance keeps its column
In Annual Details a balance column disappeared when that account sat at zero for the whole plan, so
a plan holding no Cash showed no Cash column under **Balances**. A zero balance is a fact about the
plan, and a missing column reads as though the tool does not track the account at all.
**Cash, Brokerage, Cost Basis, Total IRA, Roth, Total Wealth and Spendable now keep their columns
whatever the balance.** Columns carrying a flow or a rate, such as a tax that is never paid, still
hide themselves when every year is zero, so the table does not fill up with empty numbers.
The per-person splits stay as they were. An all-zero **Roth 2** means there is no second person
rather than an empty account, so it still hides.
### Internal structure changes
This release also carries internal changes that do not affect any number your plan produces. They
support work in progress, and the commit history has the detail.
---
## 11.16d7
### A chosen limit is now filled even after Social Security starts
**Behavior change: saved plans and shared links will not reproduce their earlier numbers.**
Once Social Security begins, a plan on a **Fill Fed Bracket** or **IRMAA Tier** limit stopped 15% of
the yearly benefit short of the limit it was told to fill, and stayed short every year after. On one
$2.8M plan at the 22% limit that was $168,500 of room left unused across 17 years. Only the taxable
part of a benefit counts toward these limits, at most 85% of it, and the untaxed remainder was being
charged against the limit as though it counted.
These plans now withdraw and convert more in those years. **Which way that lands on End Wealth still
depends on the limit**, the same as it did in 11.16b0, so re-run the Optimizer before treating a new
figure as settled.
**Optimizer row ranking is improved**, especially when *Optimize for* resulted in a small number of
matching strategies, as is common in *Roth Conversion Effectiveness*. Those unmatched were treated as
ties and the ranks assigned were meaningless. That meaninglessness led to rank instability. Now all
non-matching rows are sorted by secondary criteria.
**Separately, the "Optimize for" goal now travels with your plan.** A shared link or a saved
scenario kept every input but not the goal the table was ranked by, so it reopened on Tax Flexibility
and showed a different best plan than the one you sent. Links made before this release are unchanged.
**ACA Cliff plans are unchanged.** An ACA cap counts the whole benefit, taxable or not, so for those
plans the earlier treatment was already right.
---
## 11.16d0
### The column a note points at is now one click away
Where a note names a column in Annual Details, that name is now a link. Clicking it opens Annual
Details, turns on the category the column belongs to, scrolls to the column and highlights it for a
moment. **The columns you already had stay exactly as you left them**, and nothing about your plan is
recalculated or reloaded. If the column is all zeros, Show Zero is turned on so it is still there to
read. The same links are in the How To, on the columns and tabs it sends you to.
**ACA Cliff plans get a new acaBreach column**, marked Yes in each year the cap could not be held
alongside the Spend Goal. It is blank on plans not using an ACA cap, and hidden when it is blank
throughout. The note under an ACA cap named this column before it existed.
### The Documentation tab no longer reports a test failure that was not one
The Documentation tab read **Documentation ❌ tests failed** on every load, over one failed test
about the Limit menu's 24% entry. **No calculation was affected, and nothing in the menu was wrong.**
The tab reads 🟢 again on load, as it already did with `?runtests` on the address.
---
## 11.16b0
### Fill Fed Bracket now fills the bracket
**Behavior change. Every plan using a Fill Fed Bracket limit will withdraw and convert more, and
saved plans and shared links will not reproduce their earlier numbers.** Picking "22%" stopped short
of the top of the 22% bracket by the whole of that year's deduction - about $32,000 in 2026 for a
married couple, and more each year as the deduction grows. A tax bracket is a limit on income after
the deduction; the plan was measuring against income before it. The two are now on one footing, so a
chosen bracket is filled to its top.
**This is not free money, and which way it goes depends on the bracket you picked.** Measured across
71 plans that fund the same spending either way: terminal net worth rises in 18 and falls in 49. A
12% limit typically gains about $158,000; a 22% limit typically loses about $200,000, and a 24% limit
is close to flat. Filling a cheap bracket further is usually worth doing; filling an expensive one
usually is not, because money leaves the IRA earlier than it needed to and stops compounding there.
**So a plan on a 22% or 24% limit will usually show a lower End Wealth than this tool gave you for
the same inputs before** - about $200,000 lower at 22%, in the middle of the range measured. Nothing
about your plan got worse; the tool was under-filling the bracket you asked for, and the shortfall it
left behind happened to be worth keeping. If End Wealth is what made you choose that limit, the
comparison it was based on has moved: worth re-running the Optimizer, or trying a lower limit
alongside it, before treating the new figure as settled.
Limits that are not federal brackets - IRMAA tiers and the ACA cap - are unchanged, because those
are already measured on the right income.
### A limit broken by required distributions no longer tells you to spend less
When a plan goes over the limit you picked, the note under it now says **which** of two causes did it,
because they take opposite advice. If your spending forced the plan past the ceiling, lowering the
Spend Goal helps. If required distributions, Social Security and a pension already exceed it on their
own, lowering the Spend Goal changes nothing at all - the plan is taking no more than it must - and
the note now says so, names the required distribution doing it, and points at what does move it:
converting more before required distributions begin, a QCD, or a higher limit. A plan with both gets
both counts.
### The Limit menu says where each limit sits on the other ladder
The menu mixes two kinds of ceiling and used to print both as a bare dollar amount, which invited a
comparison that does not hold: a federal bracket is a limit on income **after** your deduction, while
an IRMAA tier is a limit on income **before** it. Each entry now names where it falls on the other
one, so `22% Fed - $211k (IRMAA Tier 1)` and `IRMAA Tier 1 - $274k (24% Fed)`. Amounts are shortened
to make room.
Underneath, a sentence describes the limit you picked, including the thing the list cannot show: an
IRMAA tier **spans** a bracket boundary, so filling Tier 1 starts in the 22% bracket and finishes in
the 24% one. **Show me** opens a picture of both ladders on one income axis, with your limit and your
plan's own first-year income marked on it.
**The amounts in that menu were each about 2.8% too high** and are now correct: they were being aged
one extra year past tax tables that were already current. The top of the 22% bracket reads $211,400
again, which is what the plan was using all along - only the menu was wrong.
### The plan now tells you when your limit could not be kept
Picking a limit did not stop the plan from going past it. When your Spend Goal could not be funded
underneath the limit, the plan paid for the spending anyway and drew above the ceiling, and nothing
on screen said so - only the BracketOverage column in Annual Details recorded it, so the headline
numbers described a plan running under a limit it had broken. A note now appears under the limit you
picked, naming how many years of the plan could not stay inside it and by how much at worst. It
counts only years the spending forced it; a conversion you chose to make going over is reported
separately, in the note under Extra Annual Roth Conversion.
That note under Extra Annual Roth Conversion also named the wrong ceiling. It said "the federal
bracket ceiling" whatever you had picked, including on plans set to an IRMAA tier or an ACA cap. It
now names the one you chose.
### A saved plan naming a strategy this version does not have now loads as the default
A saved plan or a shared link can name a withdrawal strategy that is no longer in the Strategy menu.
Until now the menu came up with nothing selected and the plan computed as all zeros, with nothing on
screen to say why. Any strategy the menu does not carry now loads as Proportional Withdraw +% at its
default setting, so what you get is a working plan you can change rather than a blank one. A plan
naming a strategy that IS in the menu is unaffected.
### Smaller
If everyone in your plan has already passed 65, the ACA income-cap options are still greyed out but
the note beside them is gone. It ended by telling you to lower your Retirement Start Age to model
pre-Medicare years, and that cannot work once the age is behind you: the plan always starts this year
at the earliest, so no start age reaches a year before Medicare. Where the start age is what puts you
past 65, the note still appears, because there lowering it does help.
---
## 11.16a9
### The Stress Test could report on the wrong plan, and Roth conversions now count toward IRMAA
**The Stress Test's first result could be computed against the plan you had before the one you
loaded.** Opening a saved plan or a shared link starts the pass once on load; if the plan finished
loading while that pass was still running, the refresh it asked for was dropped and never retried, so
the headline settled on the earlier plan's horizon. Measured on one shared link: it read "runs out of
money in 8 of the 36 worst historical periods" where the plan on screen survives all 40. **If you
have read a Stress Test result on a freshly loaded plan, re-check it.** The chart and survival table
below it now also raise the "Out of date" banner in that situation instead of appearing current.
**Behavior change. Plans using Extra Annual Roth Conversion will show higher IRMAA surcharges, and
saved plans and shared links will not reproduce their earlier numbers.** The conversion was always
taxed correctly, but never counted toward the income Medicare reads two years later to set your
surcharge, so a plan could convert year after year and never be billed for it. This reaches every
strategy, not only those targeting a bracket or tier. Nothing changes with no extra conversion set.
Two places now warn when a conversion goes past a limit you set: a note under Extra Annual Roth
Conversion naming how many years it puts you over, and an upward arrow on any Optimizer row whose
conversion lands above its own ceiling. Neither blocks - a ceiling paces ordinary withdrawals, while
a conversion moves money from IRA to Roth rather than out of the household.
**Also a behavior change:** the ACA income-cap notice no longer appears for non-ACA choices, and no
longer works its year out from your birth year plus the Retirement Start Age you typed, which could
name a year already past. That year also decides whether ACA rows appear in the Optimizer, so if you
are past the start age you typed, ACA rows may now be correctly withheld.
The "Assets at Retirement Age" heading now names the year it means - "Assets at Retirement Age
(2035)". **This tool has no accumulation phase**: it never grows your balances between today and a
later retirement year, so if your Retirement Start Age is still ahead of you, forecasting the
balances to that year is yours to do. Nothing about the calculation changed; the heading and its
documentation now say which year the numbers belong to.
Smaller: the Cash Reserve note offered "-1", a value the field will not take - it now says Off.
Hovering over an income bar on Income & Expenses shows the actual income with the approximate tax
attributed to it, not the scaled bar height. The Market Return chart names the historical year a
replayed path came from for everyone.
---
## 11.1691
### Current $ view now converts running totals correctly
**Display fix. No plan produces different totals; what changes is how the running-total columns in
Annual Details read when the Future $/Current $ switch is set to Current $.**
The running-total columns (SumTaxes, SumAdvisorFees, and Spendable, renamed SumSpendable) could
fall from one year to the next in Current $ view, which a lifetime total should never do. Each is
now built the right way for that view: every year's amount is converted to today's purchasing power
first, then added, so the total only grows. Future $ view and every per-year column are unchanged.
Hover over any of the three column headers for what each total includes.
The All RMDs figure and its QCD note, on the summary bar and in the Optimizer table, now follow the
Future $/Current $ switch the same way. They always showed future dollars, even in Current $ view,
sitting beside an All Taxes figure that converted; the two could not honestly be compared. The
Optimizer's All RMDs column also sorts on the figure it displays.
Three more readouts join them: the Advisor Fees tile's per-year average now averages the figure the
tile shows, the Optimizer's Conv Tax column converts (and sorts on what it shows), and the dollar
amounts in the Break Even explanation are stated in the dollars the switch selects. The Break Even
suggestion itself, which year to stop converting or whether to convert at all, never changes with
the switch; only the dollars that describe it do.
The Monte Carlo tab is the largest change. The survival table's Final Balance, Total Taxes and
Total Spendable columns, and the "Median ending balance" in the plan headline, each showed one
fixed basis regardless of the switch; taxes were future dollars while Spendable beside them was
today's dollars, so the two could not honestly be read together. All of them now follow the switch,
and the table sorts on the figures it displays. The percentile bands, the drawn worst-to-best
paths, and the Stress Test chart and table also follow the switch, and their Current $ figures are
now exact: every simulated path is deflated by the inflation that path actually lived through,
rather than every path sharing one average rate. High-inflation futures now look as poor in today's
dollars as they are; with the old average-rate conversion the difference reached several percent.
Hover over a line to read the value as drawn.
The Monte Carlo path count now defaults to 400 per strategy (was 500), so a Compare run finishes
sooner out of the box. The Paths box still goes up to 5,000 when you want a finer sample.
---
## 11.168e
### An advisor fee you can model, and required distributions figured the way the tax rules define them
**Behavior change. Any plan that reaches the age for required distributions will show different
numbers, and a saved scenario or a shared link can give a different answer than it did before.**
**1. Advisor and fund fees.** A new field in Assets takes the fee as a single number, plus a choice
of which accounts are billed and which pay, including billing every account except Cash but taking
the money from the larger IRA. **A small number is read as a percentage and a large one
as a flat dollar amount that grows with your CPI**, and the line under the field says which way it
read what you typed; add a % or a $ if you want to say so outright, and shorthand like 20k works.
**It starts at None, so no fee is charged until you pick accounts**, and switching back to None is
how you compare a plan with and without the fee without retyping the amount. It is charged at
the start of each year against the previous December 31 balances, the way advisors bill. **Money
taken from an IRA to pay the fee is not a taxable distribution**, so it never reaches your income or
your tax. Cash is never billed and never pays. A tile shows the lifetime total, and appears only
once you set a fee. At 1% of $2M that is roughly $20,000 in the first year, compounding for the
whole plan, which is larger than several of the differences this tool is used to weigh.
**2. Required minimum distributions** are now figured from your December 31 balance divided by the
IRS life expectancy factor. The tool was adding part of the current year's growth to that balance
first, which made every distribution larger than the amount actually required, and made it move with
the month a withdrawal happened to leave. Two otherwise identical plans could show different
distributions purely because one of them had done a Roth conversion the year before. Expect lower
distributions, lower taxable income, and usually a lower lifetime tax bill and a larger ending
balance.
**One limit worth knowing.** For a plan's first year the tool uses the IRA balance exactly as you
typed it. If that is a mid-year figure rather than the previous December 31, the first year's
distribution is an estimate. Every later year is exact.
---
## 11.1671
### Change the plan while replaying a path, and a COLA that is never a pay cut
**Behavior change. Your Monte Carlo and Stress Test results may now differ from before, because the
model behind them is more accurate.** Any path containing a deflationary year is affected.
The main projection uses the fixed growth and inflation figures you supply, so the first two changes
below show up only in the Monte Carlo, the Stress Test and a Replay, where every path carries its
own returns and inflation. That is to say, those runs now account for real-world behavior of Social
Security, tax brackets, pensions and the rest.
**1. While reviewing a Monte Carlo path you can change the inputs and watch the result.** Replaying
no longer ends when you touch the sidebar: change the spend, the withdrawal order, anything, and the
same sequence runs again against the new plan, so you can test a change against the ruin in front of
you and find out whether it survives. Starting a replay copies the run's own strategy and conversion
settings into the sidebar, which is what is being replayed, and *Exit replay* keeps every edit you
made. While a
replay is on screen an edit redraws the path and leaves the Monte Carlo and Optimizer sweeps alone.
The forward and back arrows run as one loop through every path the run kept, the ten captured Monte
Carlo paths and all the stress scenarios, about 46 in total, so neither arrow ever dead-ends. Those
kept paths are also drawn on the survival chart, five worst in red and the percentile samples in
gray, and clicking one replays it. Picking one out there can be fiddly in the early years where
several ruined paths run together, so the ▶️ Replay list beside the headline names them instead. Its
tooltip now describes the single line under the pointer rather than listing everything at that year.
**2. Social Security and a pension no longer shrink during deflation.** A cost-of-living raise can
be nothing in a deflationary year, but it is never a pay cut. The two recover differently, because they are
different instruments. Social Security measures each increase from the last one it actually paid,
per 42 U.S.C. §415(i), so after prices recover the benefit picks up where it left off rather than
being paid twice for the same recovery. A capped pension keeps what it was paid and starts again
from there, which is what plan documents say.
**3. The Market Return chart gains a real-return line**, the market change after inflation. Unlike
the two above, this one shows on every projection rather than only in a replay, and it sits flat on
a plan with fixed assumptions. The chart's key no longer shows a red swatch for bars that are mostly
green. In nerdknob mode, replaying a Historical or Stress path adds the year each one came from to
that chart's tooltip heading.
Also on the Monte Carlo tab: Input Distributions shows a fold marker, so it reads as something you
can open, and the Stress Test heading points at its own fold rather than at the tab you are already
on.
---
## 11.1662
### Taxes now follow the inflation your plan actually gets
**Behavior change. More accurate tax modeling, so any Monte Carlo or Stress Test result you saw
before this release will have changed.** A pension with a COLA also moves in the main projection,
because it now follows CPI rather than general inflation.
Each Monte Carlo and Stress Test path has always given your spending its own inflation, but the tax
code stayed frozen at the CPI you typed. On a path whose inflation ran ABOVE that figure, your
income climbed while the brackets stood still, so the plan paid tax at rates it would not really
have faced. On a path BELOW it the error ran the other way: brackets kept climbing past prices, and
the plan looked less taxed than it would have been.
Brackets, IRMAA tiers, the ACA cap, the QCD limit and Social Security COLA now follow each path,
indexed one year behind the inflation that sets them, the way the IRS and Social Security actually
do it. Both errors were real, and neither was small. What changes most are the high-inflation paths,
because history hands them inflation far above any CPI you would type while rarely handing them
inflation far below it, and those are the paths where a plan runs out of money. Plans that showed
money running out in a high-inflation path may now survive it.
CPI and Inflation stay separate, and the gap between them now does real work: the tax code is
indexed at each path's inflation less that gap. A new line under the two fields names it, so you
can see which way it runs and by how much.
**Pension COLA is now a choice of five, not a checkbox:** no increase, a 1%, 2% or 3% cap, or full
COLA. A capped pension pays the lesser of its cap and that year's CPI, so it falls permanently
behind in a high-inflation stretch instead of catching up later. That is how federal FERS and most
state and municipal plans actually pay, and a checkbox could only call them uncapped or nothing. A
pension also follows CPI now rather than general inflation, as Social Security already did. Plans
saved with the box ticked load as Full COLA; unticked loads as No increase.
Also fixed: choosing **Fill Bracket** at the top federal bracket produced no plan at all, with every
figure reading `$NaN`. That bracket has no upper limit, so there was nothing to fill up to. The
ceiling menu now ends at the highest bracket that has a top; the top federal bracket and the top
IRMAA tier stay listed, greyed, showing the income where they begin. And a saved plan using Fill
Bracket came back with its ceiling reset to "Below IRMAA" instead of the bracket you chose, so it
ran a different strategy than the one you saved. Reload any such plan to get your own ceiling back.
---
## 11.1657
### Walk one Monte Carlo path through your plan
The Monte Carlo tab reports survival rates and percentile bands, but a band is a statistic, not a
sequence any plan actually lived. Every run now captures ten of its paths - the five worst, plus one
at each of the 5th, 25th, 50th, 75th and 95th percentiles of the outcome ranking - and the Stress
Test keeps every one of its historical scenarios.
A ▶️ Replay picker beside the survival headline offers every captured path, worst to best, and
each Stress Test row has its own ▶️. The chosen sequence plays through Charts and Annual Details,
and the balance chart overlays a dashed "Plan (steady assumptions)" line - the same plan run at
your flat growth and inflation assumptions - so the point where the path pulls away from the
expected trajectory is visible rather than remembered. A new Market view in the chart row below
draws each year's market return as bars with inflation as a line, plus what day-one $10,000 still
buys on its own dollar scale; replay switches to it automatically, and it is available for
ordinary projections too. The Current $ toggle deflates the
replayed numbers by the inflation that path actually lived, and the dashed line by its own steady
assumptions.
One caveat to read replays with: taxes and Social Security are not yet adjusted for variable
inflation. Tax brackets, IRMAA tiers, the ACA income cap, and the Social Security COLA all index
at the fixed CPI input even while a path's spending inflates at its own drawn rates, so a
high-inflation path overstates tax and understates Social Security income. The plan is your own, every setting as the
sidebar has it; only that path's market returns and inflation are applied, year by year. Annual
Details marks the year the money runs out with a dark red line across its row, and under Show All
shows the year-by-year inflation and market return that caused it. The Current $ toggle deflates by the inflation that path actually lived.
A banner names what is being replayed: its rank in the run, whether it survives, and for a stress
row the historical start year. Its ◀ ▶ buttons step to the previous or next path - captured paths
in outcome order, stress scenarios in the stress table's current order. Exit with the banner's
button; editing any input or leaving Charts
and Annual Details also returns you to your plan's own projection. Replayed paths are the exact
sequences the run scored, never re-created approximations.
## 11.1642
### Roth can fill a spending shortfall, Ordered offers three more draw orders, and Monte Carlo always runs yours
1. **A new switch, *Roth before Brokerage*,** under Cycle Brokerage. Off is the default and is what
the tool has always done: when a year's spending needs more than your strategy itself withdraws,
the shortfall comes from Cash, then Brokerage, then Roth. On, Roth is taken ahead of the brokerage
account, and Cash still comes first either way. That avoids realizing capital gains, but it
cuts both ways, and not gently: across 60 test plans it gained as much as $471,000 and lost as
much as $634,000, and it was negative in 26 of them. Spending Roth preserves the brokerage
account but drains the one that grows tax-free. Your numbers do not move until you turn it on,
and the *Ordered* strategy greys the switch out, because it draws in the sequence you chose.
2. **The Optimizer sweeps both.** Every strategy except *Ordered* appears a second time in the
table, marked 🅡, so the comparison is in front of you rather than something to run twice. Those
rows appear only if you hold Roth. Nothing about a plan predicts which way it lands, which is why
the switch is swept rather than recommended.
3. **Three more draw orders for the *Ordered* strategy.** Added: **Cash→Brok→Roth→IRA**,
**Cash→IRA→Brok→Roth** and **Brok→Cash→IRA→Roth**. The list is ordered with the sequences that
most often came out ahead in testing at the top. Cash→Brok→Roth→IRA, the new first entry, was the
best sequence in more test plans than any of the three previously on offer, and in one plan it was
worth $858,000 more than the best of them. Your own selection is unchanged, and both the Optimizer
and Monte Carlo now score all six.
4. **Monte Carlo now always runs your own plan, and marks it.** Compare every strategy could leave
your plan out of the run and emphasize a different one in the chart. It affected the *Ordered*
sequences, *Guyton-Klinger*, and any plan set to an IRMAA tier or an ACA cliff: the chart's
📍 marker landed on the wrong row or on none. Your plan is now always in the table, pinned to
the top and drawn in the chart, even when it is not one of the strategies the sweep covers - a
run with such a plan is one row longer than before. Clicking a row also loads what it says: the
sequence, tier, cliff or guardrails now come back with it instead of leaving whatever was in the
sidebar.
5. **Fewer *Reduce IRA* and *IRA Draw* rows.** The Optimizer tried 16 plan lengths between 2 and 25
years and 8 draw percentages; it now tries **3, 7, 11, 17 and 23** years and **5, 7, 9, 11 and
13%**. Neighboring steps differed by very little and between them the two families were half the
table, so a run is a third shorter and the table is easier to read. Your own setting still appears
as its own row whichever length or percentage you pick, including above 13%.
---
## 11.1629
### Monte Carlo: a second synthetic model, and prices that move
1. **There is now an arithmetic model.** *Synthetic - AAM* reads the growth rate you type as a plain
yearly average, so the median it reports is the number you entered. *Synthetic - GBM* is the model
that was already there, reading the same rate as a drift in logarithms, which puts its median a
point or so lower. Neither is the more optimistic of the two about how much money you end up with;
they differ in what the rate you typed *means*. Both draw the same market shocks from the same
seed, so switching between them compares the models rather than two different runs.
2. **Both models now vary inflation.** Inflation used to be one rate repeated every year. Each path
now lives through its own, tuned to US consumer price data for 1948 to 2025: a bad stretch tends
to stay bad rather than reverting next year, and inflation leans high in the years returns are
poor - which is the pairing that actually breaks a plan.
3. **The Mode presets set how much inflation moves, and show which setting you are in.** A preset is
green while its values are the ones in effect, and clears the moment you change any of them.
*Fixed Inflation* holds inflation at your Assumptions rate and reproduces the model that shipped
before this release. *Pessimistic* leans the whole model toward a bad decade: lower growth, higher
volatility, and inflation that is more persistent, larger and more tightly tied to bad return
years. *Default* puts everything back.
**Behavior change.** A Synthetic result you saw before this release will not come back until
you click *Fixed Inflation*: the returns are identical to the digit, inflation is what moved.
Historical mode is unchanged.
### Also in this release
- Annual Details can show each year's inflation, how far prices have risen since the plan started,
and that year's market return, under **Show All**.
- loopMs - the time spent calculating each year of data - was rendering incorrectly and has been
fixed.
- Opening the page with `?tab=montecarlo` in the web address left the self-check badge stuck on its
neutral hourglass. Fixed.
---
## 11.1601
### Optimizer tab reorganized
Labels, column names, colors and symbols are now consistent with the rest of the tool: the table
uses the same names as the summary bar (End Wealth, All Taxes, All RMDs, Spendable), and what was
called both "Infeasible" and "target unreachable" is now **unreachable target** everywhere.
The table shows the columns the **Optimize for** goal actually uses. Some were never shown at all:
Maximum Roth and Roth Conversion Effectiveness both ranked partly on a Roth balance that had no
column, and Tax Flexibility ranked on a spread across your pre-tax, Roth and taxable money that was
shown nowhere. **Final Roth**, **Final IRA** and **Mix Spread** now exist. Optimize for also says in
one line what the goal ranks by, and a switch below it shows all columns at once.
**Spend Goal** and **Yrs Funded** left the table; hover over any row and both are in the pop-up.
The legend folds out of the way, and the comparison marker appears only on the row being compared
against. Under Roth Conversion Effectiveness and Earliest Break Even, the ⚓ reference row is
now the best row that actually converts, rather than a plan that never converts and so has nothing
to compare on.
**Conv Tax** is now colored: green where the extra conversions lowered lifetime tax, red where
they raised it. The stop-after-year mark on conversion rows is explained in the symbol key.
**End Wealth** and **All Taxes** are shown under every goal, directly after Rank.
A `?tab=` web address opens
the page on a chosen tab, for example `?tab=optimizer`, `?tab=annual` or `?tab=montecarlo`.
No numbers change. These are visual improvements only.
---
## 11.15e3
### A plan that has run out of everything else will now spend from Brokerage
Each year prices its taxes (up to 4 times) because the first estimate cannot know what the
final tax bill will be. A late step exists to cover the gap when the bill lands higher than expected, and
that step could draw Cash, then Roth, but never Brokerage for fear of creating a "tax death spiral".
That possibility was finally measured - and two additional taxation passes is always enough to converge.
Previously a plan that had already emptied its IRA, its Cash and its Roth
would report year after year of unpaid spending while a seven-figure Brokerage balance sat there
untouched, because the only step left refused to look at it. On the sharpest test plan that was ten
unfunded years out of twenty-four, with just over $1,000,000 in Brokerage. It is now one.
**Behavior change.** Most affected is "Fill Fed/IRMAA Bracket" with an IRMAA tier selected, the shape
that tends to drain the IRA early. Such a plan now pays for more of its spending and finishes with
less left over, because the money is spent instead of passed on. If that trade is not the one you
want, it is worth knowing the plan was never really funding those years before; it was reporting a
shortfall and holding the money back. Saved scenarios and shared links for affected plans will show
different numbers than they did.
The companion idea, letting the very last funding step spend Brokerage before pulling extra from the
IRA, was measured at the same time and **rejected**. It rescued exactly the same plans and left far
more spending unpaid elsewhere by spending the Brokerage too early and having none left later.
---
## 11.15cf
### The IRMAA ceiling was aiming two years of inflation too low
IRMAA bills the premium for a given year against the MAGI you reported two years earlier, and it
compares that MAGI against the thresholds published for the **billing** year. The engine had the
billing half right: it reads the MAGI from two years back and measures it against thresholds
inflated to the current year, which is what SSA does.
The targeting half was wrong. Every ceiling that caps *this* year's MAGI to stay inside a tier used
*this* year's threshold, when the MAGI it is capping will be judged against the threshold published
two years later. At 3% inflation that aims about 6% low. On the MFJ Tier 1 floor of $218,000 it is
roughly $13,300 a year of IRA spending or Roth conversion room that was never used, every year of a plan.
Two paths change as a result:
- **IRMAA Ceiling strategy** ("Fill Fed/IRMAA Bracket" with a tier selected). Live.
- **QCD "As Needed"**, which donates only as much as it takes to drop two IRMAA tiers. Live, and
the place the correction is worth the most.
**Behavior change.** An IRMAA Ceiling plan now converts more, a QCD "As Needed" plan donates
slightly less, and both shift the numbers in a saved scenario or a shared link. At 0% inflation
nothing moves at all, which is what makes this an indexing fix rather than a new policy.
The projection of the future threshold is always performed. A setting chooses how much extra room to
leave below the projected future threshold. "No margin (aim right at the projected
threshold)" - is riskiest because slightly less future inflation could result in exceeding the future threshold and incur an increased IRMAA penalty.
The other side of the argument is that a future threshold set too low wastes potential IRA spending or Roth Conversion space year after year.
**Behavior change: your "as needed" QCDs get smaller.**
Risking an IRMAA penalty and money leaving for charity are not the same decision. Tests show that the
money leaving the QCD to stay under a future threshold were exceeding the value of avoiding the IRMAA surcharge. For that reason,
the QCD "as Needed" contribution always uses the future threshold projection, and doesn't leave a "margin".
The withdrawal/conversions however are capped according to a projection that leaves a safety margin.
### A selectable safety margin (hidden for now)
An "IRMAA Safety margin" setting has been added to tune the behavior.
- **Half the next-tier surcharge** (default). Looks up what crossing this particular boundary
costs per year and holds back half of it, so the setback scales with the size of the cliff.
- **None**, **$1,000**, **$2,000** - fixed setbacks below the projected threshold. These are the "riskiest".
- **Half the expected inflation**, **1% less than expected inflation** - hold room back by
projecting the threshold forward at a slower rate instead of subtracting dollars.
It is gated because it is still being measured, not because it is dangerous.
---
## 11.15c9
### More accurate SALT (state and local tax) deductions
**Behavior change.** Three corrections to how state and local taxes are deducted, all affecting tax
years 2026 through 2029:
- **Property and other local taxes can now be included.** Previously only state income tax counted
toward the SALT figure. Add yours with `?ptx=25000` in the web address - state income tax is
already supplied elsewhere, so enter only property and other local taxes here. Optionally set how
it grows over time with `?ptxm=flat` or `?ptxm=custom` plus `?ptxr=2`; the default is to grow with
inflation. Sharing a link and saving a named scenario both keep the figure.
- **The higher $40,000 SALT cap now runs through 2029**, as the law allows, rather than reverting to
$10,000 a year early.
- **The SALT cap and the income limit that phases it out now step up 1% a year**, so 2026 uses
$40,400 and $505,000.
For plans converting heavily in a high-tax state, these can change the recommended conversion size.
Note that this tool tests only SALT against the standard deduction - it does not model mortgage
interest, charitable giving, or medical expenses - so the benefit is limited to households where SALT
alone exceeds the standard deduction, and it ends after 2029.
---
## 11.15a2
### Five Payment Plans - Was 3 plus Quarterly
Previously comparisons were not apples-to-applies (same time frames). Now there are 5 plans now rather than 4 - Quarterly payment was not called out as a separate plan,
so it became "Plan Q". Each plan is measured up to April 15th so they are on equal footing - previously one plan
measured to April 15th of the following year, but the others did not.
Each plan is a full computation in its own right.
Draw timing and tax-payment timing are separate levers, and the tool now says so.
A star (★) marks the cheapest plan(s)
- **Plan A, Early.** Draws and conversions in January (or next month if already past January), tax withheld at the draw. [Was Plan B]
- **Plan B, Hybrid.** Conversions early (January or next month), draws and withholding in December. Shown only when
there is a conversion to pull early; without one it would be a copy of Plan C, and the table now
says that in a line rather than letting the column vanish. [Was Plan A]
- **Plan C, Late.** Everything in December. [Same as prior]
- **Plan D, Split.** New. The spending part of each draw is taken early with no withholding, and the
tax part is held back to a separate December draw withheld up to 100%, which Form W-4R permits.
You get your spending cash on the early schedule and the tax still gets the pro-rata credit that
only withholding earns. Total draws are unchanged: D never adds a supplemental draw, because that
would create taxable income your entered tax figures do not include. Useful if you want to get your spending money early,
and pay taxes when there is more tax certainty - however there are some costs and potential gotchas.
- **Plan Q, Quarterly.** Spelled out separately, but was being compared to in the prior version. December draws with no withholding at all, and the whole liability paid
as quarterly estimates.
Some minor corrections were also made:
- Using weekdays where previously they were not being used. Payments that may
have landed at the end of the year or past deadlines are moved earlier - for example when running the Tax Planner at the end of the year, compliance requires timely payment within the year.
- In some cases the plan reported that there were no penalties, but the logic was inconsistent.
- Roth gains from an early conversion appeared as if they were costs, rather than improvements.
- The tool failed to point out that late draws carried the consequence of not getting "spending money" until late in the year. This is now spelled out.
- *Citations are clickable* - in plan descriptions the codes cited are now clickable.
- **The Income Tax Planner button** In the [IncomeTaxPlanner Tool](standalone/IncomeTaxPlanner.html) was non-operational and is now fixed.
- **Safe harbor** now states whether Federal safe harbor is met, and does so using either the input fields, or by inferring net income.
- **Withholding is credited in equal parts across every due date**, whatever date it actually
happened [IRC 6654(g)], so a December draw can still satisfy a quarter that has already gone by.
- **An estimated payment counts on the day you pay it**, so an installment already past cannot be
made timely at all.
- Plans are ranked by first-year cost, and an **underpayment penalty** is not part of that cost. When the cheapest plan is one that misses safe harbor,
the winner line and a header badge now say so, because the ranking alone would hide it.
- The planner no longer credits withholding to money that has already moved. You can indicate that some of the payments or withdrawals have already been done and the planner no longer assumes that "completed" items had the proper withholding. Instead you must
- **Tell the planner the figure** and it is credited in full, exactly as before but on your numbers.
- **Leave it blank** and the planner credits nothing. It then schedules estimated payments for the
shortfall, which overstates what you still owe rather than understating it. That is the safer
direction to be wrong in, and the note on the plan says so in as many words.
**Saved URLs will now compute correctly and may show the newer (not prior) plan numbers.**
---
## 11.1585
**Two Stress Test fixes. No change to any plan or its numbers.**
### The Stress Test now updates when you load a saved scenario
Loading a scenario refreshed everything in the summary bar except the Stress Test, which kept showing
the result for whatever plan had been on screen before. It corrected itself only if you edited a field
by hand or opened the Monte Carlo tab, so the most likely reading was that the scenario had not loaded
at all. It now refreshes with the rest of the summary bar. The same load also raises the "out of date"
notice on the Monte Carlo tab when a full comparison run is on screen, which it should have been doing
and was not.
One case is still open, and is not fixed here: if you load a scenario in the second or so while a
Stress Test pass is already running, that pass finishes last and leaves its own, now older, numbers on
the tile. Editing any field or opening the Monte Carlo tab corrects it.
### "All start years" no longer calls every year one of the worst
The Stress window offers two selections. Combined ranks every start year over each of five lengths and
runs the worst of each. All start years runs the entire record, good years and bad alike. The result
sentence described both the same way, so in All mode it read "in 13 of the 98 worst historical periods
on record" about a set that is simply every year there is. It now reads "in 13 of the 98 start years on
record", and the tooltip on both the headline and the summary tile explains the selection actually in
use. Combined is unchanged, where "worst" was accurate all along.
The README's Stress Test section had drifted from the code and is corrected: it described six ranking
lengths including a 25 year window that does not exist, and quoted a fixed count of 36 start years that
moves with the Stress sequences box.
Additional elements were added for P32 research.
---
## 11.1581
**Self-check fix. No change to any plan or its numbers.**
The page runs its full test suite at load and reports the result in the indicator at the top. That
indicator has a staleness guard: the page is told how many tests each suite on disk is supposed to
contain, and if the real number differs it refuses to report green, because a page that has lost
count of its own tests cannot honestly claim they all passed.
Tax Payment Planner release 1.1580 added 2 tests to its own suite and did not update that expected
count.
**This file had an editing pass of its own.** Including fixing dead links, an empty link, and more.
---
## 11.1553
**A teaching demo that shows, in one table, how the number of Monte Carlo paths matters. No change
to any plan or its numbers.**
### The `?montecarlo` demo
Adding `?montecarlo` to the page address (for example
`retirement_optimizer.html?montecarlo`) opens the Monte Carlo tab in Synthetic mode, exposes the
Seed, Paths and Input Distributions controls, and automatically runs a small experiment: the plan
currently in the sidebar, simulated with three random seeds at each of four path counts (5, 10, 25
and 100). Every cell goes through the same engine and reports the same numbers a manual "My Plan
Only" run would, so nothing in the table is a special demo figure.
The table lists, for each seed and path count, the sampled Equity range (the worst and best single
year across all the paths in that run) and the Inflation range. The point is visible at a glance:
with 5 paths the equity range swings widely from one seed to the next, and by 100 paths it has
settled. The Experiment button reshuffles the three seeds so a reader can watch a fresh random draw
land in the same structure. The Input Distributions panel below is drawn from the last run in the
grid (100 paths) and, in this mode, shows its worst and best sampled lines by default rather than
hiding them behind a legend click.
Synthetic inflation is still a single fixed rate today, so its column shows one value; the table is
built to display a real range automatically once Synthetic paths gain a varying inflation series.
### The Paths box can now go below 100 in advanced and demo modes
The Paths field is normally held at a minimum of 100, because the survival rate and the percentile
bands stop meaning much at very low path counts - but experiments are allowed down to 3 paths.
The default experience is unchanged.
---
## 11.152f
**A Stress Test release: two crashes fixed, the ranking windows can now be combined or skipped
entirely, and failures are graded against the length of your plan.**
**Behavior change:** the Bear-start overlay now draws its opening from the worst 3, 5 and 10 year
stretches rather than the worst decade only, which moves Historical Monte Carlo results. Nothing
else here changes any plan or its numbers.
### Asking for 85 or more stress sequences froze the tab (nerdknob mode)
The Stress Test ranks historical retirement start years and runs the worst of them. How many there
are to rank depends on the window, because a start year has to have the full window of real data
after it to be scored at all: a 5 year window leaves 94 candidates and a 30 year window leaves only
69. Asking for more sequences than the record can supply used to run off the end of the ranked list and freeze.
The count is now capped at what the window can supply and anything above that is ignored.
### Clearing the Paths box printed "NaN"
The run size line under the Run button updates as you type. Emptying the box left it reading
"NaN paths x 144 strategies = NaN simulations" until you typed a digit. Every numeric box on the tab
now falls back to its default when empty and is held inside its stated range, so a half-typed value
cannot reach the run, the time estimate or the out-of-date check.
### One window at a time was hiding most of the bad years
Windows were increased to 5/10/15/20/30 and Combined. Previously picking one (10 was the default)
caused the worst 5-year start, for example, to be skipped because it was only looking at 10 year combined
growth rates.
Two new choices in the Stress window selector:
- **Combined (5/10/15/20/30)**, now the default, scores every window and keeps the union of what
each one flags. The count you set is per window, and because the windows overlap the union is
smaller than the sum: the new default of 20 gives about 40 distinct start years rather than 100.
Hovering over a row says which windows flagged that year.
- **All start years** skips ranking and runs every start year the record holds, currently 98. At
that density the individual lines stop being readable, so survivors fade back and the failures are
drawn solid and on top. The shape you are looking for is how much of history breaks the plan, not
which line is 1966.
Everything the ranking picks is still a real historical sequence run straight through, not a snippet added
at the beginning.
A window longer than your plan is trimmed to the length of the plan, so a 12 year plan combines
5, 10 and 12 rather than considering 15, 20 and 30 year stretches.
The defaults are now 20 sequences and the Combined window, up from 10 and a single 10 year window.
### Red and amber now mean early and late in YOUR plan
A failure now is marked red if ruin occurs before the halfway point, yellow if it lands on or after the halfway point, and
green if the stress for that sequence ends in assets still present (success rather than ruin). Early ruin is worse
because it's harder to avoid.
### The scenario table reports the whole plan, and the worst stretch inside it
Real CAGR is now calculated over your whole plan, for every sequence, and it also
calculates the worst 5, 10, 15 and 20 year real return found anywhere inside it. A sequence can open
calmly and still contain the decade that breaks the plan. The old stress test might not have discovered the
true worst scenario. To save space, the equity, bond, international and inflation rates moved into the hover-over data.
### Sequences that run past the end of the record are now marked
A plan longer than the record has left after its start year wraps around and replays history from
1928. A 2015 start on a 30 year plan gets 11 real years and then 19 of replay. That has always been
the behavior, and the ranking has always excluded start years without a full window of real data
after them, so a wrapped stretch never gets a vote in which years are worst. It is now visible: the
graphed line becomes dashed. Hover over the row to reveal the year the record runs out.
### Bear-start openings are no longer all the same shape
**This one changes Historical numbers.** Bear-start overlays a bad historical opening onto a quarter
of the simulated paths, on the grounds that block sampling on its own shuffles the year order and so
under-produces the sustained bad opening that actually breaks a retirement.
It drew that opening from the worst ten *decades*, and averaging over a decade washes a crash out.
1930 is the worst opening in the whole record measured over three years, at -26.9% a year after
inflation, and only the thirteenth worst measured over ten, at -0.4%, because the decade starting
1930 contains 1933's +54% rebound. Ranked on decades, the pool came out as 1999, 1965, 2000, 1969,
1968, 1966, 1972, 1973, 1970 and 1929: six of the ten being 1960s and 1970s stagflation, 1930 absent
altogether, and only 1929 opening worse than -15% a year over its first three years. And because
every draw was spliced for a full ten years, a 1929 draw arrived with its own 1933 to 1936 recovery
attached. The overlay could not produce "crash, then whatever comes next", which is the shape people
mean by a bear start.
The pool is now the worst ten start years over each of three, five and ten years, and each one is
spliced for the length that flagged it. A three-year draw gives three brutal years and then hands
back to the random draws; a ten-year draw behaves as before. Thirty entries drawn from twenty-one
distinct years, averaging a six-year opening. A year that several lengths agree on, like 1929,
appears once per length and so is drawn proportionally more often.
The fraction (only visible in nerdknob mode) stays at 25%. Measured
over the three to five year openings this overlay now mostly draws, the frequency is about 22%, so
25% is a modest and deliberate over-weighting rather than a base rate, and the tooltip now says so.
Also fixed here: an opening longer than the whole plan used to write past the end of that path's row
and corrupt the start of the next one. It is capped at the plan length now. This only affected plans
shorter than ten years.
### Two run buttons are now visible, and each states what it will cost
The tab had one button, "Run Monte Carlo", which ran every withdrawal strategy the Optimizer knows
how to build so it could rank them. That is about 144 strategies, and the path count is per strategy,
so the default 500 was 72,000 simulations and about 43 seconds of work. A second button only runs your
own plan.
- **My Plan Only** is listed first and is the new default. It answers "how did my plan do", which is the
question most people want to answer. It costs about one 144th of the work. You get your chance of success
and the percentile chart, but no strategy ranking table.
- **Compare All Scenarios** is the old behaviour, renamed to say what it actually does.
Each button carries its own expected time, measured on the machine you are running on rather than
guessed: "My Plan Only (1.3 sec)", "Compare All Scenarios (43 sec)". The first figure shown is an estimate;
after one run it reports the time actually measured.
### Changing a Stress Test setting no longer marks the whole run out of date
Editing Stress sequences or the Stress window used to raise the "Out of date" banner over the whole
Monte Carlo sweep. That banner's Re-run button re-runs everything, so it was offering half a minute
of work on the roughly 144 strategy sweep to refresh a chart that had already refreshed itself by the
time the banner appeared. Only a change the sweep itself would have seen marks it out of date now,
and the stress pass keeps re-running on its own either way.
### Smaller things
- The Stress Test chart legend was dropped. The table underneath already names every line,
carries every number, sorts, and shows each line's color in its first column. The table made the legend
a waste of space so only the table remains.
- The hover-over data is not shown until you pick a line from the table. Click the row in the table again to bring
all the graphed lines back.
---
## 11.150b
**The Monte Carlo tab now shows you your own plan, says how big the run really is, and the Stress
Test is readable. No change to any plan or its numbers.**
**Your plan is pinned and its chance of success is stated outright.**
The survival table ranks every withdrawal strategy against every other, which answers "what is best"
but never answered "how did mine do". Your own plan was a row somewhere in a table of about 144,
sorted by somebody else's metric, with nothing marking it. It is now pinned with a 📍 to the top of
the table and stays there whatever column you sort by, it is drawn as the thick line on the chart,
and a sentence above the chart names it and gives its chance of success directly: how many of the
simulated paths it survived, and the median balance it ended with. If your exact settings are not
among the strategies the sweep covers, the page says that plainly instead of pinning a near miss.
**"500 paths" was never 500 simulations.**
The Paths box is the number of paths run per strategy, and the sweep runs every strategy it knows
how to build so it can rank them. On a typical plan that is about 144 of them, so the default 500
paths is 72,000 simulations covering 1.8 million years for a 25 year long retirement. Nothing in the
tool said so, which made a run that takes half a minute look like it should have been instant. The
box is now labeled "Paths (per strategy)" and the full arithmetic, "500 paths x 144 strategies =
72,000 simulations, 1,800,000 simulated years", appears beside the Run button and under the survival
table. 72,000 retirements simulating about 1,800,000 years is why the wait is what it is.
**The Stress Test chart is legible.**
Every line used to be labeled "1966 (eq: +6.0% inf: +7.0% real: -1.0%)". Ten of those made the
legend wider than the plot it was explaining. Each line is now labeled by its start year and what
happened to the money, for example "1966 ✗2041" or "1999 ✓", and every statistic that used to be in
the label has a sortable column in a new table under the chart: start year, outcome, ruin year, years
to ruin, equity CAGR, inflation CAGR, real CAGR and final balance. Click a row to isolate its line,
the same as clicking the legend.
The colors changed meaning. They used to run dark red to amber by how bad the opening stretch was -
ten years by default, but it may instead be 5, 15, 20 or 30 - which is the reason a sequence was
picked, not what became of your plan in it, so a scenario your plan sailed through could still be
drawn in alarm red. Red now means the money ran out inside that opening window, amber means it ran
out later, and green means it never ran out. Sequences are listed
worst first: earliest failure at the top, survivors at the bottom.
**The ten-year window is now adjustable, and what it does is documented.**
Under nerdknob there is a new Stress window control offering 5, 15, 20 or 30 years instead of the
default 10. It changes two
things at once, on purpose: which historical start years count as the worst, since they are ranked by
real return over that stretch, and where the line falls between an early failure and a late one. A
window longer than your plan is trimmed to the length of the plan, and the caption under the chart
always names the window actually in use.
Two long-standing behaviors are now written down in
[Stress Test vs Monte Carlo Analysis](README.md#stress-test-vs-monte-carlo-analysis), because both
surprise people. The
window only *ranks* start years, it is not a splice point: after it, each scenario keeps following
the real historical record for the whole length of your plan, wrapping back to 1928 when it runs past
the end of the data. And the seed does not move the Stress Test, because there is no random number
drawn anywhere in it. The seed tooltip now says so.
**Inflation is colored by whether it hurts you, not by its sign.**
Every figure in these summaries was green when positive and red when negative, which is right for a
return and backwards for inflation: 8.7% inflation was drawn in the same green as a good year for
equities. Inflation now reads red when it is above zero and green when it is at or below zero,
everywhere it appears. The other series are unchanged.
**Both banners are now the control that folds their section.**
The Stress Test starts folded, since its banner already carries the result and the chart plus
scenario table underneath are a lot of page to open unasked. The Chance of Success section starts
open.
The Stress Test had a plain "Stress Test" heading to click and a colored result banner underneath
it, so the one thing worth reading while the section was collapsed was not the thing you clicked to
collapse it. They are now the same element: the banner carries the heading, the count and a chevron,
and clicking anywhere on it opens or closes the section. The main Monte Carlo block works the same
way, with the Chance of Success banner moved up above the strategy table it introduces, folding the
table and the chart together. Both start open.
**The run size now says how many years it is simulating.**
The readout gains a third figure: "500 paths x 144 strategies = 72,000 simulations, 1,800,000
simulated years". Each of those simulations is a full plan run with a tax return computed in every
year, so the year count, not the simulation count, is what the running time is actually proportional
to. It is the number that explains the wait.
**The Historical summary line reads like the Synthetic one.**
Synthetic mode described itself in a sentence, while Historical rendered a four-row Min/CAGR/Max grid
wedged into the same line, leading with the least useful figure. Historical now uses the same
"·"-separated sentence, per asset, in CAGR then worst year then best year order, with the same red
and green coloring so a negative number is visible without reading it. Both modes now put the CAGR
first. The stress panel's own summary line changed with it.
**Reordering and two additions in the Stress Test.**
The survival table sat between the Stress Test and the chart that goes with it; it now sits after
the Stress Test, next to its chart. The scenario table gains Bonds CAGR and Intl CAGR columns, so a
sequence where equities and bonds both fell is distinguishable from one where bonds held up. Those
two figures were never computed before, only equity and inflation were. International data begins in
1970, so scenarios starting earlier show domestic equity as the proxy, which is the same substitution
the simulation itself makes. The Outcome column now says "Ruin" for both early and late failures: the
color already carries that distinction and the Ruin Year and Yrs to Ruin columns give the exact
answer.
**The historical record label said 1928-2024 and had not been true for a while.**
The embedded data runs through 2025 for equities, bonds and international, and the code has always
used all of it. Only the labels were stale. They now say 1928-2025.
**You can now run Monte Carlo against your own plan alone (nerdknob).**
*This part is deliberately absent from the short changelog inside the page. It adds a control
without changing what anyone gets by default, so it is recorded here for the record rather than
announced on the Documentation tab.*
The tab has only ever had one kind of run: simulate every withdrawal strategy and rank them. That is
about 144 strategy arms, and it is why a run takes half a minute. Most of that work answers "which
strategy is best", not "will my plan hold up", and the second question needs 500 simulations rather
than 72,000.
Under nerdknob there is now a second button, **Run My Plan Only**, which runs just the plan in the
sidebar. On the default scenario it finishes in under two seconds instead of around thirty-seven, and
it produces the same answer: the same 66.6%, the same 333 of 500 paths that the pinned row reports
after a full comparison run. You get your chance of success, the percentile chart and the Stress
Test; you do not get the strategy comparison table, because a one-row ranking is not a ranking, so
that table is hidden in this mode.
**Nothing changes unless you press the new button.** "Run Monte Carlo" still means compare every
strategy, and that is still what runs automatically when you open the tab, which is what people
without nerdknob get. Switching between the two marks the previous result out of date rather than
leaving the other mode's answer on screen.
This was cheap to add because it required no engine change at all: a single-variation run is exactly
the shape the Stress Test pass has used since it was introduced, so the worker was already able to do
it.
**Fixed a potential freeze when running Monte Carlo,** along with a progress bar that could sit at
the same number for a whole run and a Cancel button that would not take effect until the run ended
on its own. Only the downloaded copy of the tool, opened directly as a file, could freeze; the
version served from the web was never affected.
**Three fixes came out of the same work.**
Toggling Current Dollars re-drew the main chart but not the Stress Test chart, which silently stayed
in nominal dollars underneath it. Both now switch together, along with the new table. Sorting a
column whose blanks are stored as "no value", such as Exhausted or Total Taxes, compared two blanks
by subtracting them, which produces a meaningless result and left those rows in arbitrary order; the
comparison is now done properly. And the "paths" readout beside the Cancel button described the
previous run rather than the one in flight. Hovering over a line on the main chart could also name a
different strategy than the one its legend entry named.
---
## 11.1508
**Self-check fix and documentation. No change to any plan or its numbers.**
The page runs its full test suite at load and shows the result in the indicator at the top. That
indicator has a staleness guard: the page is told how many tests each suite on disk is supposed to
contain, and if the real number differs it refuses to report green, because a page that has lost
count of its own tests cannot honestly claim they all passed.
Release 11.14e1 added 11 tests to the engine suite and did not update that expected count, so the
guard fired as designed and the indicator read "test counts changed" with the console
detail `optimizer_core: 244 tests on disk, 233 expected`. Nothing was failing. The expected count
is now 244, which is the number `node optimizer_core.tests.js` actually reports, and the indicator
is green again.
Also in this release:
- **[README.md](README.md) now explains [Monte Carlo and "Chance of Success"](README.md#monte-carlo-and-chance-of-success-accuracy) properly.**
Walks through what a Monte Carlo simulation is doing, why results from two different
tools are not comparable, why more iterations cannot repair a sampling method that does not
resemble the real world, why a model that holds inflation fixed is not good enough, why real
markets trend rather than lurch randomly year to year, and why a seeded (repeatable) random
sequence matters. It ends with what you can and cannot conclude from any tool's success
percentage.
- **The [tax-withholding maneuver](README.md#why-end-of-year-vs-quarterly) section gains "the *other* maneuver":** using the once-per-year
IRA-to-IRA rollover, with repayment, to pay taxes through withholding, and the five separate
jobs that one pair of transactions can do.
---
## 11.14e1
**Research groundwork. No change to any plan or its numbers.** Nothing on the page behaves
differently.
What was added, and why:
- **Four engine research inputs**, used only by the offline study harnesses.
- **Five research studies were run and recorded** (results live beside the harnesses). The
highlights below will shape future releases:
[Brokerage Research](research/BROKERAGE_DRAW.md),
[Proportional Draw Research](research/STRATEGY_FAMILY_RANKING.md)
(it's not as optimal as thought), and
["EndGame" Research](research/ENDGAME_DRAW_ORDER.md) -
what is the best strategy once the IRA is under control.
---
## 11.14dd
**Behavior change** (Ordered strategy, RIBC and BIRC sequences only).
The Ordered strategy draws your accounts in a strict sequence you choose (CBIR, RIBC, or BIRC). Any
year your guaranteed income and required withdrawals come to more than you spend, the leftover has to be
banked somewhere. Until now it always went to Cash, regardless of the sequence. For the two sequences
that draw Cash last (RIBC and BIRC), that stranded the surplus in the one account the strategy would not
touch again until everything else was gone: it earned the cash yield instead of the market return, and
it was not spent until far too late.
Now the fill follows the same order as the draw. The surplus is banked in the first account your
sequence would draw that can actually receive a deposit, which is Cash or Brokerage. Roth and the IRA
are contribution-limited and cannot take an arbitrary after-tax deposit, so they are never fill targets.
That resolves to:
- **CBIR** (Cash first): Cash, exactly as before. No change to any CBIR plan.
- **RIBC** (Roth, IRA, Brokerage, Cash): Brokerage, because Brokerage is drawn before Cash.
- **BIRC** (Brokerage, IRA, Roth, Cash): Brokerage, because Brokerage is drawn first.
An explicit Cash Reserve target, or Cycle Brokerage, still takes precedence over this rule where set.
Effect: RIBC and BIRC plans that ran a surplus now end with more wealth, because the banked money grows
at the investment return rather than the cash yield and is drawn back sooner. On one 30-year test the
gain was roughly 4% of ending wealth for RIBC and 7% for BIRC. CBIR plans are byte-for-byte unchanged, as
are all other strategies. The withdrawal sequence itself was already correct: it restarts from the top
every year and uses whatever balances exist that year, so a refilled account is drawn again in its
proper position.
---
## 11.14c6
**The suggested spending goal** (the figure behind the ⓘ next to the spending goal) is now
solved by the plan itself, and considers how long the retirement is, and when any pension
is planned to start.
The old suggestion used 5% of every account, plus Social Security and pension,
taxed once. It ignored the length of the plan entirely, so a 12-year retirement and a 40-year
retirement got the same 5%. That is contrary to what the withdrawal-rate research suggests.
A shorter horizon can support a meaningfully higher rate; a longer one needs a
lower one.
The suggestion now runs your current strategy at many spending levels and reports the highest one that
still leaves a cushion in the portfolio at the end. "Cushion" means the last modeled year still
holds at least five years of the spending the portfolio
must cover after Social Security and pension. Because it runs the real engine rather than a
formula, taxes, required minimum distributions, and the timing of a deferred pension or an
unclaimed Social Security benefit are all handled correctly - the previous method did not.
A closed-form amortization of the invested portfolio (using your own growth and inflation
assumptions) seeds the search, and the tooltip reports the result as a percentage of that naive
amortization so you can see how much the taxes and the end-of-plan cushion shaved off.
What to expect: the calculation will vary with the withdrawal strategy, and may differ from what the
Optimizer calculates under "Maximize Spend". Do remember that the proposed spend is a THEORETICAL, not
a "best" option - and does not properly account for Sequence of Returns Risk (SoRR).
Anything suggested should be vetted first against the Monte Carlo (especially the stress test)
analysis, and ultimately scrutinized by a CFP (Certified Financial Planner).
More thorough proposed spend goals are being considered for the future.
---
## 11.1499 (behavior change)
When someone dies, the cost basis of a taxable brokerage account resets to what the account is
worth that day. The gain that built up during their lifetime is never taxed, to them or to their
heirs. The tool was ignoring this in two separate places:
1. **At the first spouse's death.** The survivor's cost basis carried straight on as if nothing had
happened, so every later sale from that account realized more gain than it really would, and
paid tax on it.
2. **At the end of the plan.** The final year valued whatever brokerage was left over as if it had
been sold and the capital-gains tax paid. It is not sold, it is inherited, and that tax is never
owed.
Both are now modeled. The size of the first basis adjustment depends on where you live - the underlying
property law of the state dictates what happens. In a community-property state the whole account resets; everywhere else only the
share that belonged to the person who died, which the tool treats as half. This follows the State Tax selection, because it is a matter of law
rather than preference. The seven community-property states this tool models are AZ, CA, ID, NV,
TX, WA and WI. Louisiana and New Mexico are community-property states too, but they are not among
the 38 jurisdictions the tool models at all, so they cannot be selected. Alaska is deliberately
treated as common law, because community property there is something spouses have to opt into by
written agreement.
**What to expect.** Ending balances rise, lifetime tax falls, and the spending goal is funded in
exactly the same years as before. The demonstration plan the tool opens with goes from $625,885 to
$637,024 and pays $10,513 less tax over its life, with spending identical to the dollar. That plan
happens to spend its brokerage account down to nothing, so all of its gain comes from the first
death; a plan that still holds a large taxable account at the end can gain considerably more. One
test plan holding $900,000 of brokerage gained $398,712 in the final year alone.
**Why this matters for Roth conversions.** The error only ever ran one way. It penalized money
kept in a taxable brokerage account and left Roth and cash untouched, so every comparison between
converting and not converting was tilted toward converting. Correcting it makes conversions look
worse than they used to, and some plans that previously reported a conversion benefit no longer
report one. In one test case the conversion advantage reported was worth less than what was gained
by the step-up in basis.
**Break Even is deliberately unchanged.** It still measures the year conversions pull permanently
ahead, valued as though the account were sold rather than inherited. Break Even during your
lifetime and break-even measured against what your heirs receive are different questions, and the
second one is being left for its own release rather than folded in silently here. Break-even years
reported today are the same ones reported before this change, on the same basis.
**On the chart.** Every death now carries a marker, and the marker carries a small circle showing
which step-up applied: half filled for a half step-up, solid for a full one. A legend under the
chart says which is which, and it only describes the marks your plan actually has. A half step-up
cannot happen in a community-property state, or to a single person whose estate goes straight to
their heirs, so those plans get the solid circle explained and nothing about halves. The last
death is marked which is where the final step-up happens.
Also in this release: a brokerage cost basis larger than the brokerage account value is now corrected rather than
carried. A market decline can cause a declining basis - as would be observed in Monte Carlo Stress Cases.
Ordinary projection will not see a basis decline due to declining account value unless you put a
[negative growth rate in place purposely](https://tools.netcitizen.us/retirement_optimizer.html?str=ordered&os=RIBC&bb=90k&g=-1.0).
The place you would observe the declining basis is in the "Annual Details" -> "Brokerage" column "Basis".
---
## 11.147c
**Self-tests improved.** No change to your plan or its numbers. The self-check that runs at page
load now covers 510 tests instead of 245; hover over the indicator for the count. A brief hourglass
before the green dot is normal.
---
## 11.1478 (behavior change)
**Four tax and accounting corrections. The big one: dividends and interest were counted twice, so
plans have been showing more money than they should. Ending balances go down. Please read this even
if you skip the rest.**
### 1. Dividends and interest were credited twice
When your brokerage paid a dividend, or your cash earned interest, the same dollar was doing two
jobs. It was added to your balance, which is right. It was also treated as income available to
spend, so the tool withdrew that much less from your accounts to cover your spending. The dollar
bought your groceries and stayed in your account. Nothing ever took it back out.
The longer the plan and the higher the rates, the more money this invented. A cash account earning
4% grew at closer to 8%. In one test a plan spent $800,000 over twenty years while taking only
$2,449 out of the cash account. That money tree - unfortunately - does not exist so the modeling now matches reality.
You can check the old behavior yourself. Take a portfolio returning 8% a year. Have it earn that as
[8% growth with no dividend](https://tools.netcitizen.us/retirement_optimizer.html?str=ordered&os=RIBC&bb=90k&dr=1&g=8&div=0), then again as [6% growth plus a 2% dividend](https://tools.netcitizen.us/retirement_optimizer.html?str=ordered&os=RIBC&bb=90k&dr=1&g=6&div=2) that is automatically
reinvested. Same total return, same reinvestment, but the dividend version pays tax every year that
the growth version does not, so it must end up **behind**. After the fix, the 6%+2% example lands at $1,231,249 vs $1,250,627 net wealth (current dollars) for the 8% DRIP.
Dividends and interest are still income, still taxed exactly as before, and still land in your
account. The withdrawal strategy decides whether a dividend is spent or banked, and how the tax on it is paid (e.g. you can earn interest and pay the tax on that interest from an IRA, or from cash).
**Expect ending balances to fall, typically between 4% and 23%** on a plan using the default 3% cash
yield and 0.5% dividend rate. Your spending goal is still funded in the same years it was before.
Tax goes up slightly, because the invented money had been quietly absorbing some of it. A
Guyton-Klinger plan can also shift its spending path, since that strategy sets spending from what was an overstated - now correct - portfolio balance.
A plan that looked like it worked may now need a lower spending goal, a later start, or a different strategy.
### 2. Two 2025 tax-law provisions were implemented but never switched on
The senior deduction and the higher SALT cap from the 2025 tax act were both built into the tax
engine and then never actually applied to any plan. The result was federal tax that was too **high**.
The **senior deduction** is an extra $6,000 for each filer aged 65 or over, for tax years 2025
through 2028, phasing out above $150,000 of income for a couple. On a plan with two filers over 65
this is $12,000 a year of additional deduction. Both provisions now switch themselves off in the
year the law says they expire.
The **SALT cap** rise, to $40,000 through 2029, only matters if your state and local taxes are large
enough to make itemizing worthwhile, which for most retirees they are not. It is worth up to about
$950, in a narrow band of high income in a high-tax state, and nothing outside it.
### 3. States that exempt retirement income were taxing it anyway, in some years
16 of the 38 states the tool models exempt some or all of your pension and IRA income: Connecticut,
Georgia, Illinois, Mississippi, Iowa, Maryland, Michigan, New York, Pennsylvania, Virginia, Alabama,
Colorado, Kentucky, Maine, Ohio and Wisconsin. The tool works your tax out up to four times in a
year as it settles the final withdrawal amounts, and one of those passes was not being told which
part of your income was retirement income. Any year that reached that pass was taxed as though your
state had no exemption at all.
**Ordered** strategies took the worst of it, because for other strategies a later pass works the tax
out again correctly and hid the problem. Ordered strategies do not run that later pass, so the
faulty figure was the final answer. Across a 16-state test sweep the correction removed **$732,133**
of state tax that was never owed, **$685,487** of it on Ordered rows, and **23 of 36 Ordered plans
went from reporting failure to funding the whole plan**. On one Pennsylvania test plan whose only
income was Social Security, a pension and IRA withdrawals, all three of which Pennsylvania exempts,
lifetime state tax fell from **$28,055 to $9**. Plans in the other 22 states are unchanged.
### 4. Documentation that no longer matched the tool
The funding backstop was still described as a Fill Bracket / IRMAA feature. It has not been one
since 11.146a: once Cash, Brokerage and Roth are exhausted, nearly every strategy now draws the
extra IRA needed. What is specific to those two ceilings is that their draw goes *above* the
ceiling, which is what makes them soft.
**Ordered's exclusion was documented nowhere.** Like ACA Cliff, it reports a shortfall by design,
because the account sequence is the one you picked and it will not step outside it. A small
remaining shortfall on a CBIR, RIBC or BIRC row is the answer, not a bug. That is now said in the
README, in the strategy description, on the Annual Details legend and in the shortfall column
tooltip. The shortfall tooltip also used to say a shortfall was "likely due to errors in the
calculation", which was backwards.
One of the Ordered strategies was leaving a small amount of spending unfunded in two years; the
amounts moved and now sit between $10 and $161 across three years on the test plan. These are
rounding-scale residuals from the order in which tax is recomputed, not stranded capital.
---
## 11.146a (behavior change)
**Proportional, Reduce, Guyton-Klinger and the default strategy were sizing withdrawals as if Social
Security, pensions and RMDs arrived tax free. They now work the withdrawal out correctly, and they
draw the IRA needed to cover the whole spending goal, so plans that used to report a shortfall next
to a large IRA are funded.**
**What was wrong.** These strategies worked out the withdrawal by subtracting your guaranteed income
from your spending goal, using the full **pre-tax** value of that income. The tax owed on the Social
Security, pension and RMD money was never included, so every year came up short by roughly the
amount of that tax bill. While there was money in Cash or Brokerage the gap was quietly covered and
nothing looked wrong. Once those accounts ran dry the gap had nowhere to go, and the plan reported
unfunded spending while the IRA still held a large balance.
**What changed, part one: the plan can reach the money.** When Cash, Brokerage and Roth are
exhausted and spending is still unfunded, these strategies now draw the additional IRA needed, the
same backstop Fill Bracket and IRMAA Tier have always had. The amount appears in the `ForcedIRA`
column.
**What changed, part two: the first withdrawal is the right size.** Part one fixed the symptom, but
the reason the plan came up short every year was still there. The first withdrawal is now sized
against what your guaranteed income is actually worth **after** tax. That tax is computed, not
estimated with a single rate, because your guaranteed income is not taxed at one rate: Social
Security is between 0% and 85% taxable depending on your other income, qualified dividends have
their own 0/15/20% schedule, and pensions and required distributions are ordinary income. Applying
one blended rate to the whole amount overstated the tax by several times on a Social
Security heavy household and drew far too much.
**What this does to your numbers.** A plan that reported a shortfall in one or more years may now succeed.
**Which strategies change.** Proportional, Reduce, Guyton-Klinger, the default strategy, an ACA
Cliff plan after its cap ends at Medicare, and the Cyclic Brokerage option. Fill Bracket, IRMAA Tier,
IRA Draw % and Ordered set their withdrawal by their own rule, so they produce identical results to
before.
**ACA Cliff is deliberately excluded, and that is not an oversight.** While the cap is in force it
will still report a shortfall rather than draw more from an IRA. An IRA withdrawal is taxable income,
and going a single dollar over the cap forfeits the entire premium subsidy, so drawing more would
cost far more than the spending it funds. A shortfall on ACA Cliff means the spending goal could not
be met from non-taxable sources, which is the answer that strategy exists to give. Once the cap ends
at Medicare there is no subsidy left to protect, and from that year the plan is funded like any
other. On a plan whose cap runs to age 65, that shows up as seven capped years that still report a
shortfall, followed by twenty-one funded years.
**Also.** A plan that genuinely runs out of money still reports a shortfall. The change only affects
spending that could have been funded from an IRA that was sitting there.
---
## 11.1464 (behavior change)
**The ACA Cliff strategies are now available to everyone, the income cap ends when Medicare begins
instead of being enforced for the rest of your life, and the warnings tell you which year they are
talking about.**
Only plans using an ACA Cliff strategy change their numbers. Everything else is about what you can
see and select. As for what "Spend Goal" you should enter: enter your AFTER ACA spend goal. While ACA limits are in
place, your spending is capped to meet them. If you have lots of cash or Roth, or a high basis Brokerage,
you may be able to meet your spend goals in the first years of ACA by consuming those resources.
**1. The cap now ends at Medicare.** ACA premium subsidies stop being available once you are
eligible for Medicare. The tool was not checking ages at all: pick an ACA Cliff strategy and it held
your income under the Federal Poverty Level multiple you chose at 65, at 80, at 95, protecting a
subsidy that had ended decades earlier. On a plan that opens after 65 the cap was being enforced for
every single year.
From the first year in which **every living person in the plan** is old enough for Medicare, the cap
is dropped and the strategy behaves as **Proportional 0%**, drawing across your accounts to fund
your spending goal. Two details worth knowing:
+ It is **every living person**, not every person. If one spouse dies before reaching Medicare age
and the survivor is already past it, the cap ends that year. Those survivor years are exactly where
a single filer's narrower tax brackets used to strand spending under a cap that was protecting
nothing.
+ Until then the cap is measured against **household** income. If one spouse is already on Medicare
and the other is not, the older spouse's required distributions and Social Security still count
against the younger spouse's limit. The warning under the strategy selector now says this; it
previously claimed the limits applied only to the younger person.
On a couple aged 66 and 67 at the start with a $2.1M IRA and a $160,000 spending goal, the ACA 400%
arm changes like this:
| | before | now |
|---|---|---|
| years the cap blocked spending | 24 of 24 | 0 |
| spending left unfunded | $735,010 | $304,331 |
| total spending funded | $3,832,599 | $4,263,278 |
| ending net worth | $1,888,543 | $684,010 |
Ending net worth **falls**, and that is the point rather than a side effect. The old behavior looked
wealthier only because it refused to fund the spending goal and left the money in the IRA.
**2. ACA Cliffs: the four Federal Poverty Level ceilings
(200/250/300/400%) now appear for everyone**. The Optimizer sweeps those rows
for everyone. The tool models the income cap and **none** of the premium subsidy
that cap buys, so an ACA row shows you what staying under the limit costs and not what it saves. Read
one as a constraint study, never as a recommendation. The options disappear once both people
are on Medicare at the start of the plan.
**3. On Optimizer rows a ⚠️ indicates that the spend goal is infeasible.**
If one ACA row is flagged every lower one is flagged too.
**4. The Medicare warning now names the year and your ages in it.** It used to say only that you
would be on Medicare "at retirement start", while the age shown next to your birth year is your age
**today**. If your plan starts years from now those are two different numbers, and being told you are
on Medicare while the field beside it reads "Age 59" looked like the tool had stopped paying
attention. It now reads, for example, "At retirement start in 2031, you will be 65 and your spouse
79", and tells you that lowering **Retirement Start Age** is what brings the ACA years back.
---
## 11.1447 (behavior change)
**Roth conversion strategies now work in states with no income tax and in states with a single flat
rate.**
If your state is one of the 21 the tool models with a single tax rate, the strategies that fill a tax
bracket were converting almost nothing, however you set them. Those states are Alaska, Arizona,
Colorado, Florida, Georgia, Illinois, Indiana, Iowa, Kentucky, Massachusetts, Michigan, Nebraska,
Nevada, New Hampshire, North Carolina, Pennsylvania, South Dakota, Tennessee, Texas, Washington and
Wyoming.
The cause: a bracket strategy takes the tighter of the federal limit and your state's limit. To find
the state limit the tool looks up which bracket your income falls in, and a state with one flat rate
has no bracket boundary to find. It read that as "there is no room at all" rather than "this state
imposes no limit", and because it then took the tighter of the two, the missing state limit wiped out
the federal limit as well.
Measured on a single filer with a $600,000 IRA and a $60,000 spending goal, Fill Bracket at 22%:
| state | converted before | converted now |
|---|---|---|
| Nevada, Texas, Florida | $465 | $163,686 |
| Illinois | $6,740 | $163,686 |
| Pennsylvania | $4,047 | $163,686 |
| Arizona | $775 | $151,101 |
| California, New York | $131,832 / $141,348 | unchanged |
Two further problems came from the same cause:
- **Spending was reported wrongly in those 21 states.** For any strategy that is not bracket based,
the plan's yearly spending target was set to zero. The Summary then reported the plan as successful
while it funded no spending at all, and total spending could come out negative.
- **"Below IRMAA" converted little or nothing in every state, not just those 21.** That strategy
stops at the first IRMAA surcharge tier, and the same missing boundary applied whenever your
spending goal sat below that tier, which is most plans. The same single filer above now converts
$13,698 in California and $163,686 in Nevada, against $0 and $465 before.
**What this means for a saved plan.** If your state is one of the 21, or if you use the "Below IRMAA" bracket,
your saved scenarios and shared links will produce different numbers than before. In every
case measured the new numbers are the correct ones and the old ones understated what the strategy
would do. The 17 states the tool models with graduated brackets are unaffected for every strategy
except *Below IRMAA*, and that was verified state by state rather than assumed.
---
## 11.13d0
**A link labeled with a file name now shows the page it actually opens.**
The previous release pointed the Change Log's "Details" links, and the link to the full change log
itself, at the formatted page the site publishes. It changed where those links went but not what they
said, so the Documentation tab read "optimizer_changelog.md" while the link underneath opened
`optimizer_changelog.html`. Naming one file and opening another is confusing in both directions: it
looks like the fix never shipped, and anyone who copied the visible name got a file that downloads
instead of displaying.
The label now follows the link. This applies only where the visible text *is* the file name.
Descriptive wording, "Details" or the theme's "Improve this page", is never rewritten, and neither is
the name README.md, which is still the honest name of what sits at the site root.
Nothing changes when you run the tool from your own disk or your own local web server. There the
formatted pages do not exist, so both the link and its label stay on the plain file, which is what is
really there.
---
## 11.13c5
**The "Details" links in the Change Log now open a properly formatted page instead of downloading a text file.**
Every release write-up lives in a Markdown file, and clicking through to one used to hand your
browser a raw `.md` file. Most browsers download that rather than displaying it, and a link to a
particular release inside it did nothing at all. So the links went to a file you had to open by hand,
in whatever program claims `.md` on your computer.
The site already publishes a formatted version of each of those files; the links just were not
pointing at it. They are now, and the link to a specific release lands on that release. Running the
tool from your own disk or your own local web server still opens the plain file, because the
formatted version only exists on the live site.
Two side fixes came with it. The three links on the front page to the planning notes, under "What is
coming next", were broken and now go to where those notes can actually be read. And the architecture
document's diagrams, which were showing as a wall of text, are now labeled as diagram source with a
link to a page that draws them.
---
## 11.13c3
**The Tax Payment Planner now receives your brokerage position, so it stops guessing at it.**
Clicking a year in Annual Details of the Retirement Optimizer opens the Tax Payment Planner with that year's figures. Three of
them were never being sent, and the planner quietly fell back to its own built-in assumptions
instead:
- **Brokerage value and cost basis.** The planner uses the unrealized-gain share of your brokerage
account to price the option of raising the tax money by selling shares. With nothing handed over
it assumed 40% of the account was gain, no matter what basis you had entered here. On the default
scenario the real figure is 60.4%, which nearly doubles the capital-gains cost of that option.
- **The blended long-term capital gains rate**, federal plus state, which had been fixed at 20%.
- **The high-income flag.** Above $150,000 of prior-year income the estimated-tax safe harbor rises
from 100% to 110% of the prior year's tax. The planner had no way to know, so it always used 100%.
Nothing the Optimizer computes changes; this only affects what the Tax Payment Planner is told when
you open it. On the planner side, the single "Brokerage Appreciation (%)" field has been replaced by
**Brokerage Value ($)** and **Brokerage Basis ($)**, with the gain share shown as a computed
readout. That field held a fraction of account value, not a growth rate, but it sat among four
genuine rates and was widely read as one. Shared links using the old field still work.
---
## 11.13bd
**Two controls are now available to everyone: "Stop conversions after (year or age)" in the sidebar and "Fed Tax Creep %/yr" in Assumptions.**
Neither does anything until you fill it in, so no existing
plan, saved scenario or shared link produces a different number because of this release.
**Stop conversions after (year or age)** sits just under Extra Annual Roth Conversion. Enter a
calendar year (2031) or your age (71) and that is the last year conversions run; they cease the year
after. Blank means convert for the whole plan, which is what the tool has always done. The dropdown
beside it chooses the scope: "all conversions" stops your strategy's own bracket filling as well as
the Extra Annual Roth Conversion, while "extra only" stops the extra conversion and lets the strategy
keep filling brackets.
This is the field the Break Even ⓘ has been filling in all along. When the diagnostic finds a year
worth stopping at, it offers a "Stop after YYYY" link, and clicking it writes the year here and
re-runs. Converting past the point where the tax paid stops earning its keep can leave you worse off
than never converting at all, so the year that maximizes after-tax wealth is worth knowing.
**Fed Tax Creep %/yr**, with **Creep Starts** beside it, sits in Assumptions under the growth row. It
multiplies the federal ordinary-income bracket rates by a little more each year from the start year
onward, so a 1% creep turns a 22% bracket into a 26.8% bracket in 20 years. Bracket thresholds still
track CPI as before, and state tax, capital gains, the NIIT surtax and IRMAA are unaffected. Why is tax creep here at all?
Many think a higher-tax future is likely, so this is a simple way to model increasing taxes. We are not forecasting that
tax rate increases will come - after all, tax law changes at the speed of Congress and that's to say... slowly.
Leave tax creep % at 0 to keep today's rates for the whole plan. There is no model for State tax increases, and no plan to
add them, so if you think your state taxes will be creeping up, you can make the creep rate a little higher to compensate.
---
## 11.13a1 (behavior change)
**Social Security is now counted from your birth month, both when you claim it and when a spouse dies.**
Until now the tool treated benefits as an all-or-nothing yearly amount. Someone claiming at 70
collected a full twelve months in the year they turned 70, whichever month their birthday fell in.
The claim year is now paid from the birth month onward: a June birthday collects six months, a
January birthday eleven. The birth month defaults to December, which means no benefit at all in the
claim year and one more year before the money starts. If that is not your birthday, set it under
Profile. It was already used for the age 70 and a half rule on charitable distributions, and it now
matters for Social Security too.
The year a spouse dies works the same way, with the death treated as falling in the deceased
spouse's birth month. The months before it pay both spouses' own benefits, because both were
genuinely collecting; the months after pay the survivor benefit. Previously that whole year paid only
the survivor benefit, which understated it, since the survivor benefit is the higher of the two
benefits and never their sum.
This moves the numbers in almost every plan, and not by a fixed amount. Less income early on changes
how much room a Roth conversion has, how much has to come out of the portfolio, and how much of your
benefit is taxable, and the Medicare surcharge look-back brings the difference back two years later.
**Survivor benefits now use the deceased spouse's real Full Retirement Age instead of assuming 67 for everyone.**
Full Retirement Age is 66 for anyone born between 1943 and 1954, rises two months per birth year
through 1959, and reaches 67 only for those born in 1960 or later. The tool assumed 67 for everybody.
Because a survivor benefit is worked back from what the deceased was collecting, that assumption made
an early claim look like a bigger reduction than it was, and paid the survivor more than they are
actually due.
On a couple born in 1950 and 1952 where the higher earner claimed at 62, the survivor benefit drops
from about $51,076 to $49,148 in its first year. Plans where both people were born in 1960 or later
are unchanged to the dollar, and so is any plan where both spouses claim at or after their Full
Retirement Age, because a late claimer's survivor baseline is their actual benefit rather than a
figure derived from it.
**The chart marks when Social Security actually starts.**
New markers show the first year each person receives a benefit, and the year a survivor benefit
begins. With a December birth month that marker sits on the year after the claiming age, which is
when the money genuinely arrives. Someone already collecting when the plan starts gets no marker,
since the tool cannot tell that apart from a benefit beginning in the first year.
**The Monte Carlo tab no longer shows you the previous plan without saying so.**
Changing a setting while the Monte Carlo tab was open used to leave every chart, the survival table
and the stress test quietly describing the plan you had before the change, until you clicked away to
another tab and back.
The stress test now re-runs by itself, because it is only about ten simulations. The main chart and
survival table are a different matter: they are hundreds of simulations across every strategy, which
takes long enough that recomputing them on every edit would leave you waiting. Those are marked "Out
of date" with a Re-run button instead.
**A Stress Test result now appears in the summary bar at the top of every tab.**
It reads, for example, "7 of 10 fail", meaning your plan runs out of money in 7 of the 10 harshest
return periods in the historical record. It is colored green, amber or red on the same scale as the
Monte Carlo survival table, and the tooltip names the median year the money runs out.
These sequences are deliberately the worst on record, so this is a durability test and not a
forecast: failing some of them does not mean the plan is likely to fail. Because the stress pass is
cheap, this number recomputes whenever you change an input, which makes it the one Monte Carlo figure
that is always current. On the Monte Carlo tab the stress test now appears first, above the
percentile chart, with a headline count and a fold-away control for readers who only want the main
projection. In Synthetic mode there is no historical record to stress against, so the tile reads a
dash.
**The stress test is always available, and the Optimizer says when it is working.**
The stress test used to be skipped entirely in Synthetic mode. It never needed the main projection
to be Historical: it builds its own set of worst-case sequences from the real record, so it now runs
in both modes. Choosing synthetic returns for the projection is not a reason to hide the question of
whether the plan would have survived the worst of what actually happened.
The Optimizer blocks the page for a second or more while it tests every strategy, which looked like
a freeze with the previous run still on screen. It now shows a "Calculating strategies" banner
first. The comparison feature also gained a permanent one-line explanation above the table, and the
button to end a comparison now says so plainly.
**Any two strategies in the Optimizer table can now be compared directly against each other.**
The two Δ columns have always measured every strategy against the anchor ⚓ baseline, which answers
"is this better than doing nothing clever" but not "is this better than that one". Every row now
carries a ⚖ button. Click it and both Δ columns are re-measured against that row, so the numbers
beside every other strategy tell you directly how it compares with the one you picked.
The chosen row reads zero by definition, a note above the table names what the columns are measured
against, and the column headings gain "vs ⚖" so the change in meaning is never silent. Click the same
⚖ again, or use the button in that note, to go back to the baseline. The choice survives a re-run:
change a setting and the tool re-finds the same strategy in the new results rather than losing your
place. If a change removes that strategy from the table entirely, the comparison reverts to the
baseline. Clicking ⚖ does not load the strategy into the sidebar; clicking anywhere else on the row
still does.
Your current plan is now marked (📍) in exactly one place when there is nothing else to say. If another 📍
appears it indicates a slightly DIFFERENT variation from your plan.
The pinned rows read ⚓ BASELINE and 📍 CURRENT, marker first in all cases, and a row that wins a
metric in the Best table keeps its marker there too, so it is recognisable as the same row without
repeating the words.
The ⚖ control has its own column at the start of each row, and the outcome marker beside it is part
of the same control, with a gap separating both from the rest of the row. The highlighted ⚖ always marks
whichever row the columns are currently measured against, which is the ⚓ baseline until you pick
something else, so the table shows where the comparison point is from the moment it opens. Clicking
the highlighted one puts everything back, exactly like the Stop comparing button.
---
## 11.1387 (behavior change)
**Your own plan now appears in the Optimizer table, so you can compare it against every strategy it suggests.**
A pinned 📍 CURRENT row sits under the ⚓ baseline showing your sidebar settings run exactly as they are, including whether conversions are on, any Extra Annual Roth Conversion, and any stop year. That matters because every strategy the Optimizer sweeps is run with conversions turned on and your own conversion settings set aside, so until now nothing in the table was actually your plan, even when the strategy and its parameter matched. Your plan is also ranked with the others, so the Rank column tells you where it stands, and it can win a metric in the Best table if it genuinely beats everything else. The same 📍 marks the swept row that uses your strategy and parameter, which lets you see where your setting sits on that family's curve. If your parameter falls between the standard steps, for example a 7% proportional boost or an 18 year drawdown, that exact value is now swept as its own row instead of being skipped. The Break Even column is filled in for every strategy now, not just the ⇌ conversion rows, so sorting by it works and there is a new "⏱ Earliest Break Even" entry in the Best table: the strategy whose conversions pay for themselves soonest, with ties going to the one that leaves more after-tax wealth. Choosing "Earliest Break Even" under Optimize for now reorders the table, which it previously could not do because the column had no values to sort. Three fixes came out of the same work. Clicking an Ordered row now brings its account sequence with it instead of leaving whatever was in the sidebar. The Monte Carlo stress test now recognizes Guyton-Klinger and Ordered plans as your current strategy rather than falling back to a generic one. And the summary numbers at the top of the page, including Break Even, now refresh when you change a setting while the Optimizer tab is open: they used to keep showing the previous plan until you clicked over to Chart or Annual Details, which is what made a change like turning Maximize Conversions on look like it had no effect.
---
## 11.137f (behavior change)
**The suggested year to stop converting is now measured against the plan you would actually get.**
The tool decides whether to take money out of your accounts in January or in December based on whether that year makes a Roth conversion, because a conversion year is better done early. When the Break Even ⓘ searched for the best year to stop converting, it built each trial plan in a way that failed to set that switch, so every year it compared was run with December withdrawals even where conversions were happening. It also treated "convert nothing" as a January-withdrawal year, which no plan with no conversions would ever be. The result was a suggested stop year and dollar gain measured against plans that were not quite the ones you could load, and the two "stop scope" choices could report different gains for the very same cutoff. On a large-IRA example the gain compared with never converting was overstated by about $8,900, and the two scopes disagreed by about $2,200 on a cutoff where they should have matched. All of that now lines up: click the suggested year and the plan you get is the plan that was scored. A stop year set at or before the first year of the plan is also now correctly treated as converting nothing. Plans with no stop year, including the whole strategy table, Monte Carlo and Annual Details, are unchanged to the dollar.
---
## 11.1370 (behavior change)
**When converting more does not help, the Optimizer now tells you what would have to be true for it to help, and it can find conversions it used to miss.**
Optimize Conversions previously reported only that none of the strategies it examined improved by converting more, which is accurate but a dead end. It now names the future tax rate your plan is assuming, and a link works out the lowest future tax rate at which converting would start to pay, along with the amount and the gain at that rate. If no plausible rate makes conversions worthwhile, it says so plainly, because that is a real answer about your plan rather than a missing one. Separately, the search itself was too narrow: it only ever tested converting the same amount every year for the rest of the plan, so a plan that should convert heavily for a few years and then stop could only be told to convert nothing. When a flat conversion does not help, the Optimizer now also tries converting for a limited number of years, and any plan it finds this way appears as a ⇌ row tagged with the year conversions stop (⏹). Clicking it fills in both the amount and the stop year. This changes which conversion rows appear and the amounts they suggest. Two columns also changed: the Avg BETR column and its summary-bar tile were removed, because testing showed the Break-Even Tax Rate is not reliable enough to decide anything (it is still available per year in Annual Details), and "Conv Savings" is now called "Tax Paid Δ", since it only counts tax paid during the plan and can look positive on a plan that ends up worse off. Break Even remains the column to trust.
---
## 11.1340 (behavior change)
**Cash Reserve now controls what happens to money you withdraw beyond your spending needs.**
Most retirees with a large IRA are forced to take out more than they spend once required distributions begin. Until now the tool left all of that surplus sitting in cash, earning only the cash yield, for the rest of the plan. That understated how well a plan does when those extra dollars are reinvested, and it quietly tilted the Roth conversion comparison in favor of converting. The Cash Reserve field now decides where the surplus goes. Leave it blank (or enter -1) to keep the original behavior: all surplus stays in cash. Enter 0 to reinvest every surplus dollar into your Brokerage account, where it grows at the market rate. Enter a dollar amount to keep that much as a cash cushion, in today's dollars, and reinvest only the surplus above it; spending will not draw the cushion down unless every other account is exhausted. If you load a saved plan or a shared link that already sets a Cash Reserve, a note reminds you that results now differ from older versions and that blank or -1 restores the old behavior. This one setting can meaningfully change Break Even, the suggested year to stop converting, and the final balances, so it is worth trying both ways.
---
## 11.1330
**The Break Even note now tells you the best year to stop converting, and offers one click to do it.**
When your plan makes Roth conversions, the ⓘ next to Break Even now names the year that leaves you with the most after-tax wealth if you stop converting after it, along with how much more that keeps than converting all the way through, and how much more than never converting at all. This replaces the older note, which pointed at the year conversions first turned unprofitable; that year is usually not the best place to stop, because it ignores how much a plan can still gain by stopping earlier. The point is that converting past a certain year can leave you worse off than never converting, since the tax paid on a late conversion no longer earns its keep before the plan ends. Click the suggested year to fill in a new "Stop conversions after" control and re-run: enter a calendar year like 2031, or your own age like 71, and conversions run through that point and then stop. You can choose whether stopping applies to all conversions or only the Extra Annual Roth Conversion; stopping all of them is the stronger lever. Spending is unchanged by where you stop, so this is a pure comparison of what your heirs receive after tax.
---
## 11.12fd
**The charts now mark the year each person's RMDs begin.**
Two new dashed lines, "Your RMDs begin" and "Spouse RMDs begin", join the existing markers on both charts, and because the date depends only on a birth year they also appear on the Monte Carlo chart. The line is drawn for the year you reach your RMD age (73 or 75, depending on when you were born), and only when that birthday falls inside the plan, so a plan that starts after your RMDs have already begun does not get a line. It is drawn whether or not there is anything left in the IRA to withdraw, since the date itself is worth seeing. The "Show milestones" tooltip now lists every marker the charts can draw, which it had stopped doing some versions ago.
---
## 11.12fb
**The Withdrawal Rate stat was reading far too low and has been corrected.**
It now measures what it says: the money actually pulled out of your accounts to pay for spending and taxes, divided by what the portfolio was worth at the start of that year. Three things were wrong before. Social Security and pension income were being subtracted from the withdrawals, so a plan drawing $164,000 against $81,000 of Social Security was reported as if it had withdrawn only $19,000. The bottom of the fraction used an after-tax estimate of your wealth rather than the actual account balances, and the very first year used a different basis than every year after it. On a typical plan the stat moves from about 1.1% to about 2.4%, which is now directly comparable to the classic 4% rule and to the guardrail rate the Guyton-Klinger strategy uses internally. The tile is relabeled simply "Withdrawal Rate", and hovering over it shows two extra views: a dollar-weighted average, and a net depletion figure that goes negative when your portfolio is growing faster than you draw it down. The rate itself can no longer go negative, because you cannot withdraw less than nothing. A related fix: in years when Social Security covers everything, a required distribution that gets reinvested is no longer counted as a withdrawal, since those dollars never left the portfolio. Also in the header, the summary stats are left-aligned on wide screens instead of being pushed to the right edge, and the "what you last changed" note moved to sit after the numbers alongside the save and load messages.
---
## 11.12f7
**New "Optimize for" choice at the top of the Optimizer ranks every strategy by the goal you care about.**
Pick a goal and the table sorts to put the best plan for it on top, moves the baseline to match, and renumbers the Rank column. It changes only which plan the tool puts in front of you, never the underlying numbers, and clicking any column header still sorts by that column. The choices, in the order most people asked for them: Tax Flexibility (the default, which favors plans that end with your money spread evenly across pre-tax, Roth, and taxable accounts so you can draw from whichever is cheapest in any given year, compared only among the plans that also finish among the wealthiest); Maximum Net Wealth; Avoiding Widow and RMD Tax (favors plans that leave less in the pre-tax IRA, since that is what drives forced withdrawals and the higher single-filer tax a survivor pays); Minimum Lifetime Taxes; plus Maximum Spending, Maximum Roth, and a Balanced blend. The Documentation tab explains what each one measures, including why Minimum Lifetime Taxes and Avoiding Widow and RMD Tax can legitimately disagree. A few smaller fixes came along with it: the "Best" summary no longer lists a strategy that cannot actually be run; Annual Details shows the Roth Conversion column when you turn on the Opp. Cost view; when a spouse is already on Medicare, all of the ACA income-limit rows are marked not applicable rather than just one; and the Score column was dropped since the row order and the Rank column already show the ranking.
---
## 11.12e5
**Optimize Conversions now looks at the best plan from each family of strategies, not just the five with the highest ending wealth.**
The strategies that benefit most from converting are often not the ones that finish with the most money, so the old five-highest list could be filled entirely by strategies that gain nothing from converting, leaving the ⇌ table empty even when other strategies had a real conversion to recommend. It now considers the best Proportional, Reduce, Fill Bracket, IRMAA Ceiling, IRA Draw, Ordered, and Guyton-Klinger plan (each with and without the cyclic brokerage option), so a strategy that converts well gets a seat. It also judges how much to convert by the same measure it uses to rank everything else - after-tax wealth left over plus the money you actually get to spend, in today's dollars - rather than raw ending wealth, which used to reward simply hoarding in the IRA. Because of these two changes, the recommended conversion amounts and the strategies shown may differ from before on the same plan. When no strategy benefits from converting more, the tool now says so plainly instead of showing an empty table.
---
## 11.129d
**Annual Details now shows Roth conversions as the IRA withdrawals they are, and conversions draw from the larger IRA.**
A Roth conversion is money leaving the IRA, so the IRA WD, IRA1- and IRA2- columns now include it. Before, a year could show a large conversion with little or no IRA withdrawal, which looked wrong even though the balances were correct. RMD stays in its own columns (it is the involuntary draw and is never a conversion). The Federal and State tax columns now include the tax on conversions, so they add up to Total Tax and the Taxation chart shows the real tax. Conversions are now taken from the larger IRA first (spilling to the smaller only when the larger cannot supply the full amount) instead of being split by size, which is what you would actually do and avoids converting a token slice out of a tiny IRA. This changes each IRA's balance over time and therefore each person's future RMDs; the combined yearly tax is unchanged. Clicking a year to open the Tax Payment Planner now sends the correct per-IRA voluntary, conversion, and tax figures.
---
## 11.1287 - cash-funded conversion taxes
**Maximize Conversions now actually maximizes: available Cash can cover conversion taxes.**
Previously nothing in the tool paid a conversion's tax bill from outside funds, so a $20,000 Extra Annual Roth Conversion only landed about $13,700 in Roth - the rest went to the tax on its own withdrawal. Standard practice is to cover that tax from Cash so the full amount gets to grow tax-free, and Maximize Conversions now does exactly that, on top of its original behavior of routing leftover IRA withdrawals into Roth. It uses only the Cash you have: when Cash is short it funds what it can, and at $0 Cash nothing changes, so existing plans are unaffected until you turn it on. Two switches beneath it, "Convert Excess to Roth" and "Use Cash", let you control the two behaviors separately.
---
## 11.1287 - Optimize Spend and Optimize Conversions move to the Optimizer tab
**Optimize Spend and Optimize Conversions moved to the Optimizer tab, and Optimize Conversions is now on by default.**
Both only ever affected the Optimizer, but sat in the sidebar next to the strategy settings that drive every tab, which made them look like plan settings. They now live at the top of the Optimizer tab under "Search options", where they apply. Optimize Conversions starts on so the conversion-optimized ⇌ rows are found without having to know to ask for them; turn it off if you want a faster sweep. Note that ⇌ rows only appear where an extra conversion actually improves the result for one of the top-ranked strategies, so a plan whose best strategies do not benefit from converting more will correctly show none. Your saved scenarios and shared links keep working unchanged.
---
## 11.1287 - smaller fixes
**Smaller fixes.**
The Roth Conv column in Annual Details had silently lost its explanatory tooltip to a naming typo (as had the Roth Growth column); Roth Conv's now spells out why it can read lower than the amount you asked to convert. The Break Even explainer is a compact ⓘ next to the year instead of a standing button, and it no longer makes you click before telling you anything - hover over it to read the reason, click pins it open, click again closes it.
---
## 11.1271
**Three fixes from testing the Extra Conversion field and Break Even.**
(1) The Optimizer's conversion-amount search could recommend a Guyton-Klinger conversion so large that GK could only "afford" it by continuously cutting your future spend behind the scenes -- the search now rejects any amount that isn't actually sustainable for GK, the same stability check already used elsewhere in the optimizer. (2) The "Extra Annual Roth Conversion $" tooltip now says plainly that it is capped only by the remaining IRA balance, not by your IRA Goal -- a conversion moves money IRA-to-Roth rather than out of the household, so it is allowed to draw the IRA below that goal on purpose. (3) When Break Even shows “ - ”, a new ⓘ next to the stat identifies the specific conversion year that erases an otherwise-sustained lead, instead of leaving you to guess why.
---
## 11.1253
**Optimizer strategies loaded from the table now actually match what the table showed.**
Clicking a "⇌ Optimize Conversions" row to load it previously dropped the extra conversion amount that made that row special, so the loaded plan's Break Even (and everything else) could disagree with the Optimizer table. A new "Extra Annual Roth Conversion $" field now holds that amount, is filled in automatically when you load a ⇌ row, and is shareable/saveable like any other input. Loading a cyclic-brokerage or IRMAA/ACA-ceiling ⇌ row now also correctly carries those settings over. Also clarified the Conv Savings tooltip: it only counts tax actually paid so far, so it can look good even when Break Even (which also counts the tax still owed on money left in the IRA) says the conversions never really paid off.
---
## 11.1247
**Optimizer: Break Even now shown for Optimize Conversions strategies.**
When the Optimize Conversions checkbox is on, the top 5 strategy rows now report a Break Even year (same permanent-crossover definition as the single-scenario stat) in a new column, and a new "Earliest Break Even" objective ranks strategies by how soon their conversions permanently pull ahead. Strategies that never sustain a lead show “ - ”.
---
## 11.1240
**Break Even now requires the lead to hold for the rest of the plan.**
A rare scenario could report Break Even off a single year where the converting (or excess-withdrawal) plan brushed even before falling behind again, sometimes for good. Break Even is now the earliest year the plan pulls ahead and stays ahead through the end of the simulation, showing “ - ” if that lead is never sustained. Same fix applies to excessOC.
---
## 11.11ff
Refactor: No behavior changes. If the page looks stale after this update, do a hard refresh.
---
## 11.11f3
Internal cleanup: source files renamed to a shorter, consistent scheme (optimizer_core.js, optimizer_ui.js, optimizer_tests.js, optimizer_text.js, optimizer_styles_responsive.css). The simulation engine and the page UI now live in separate files. No behavior changes. If the page looks broken after this update, do a hard refresh (Ctrl+Shift+R) to clear the old cached files.
---
## 11.11dc
**Break Even and Opp. Cost are now measured with a true no-conversion re-simulation.**
The tool re-runs your whole plan with the Roth conversions removed and compares after-tax wealth year by year. The no-conversion plan keeps the money in the IRA and pays its own larger RMD taxes and IRMAA surcharges later, so the benefit of avoiding those is now fully counted. The old approximation could report a Break Even year when there were no conversions at all, and could fail to report one for conversions that clearly paid off. Break Even now appears only when conversions actually happen. The same rework applies to the excess-withdrawal opportunity cost (excessOC).
---
## 11.11c8
Improved bar charts. Hover over or click legends to isolate that value. Double click to restore. Line legends are unchanged. IRMAA is now presumed to occur in years 1 and 2 (matching the future value) - a bug caused it to only show up on year 2.
Stress is now performed automatically when Historical Monte Carlo is selected and applies to the current strategy. Line charts now also have the "isolate" behavior in addition to the "click to dismiss, click to reshow."
Cycle Brokerage will now fill up the 0% or 15% tax bracket (whichever falls below any current limits). Previously it was only pulling enough to meet spending.
Other minor UI improvements also made.
---
## 11.11ae
Income & Expenses chart: added a note that shown incomes are after-tax (pre-tax figures are in Annual Details).
---
## 11.11ad
Added Spending to Annual Details to focus on spending.
---
## 11.1133
**Architectural improvements.** No features changed.
---
## 11.1125
**IRMAA now respects Medicare age.**
- **Age-65 gate, per spouse.** The IRMAA surcharge is charged only for spouses who are 65+ (actually on Medicare) - a couple aged 61/59 no longer pays IRMAA no matter how large a Roth conversion is. With one spouse 65+ and one younger, only the older spouse's half is charged.
- **Tier display fix.** The IRMAA Tier column (and chart milestone) previously appeared one year before the surcharge was actually charged; both now use the same 2-year MAGI lookback as the dollar amount, and show -none- before age 65.
- **IRMAA Ceil strategies unlocked pre-63.** Before any spouse is 63 (the earliest year income can affect a premium, given the 2-year lookback), the IRMAA-tier conversion ceiling relaxes to the top of the federal bracket containing it - no more pointlessly capped conversions in your early 60s. Same for “Lesser of IRMAA or Bracket.”
- **Medicare Parts B+D base cost** now shown on charts and table for spouses 65+ - illustration only, not deducted from spendable.
- **Distinct chart colors.** IRMAA is now deep pink and Medicare teal on all charts (they previously shared nearly identical pinks); the IRMAA milestone marker matches the pink.
---
## 11.1119
**State retirement-income taxes are now much more accurate.**
16 states (Alabama, Colorado, Connecticut, Georgia, Illinois, Iowa, Kentucky, Maine, Maryland, Michigan, Mississippi, New York, Ohio, Pennsylvania, Virginia, Wisconsin) now apply their real pension/IRA exemptions, caps, and credits instead of taxing all retirement income as ordinary income. Several states have limitations; limitations are listed below the state once it's selected. See, e.g., CA, MA, NC, SC, NE, AZ, and MT.
---
## 11.1102
**State retirement-income exemptions (Illinois & Pennsylvania):**
IRA/401k/pension distributions are no longer taxed by states that exempt them. Illinois and Pennsylvania now correctly exempt these withdrawals (interest, dividends, and capital gains remain state-taxed). Other states are unchanged.
---
## 11.10ee
**Pension start age, optimizer symbol legend, and a Guyton-Klinger fix:**
- **Pension start age.** Model a pension that begins after you retire (e.g. retire at 60 while the pension starts at 65). Leave the start age at 0 to begin at retirement.
- **Optimizer symbol legend.** A new legend explains the symbols shown on strategy names (✓ ✦ ▼ 🗘 🔄 ⇌ ⚠️ 🟢 🚨 ⚓).
- **Guyton-Klinger sustainable-spend fix.** When no strategy can fund your spend goal, the suggested fallback spend no longer reports a Guyton-Klinger value that only “works” by cutting spending every year - and the suggested strategy is now clickable to load.
---
## 11.10cf
Numerous user experience improvements: IRA Draw strategies now tested to 20%, After-Tax-Spend searches higher spend rates and a suggested value is offered. Other changes include tooltip clarity improvements, chart improvements, and showing Charts first by default.
---
## 11.10a2
Added additional charts for clarity: Taxation, Income vs Net Income, Inflows vs Outflows, Earnings vs W/D.
---
## 11.1099
**Baseline strategy is now ranked by total economic value, not terminal wealth alone.**
The pinned ⚓ baseline is the best no-conversion strategy. **Guyton-Klinger Optimize Spend no longer reports an unsustainable spend.** Optimize Spend now applies a stability floor and accepts a higher initial spend only if the guardrails never cut real delivered spending below one guard band of the initial.
---
## 11.1091
Fixed a NaN in a Monte Carlo.
**Fixed a shortfall**. Fill strategies will exceed the threshold when needed to meet the spend goal.
However, the ACA Cliff ceiling is now its own strict strategy: it never breaches the cap. If spending can’t be met under the target the plan is flagged untenable (⚠️ in the Optimizer) rather than silently overspending.
---
## 11.1060
**Stress mode now ranks worst decades by real inflation-adjusted Compound Annual Growth Rate (Real CAGR):**
real CAGR = (1 + equity) / (1 + inflation) − 1. High-inflation decades like the 1960s–70s stagflation, and the Lost Decade (1999-2010) correctly rank as worse for retirees than a pure equity crash with near-zero inflation. Depression-era crashes also had high deflation which is now clamped to a minimum of -1% (rather than -9%).
---
## 11.1042
**Guyton-Klinger Guardrails withdrawal strategy:**
- **Dynamic spending rules.** The Guyton-Klinger strategy adjusts annual spending based on portfolio withdrawal rate. It skips inflation adjustments when the prior year return was negative, and cuts spending 10% when the withdrawal rate exceeds the upper guardrail; raises spending 10% when withdrawal rate falls below the lower guardrail. Supports higher initial withdrawal rates (~5–5.5%) by absorbing sequence-of-returns risk through spending flexibility.
- **Four configurable parameters:** Upper/lower guardrail (default ±20% of IWR) and cut/raise percent (default 10%). Annual Details Income category shows gkSpend and gkAdj columns.
---
## 11.1001
**Best "Baseline" is used for strategy comparison:**
- **Honest baseline.** The optimizer now also runs every strategy family with no Roth conversions and no cyclic brokerage maneuvering, no QCDs, and pins the strongest of those as a ⚓ BASELINE reference row at the top of the results. Every other strategy is measured against it - because "strategy A beats strategy B" only means something relative to the best you can do without conversion/brokerage tricks.
---
## 11.1048
**Shorter share URLs:**
Share URLs are now up to 70% shorter; older, longer URLs still load unchanged.
---
## 11.1019
**Cyclic-brokerage drawdown fixed.**
---
## 11.1018
**Δ columns honor Future/Current $; tax tooltips clarified.**
- **Cleaner table.** Infeasible (bracket-unreachable) rows are hidden by default - click the Infeasible legend to reveal or hide them. Failed plans always sort below successful ones. Every column header now has an explanatory tooltip.
---
## 11.fed
**Charitable Giving (QCDs) & UX Improvements:**
- **Qualified Charitable Distributions (QCDs):** New Charitable Giving (QCD) section in the sidebar. Enter an annual household QCD maximum - the simulation transfers that amount directly from the larger eligible IRA to charity each year (age 70½+ per person). QCDs satisfy the RMD requirement without adding to taxable income, reducing IRMAA exposure.
Two modes: Always donates the full amount every eligible year; As Needed applies only enough to drop two IRMAA tiers (or escape the surcharge entirely, whichever requires fewer dollars). The per-person limit is $111,000 for 2026 (CPI-indexed annually per SECURE 2.0). QCDs appear as a gray bar in the Income & Expenses chart.
- **Dollar display toggle:** The Future $ / Current $ toggle has moved to the left of the tab bar (always visible). Future $ shows nominal amounts as they will actually appear; Current $ restates everything in today's purchasing power for easy year-to-year comparison.
- **Strategy panel switches:** Maximize Conversions (formerly "Max Conversion"), Cycle Brokerage (formerly "Cyclic"), Optimize Spend, and Optimize Conversions (formerly "Conv Optimizer") are now all in the strategy sidebar with labels to the right of each switch. Toggle labels are positioned so that the green (on) position visually aligns with the label.
---
## 11.ecc
**Withdrawal Rate:**
Corrected Portfolio Withdrawal percent calculations. Added netOutflow, grossOutflow and inFlow columns for summary information.
---
## 11.ecb
**Withdrawal Timing:**
Each simulation year now auto-selects Early (January) or Late (December) withdrawal timing. Conversion years use Early - maximizing Roth compounding duration. Spending-only years use Late - the full portfolio compounds before the withdrawal exits, gaining D×r per year (~$3,500/yr on a $50k draw at 7%). No manual toggle; the algorithm tracks prior-year conversion activity. A new Timing column in Annual Details shows Early(Conv) or Late(Spend) for each year.
---
## 11.eca
**Bear-Start Overlay (Historical Monte Carlo):**
In Historical mode, the bottom 25% of paths begin with a randomly-sampled worst-decade historical sequence before continuing with random draws. The likelihood of a "Bear start" market is 23%, so this adds realistic sequence of return risks (SORR).
---
## 11.ec9
**Stress Mode:**
Monte Carlo "Stress (worst sequences)" runs the worst historical retirement-start sequences scored by first-decade real (inflation-adjusted) CAGR. Shows sequence-of-returns risk directly - each line is a real historical scenario. Chart labels show nominal equity CAGR, inflation CAGR, and real CAGR for each scenario. Clicking on a legend hides all other plots. Will your retirement withdrawal strategy survive the stagflation of the 1960s–70s? The tech bubble of 2000? This will tell you.
---
## 11.ec8
**State Bracket Inflation Fix:**
MT, ND, AL, OH, and SC have statutory fixed income tax brackets (they were being incorrectly inflated which would understate future state taxation). Other states are unaffected. Known limitation: the standard deduction for AL, OH, and MT is fixed by statute but is being adjusted - future fix to be provided.
---
## 11.ec7
**State Tax Expansion:**
Added AL, AZ, CO, IN, KY, MA, ME, MN, MT, ND, OH, SC, WI to the state income tax dropdown. All states adjusted to latest available information.
---
## 11.ec6
**Cycle Brokerage:**
New Cycle Brokerage toggle in the strategy box enables alternating IRA draw years and brokerage LTCG harvest years.
IRA-draw years and brokerage harvest years alternate. The ratio of IRA to brokerage cycles is based on the ratio of the account balances. E.g. if the brokerage is 1/3 the size of the IRA total, the cycle will be 1 year brokerage, 2 years IRA draws.
As the asset balances change, the ratio may also change.
In Brokerage harvest years spending is funded from brokerage gains at capital-gains rates (often 0–15%). Excess brokerage funds are reinvested (DRIP) so dividends stay inside brokerage and don't
leak to Cash as ordinary income. Two orderings are supported: IRA-first (🗘, red) harvests after N IRA years; Brokerage-first (🔄) harvests at the start of each cycle.
The Optimizer and Monte Carlo always run all three variants (baseline + IRA-first + Brokerage-first) and surface the best. The subCycle column in Annual Details shows IRA/Brok/⚠Brok per year.
One benefit of Cycle Brokerage is earlier reduction of brokerage assets that would otherwise grow increasing later capital gains.
---
## 11.e64
**BETR (Break-Even Tax Rate):**
Annual Details now shows the break-even future tax rate for each year's Roth conversion.
Avg BETR summary stat shows the average across all conversion years. "Conv Optimizer" determines whether conversions realize any tax savings and are indicated with (⇌)
**Bug fix:** eliminated shortfalls in cases involving brokerage withdrawals.
---
## 11.e52
URL compression: all share URLs now use short parameter names (57% shorter). Share button opens a copyable popup panel on all tools matching Income Tax Planner style. Income Tax Planner gets an "Open in Tax Planner →" button that pre-fills Retirement Tax Planner with state, SS income, and capital gains. Backward compatible - existing long-key bookmarks still load.
---
## 11.e4f
User Experience polish throughout. Roth conversion & opportunity cost tracking calculates when conversions or cash withdrawals "pay off". Break Even at the top says when that year is. New Opp. Cost category and columns added to Annual Details. Future IRA Tax % marginal rate when IRA is eventually distributed; auto defaults the last simulation year marginal rate. Choose your heirs total tax rate.
---
## 11.e4a
Fixed start age bug (table now starts at retirement year on URL load). Updated withdrawal strategy descriptions in How To. Added color legend to Annual Details table.
---
## 11.e1a
New feature: view input distributions in Monte Carlo.
---
## 11.df0
Monte Carlo uses historical inflation 1928–2024 and growth (based on S&P), that historical inflation also is used to escalate the spend-goal rather than using a fixed rate. Stats display CAGR (Compound Annual Growth Rate) instead of arithmetic median - e.g. the CAGR growth rate is the "true" rate where the median rate is the "middle" rate over the entire period. Example: -20% growth for 3 years, 10% growth for one year, then +20% growth for 3 years the CAGR is about -0.38%, but the median is 10%. CAGR gives you the real net growth.
---
## 11.dd9
Monte Carlo uses Historical data from real S&P 500, bond, and international returns (1970–2024). Account Composition shows Est.Rtn advisory column (median before inflation); Monte Carlo shows per-asset-class return ranges (Eq/Bonds/Intl).
---
## 11.dd6
Added ordered withdrawal strategy (CBIR/RIBC/BIRC). Fixed error where Spend Goal was inflated prematurely. Default bracket strategy changed to `Below IRMAA`. Fill provides feedback: ✓ or ⚠ status with max feasible spend estimate; clicking ⚠ sets After-Tax Spend to the bracket maximum and triggers recalculation.
---
## 11.dad
Ages (current age and RMD start age, live-updating) added; optimizer always runs with Max Conversion on (it was almost always better); Strategy defaults were changed; other defaults changed, too.
---
## 11.da6
Bug fixes: survivor SS benefit was ~2× correct (delayed credits were accruing past claiming age - fixed); dividends now correctly compound inside IRA and Roth accounts (previously only Brokerage received the dividend growth component). New feature: Retirement Start Age - enter the age you plan to retire; the simulation inflates brackets, SS COLA, and spend goal forward to that year automatically. Start age defaults to your current real-world age (birth-month aware). You may have retired already, but this tool starts where you are. Monte Carlo: click any result row to load that strategy. Annual Details: Roth1/Roth2 per-person balance columns now available under Roth Δ. Minor Version numbering changed: lowercase hex((day-of-year × 24) + hour).