[TR-04] Practice Retirement

A six-month trial costs 45% of one marginal working year and resolves a band 4.6 times its own size — and run naively it hands you exactly the flattering number it was built to catch

Pillar: Approaching Retirement & Transitions · Applies to: Anyone whose retirement date is being argued rather than decided, and anyone about to leave on a spending number they have estimated rather than measured Last verified: August 2026 · Refresh cadence: Evergreen (the arithmetic and the statutes); figures via FN-02 Related: TR-01 Five-Year Runway · TR-03 One More Year · ER-01 Safe Withdrawal Rates · ER-06 Sequence of Returns Risk · ER-07 Coast and Barista FIRE · HC-01 ACA Bridge · HC-05 Non-ACA Alternatives · TX-07 Guardrails

Not advice. Whether you can take a leave at all is your employer’s decision, not a right — the one federal statute that guarantees your health coverage during unpaid leave does not cover this purpose, and §6 is the reason to get the answer in writing before you plan around it. The arithmetic here is exact; the spending number it measures is the only one you will ever have that isn’t a guess, and it is still only six months long.

---

TL;DR

• A six-month trial costs about 45% of one marginal working year — $3,176 of annual sustainable spending for the reference household, against $7,062 for a full extra year of work (TR-03 §1). Half a year of time, under half a year’s value.
• It resolves a band 4.6x its own size. A ±10% error in the spending estimate moves the funding ratio 18.2 percentage points — from 82.0% to 100.2%. The trial costs 4.0 of those points.
• The two purchases are different goods. An extra year buys margin; a trial buys knowledge. Sizing margin before measuring what it protects is the wrong order, and it is the order almost everyone uses.
• Run naively, a six-month trial understates annual spending by 15% — it captures the smooth $68,000 and misses the lumpy $12,000. The instrument built to remove TR-03 §5’s flattering number will manufacture it instead unless corrected.
• FMLA does not cover this. It guarantees coverage continuation for birth, adoption, family care, your own serious health condition, or a military exigency ✅ — not for testing a retirement. An unpaid personal leave can be a COBRA qualifying event while you are still employed ✅.

---

1. What a trial costs

The reference household is TR-03’s, deliberately, so the two numbers are denominated the same way: both 58, portfolio $1,600,000, spending $80,000, saving $50,000, planning to 95. Growth is a stated 5% real; sustainable spending is ER-01 §3’s benchmark — amortised at a guaranteed 3% real over the remaining horizon.

Model the leave year honestly: half a year of saving still happens ($25,000), and half a year of spending comes out of the portfolio ($40,000) instead of out of salary.

Path	Portfolio at 60	Sustainable	Funded
Worked straight through	$1,866,500	$86,866	108.6%
Took the trial, coverage maintained	$1,798,250	$83,689	104.6%
Took the trial, leave triggered COBRA	$1,784,434	$83,046	103.8%

The trial opens a $68,250 portfolio gap — $3,176 of annual sustainable spending, or 4.0 points of funding ratio.

Against TR-03 §1’s marginal year at $7,062, that is 45%: half a year of time for under half a year’s value. The trial is not free, and it is much cheaper than the thing people do instead.

The COBRA line is the surprise, and it runs the reassuring way. Six months of premiums at HC-05 §3’s stated family rate is $13,158 — a serious number against a salary and a small one against $1.8M. It raises the trial’s cost from $3,176 to $3,819, about a fifth, not a multiple. The healthcare exposure in a trial is a structural problem (§6), not a financial one.

2. What it buys

Sustainable spending at 58 is $72,179 and does not move — it is a fact about the portfolio and the horizon. The funding ratio moves a great deal, because its denominator is an estimate.

If true annual spending is…	Funding ratio at 58
$72,000	100.2%
$76,000	95.0%
$80,000 (the estimate)	90.2%
$84,000	85.9%
$88,000	82.0%

A ±10% error spans 18.2 percentage points. That is larger than two years of work adds, and it is invisible, because the spreadsheet reports the ratio to one decimal place regardless of whether its denominator was measured or guessed.

So: 4.0 points to resolve 18.2 points — 4.6x.

One honest qualification, and it matters. These are not the same good. The extra working year buys margin; the trial buys knowledge. You cannot straightforwardly trade one for the other, and the trial does not eliminate the band — it collapses it toward a measurement, over six months, subject to §4.

But the ordering claim survives the qualification, and the ordering claim is the point. Margin is sized against a spending number. If that number is wrong by 10%, the margin is being calculated against the wrong target, and more of it does not fix the error — it disguises it. Measure first, then size. Almost everyone does it backwards, and TR-03 §1 explains why: the extra year always looks productive, and the trial always looks like an interruption.

3. Against one more year

TR-03’s household at 90.2% funded has two ways to spend the next twelve months.

	Costs	Buys
One more year of work	12 months	+8.8 points of funding ratio (90.2% → 99.1%)
A six-month trial	6 months, 4.0 points	Whether the 90.2% was ever the right number

These are not competitors and the table is not a scoreboard — the honest answer for most households at this point is both, in that order. What the table shows is that the trial is the cheaper of the two by time and the more informative by a wide margin, and it is the one that gets deferred.

The reason it gets deferred is structural, and TR-03 §2 already named it. An extra year of work requires no decision — it is the default, it happens by not acting, and its cost never appears on a statement. A trial requires asking an employer for something, explaining why, and accepting a visible, quantified hit to the balance. One of these feels free and is not; the other feels expensive and is 45% of a year.

Where the trial is decisive rather than merely useful: when the funding ratio sits between roughly 90% and 110%. Below that the answer is not in doubt and a trial is an expensive way to confirm it; above it, the margin absorbs a spending error and the question is no longer financial (TR-03 scenario B). In the band, the spending error is larger than the gap being argued about, which is exactly when measuring beats estimating.

4. The trial that lies to you

A six-month trial has a systematic bias, it runs in the flattering direction, and it is large.

Split the reference household’s $80,000: about $68,000 is smooth — housing, food, utilities, insurance, routine medical — and about $12,000 is lumpy: the roof reserve, the car reserve, travel, irregular medical, the family events that arrive without a schedule.

A six-month window captures $34,000 of the smooth half and, typically, none of the lumpy items. Double it and you get $68,000 — an annual estimate 15% below the truth.

So the instrument built to remove TR-03 §5’s flattering-number error manufactures that exact error instead. Worse, it manufactures it with the authority of a measurement, which is harder to argue with than an estimate.

The correction is mechanical, not subtle:

• Add the lumpy items back from records, do not try to measure them. Five years of bank and card statements will show the roof, the car, the vet, the wedding. Six months will not.
• Reserve for them monthly during the trial so the cash-flow experience is honest too — a trial that feels comfortable because nothing broke has tested less than it appears.
• Count the trial’s own artefacts out. People spend differently when they know it is temporary, in both directions: some economise for the experiment, some treat it as a holiday.
• Twelve months beats six for exactly this reason, where it is available. The lumpy items are annual, so an annual window is the first one that samples them.

5. What a trial cannot test

The honest list, because a trial oversold is worse than a trial not taken.

Question	Does a trial answer it?
What do I actually spend?	Yes — this is the one it is for, subject to §4
Can I fill the time?	Partly — and six months of novelty is not thirty years of structure
Sequence-of-returns risk	No. Six months is not a sequence (ER-06)
The healthcare cost	Usually no — you are typically still on employer coverage (§6), so the trial omits the largest new line in an early-retirement budget (HC-01)
Longevity, inflation over decades	No
Irreversibility	No, structurally — you know you are going back, which is the variable that makes retirement different

Read the fourth row twice. A trial that leaves you on employer coverage measures a household budget that is missing its bridge premium. For a household under the 400% FPL cliff that omission can be five figures a year (HC-01); over the cliff it can be much more. Add the modelled bridge cost to the trial’s result before comparing it to anything — the trial measures your spending, not your retirement’s spending.

And the last row is the one that limits the whole exercise. TR-03 §4 showed the retirement decision is largely reversible; a trial is entirely reversible, and that changes the experience of it. A trial can tell you what things cost. It cannot tell you how you will feel about a door that has closed, because during the trial it hasn’t.

6. The mechanics: leave, coverage, and what the statute does not do

This is where trials fail in practice, and the failure is nearly always about health coverage.

FMLA guarantees coverage continuation — for reasons that do not include this one. The Act provides eligible employees 12 workweeks of unpaid, job-protected leave and requires the employer to maintain group health coverage on the same terms as if the employee were working ✅. Eligibility is 12 months of employment, 1,250 hours in the prior 12 months, and a worksite with 50+ employees within 75 miles ✅. The qualifying reasons are birth, adoption or foster placement; care for a spouse, child or parent with a serious health condition; the employee’s own serious health condition; and qualifying military exigencies ✅.

“I would like to test my retirement” is not on that list, and the protection people vaguely assume covers a sabbatical does not reach it ✅. This is the single most important sentence on the page for anyone planning around a leave.

So a practice-retirement leave is discretionary, and its coverage terms are whatever the plan says ◻️. Which raises the trap:

An unpaid personal leave can be a COBRA qualifying event while you are still employed. COBRA’s triggering event for a covered employee is termination or reduction of hours ✅ — and a leave that drops your hours below the plan’s eligibility threshold is a reduction of hours, even though you remain on the payroll and still have a job to return to. Where it applies, the continuation period is 18 months ✅.

Three consequences worth planning for:

1. Ask in writing whether the leave maintains active coverage or triggers COBRA, before agreeing to anything. Two employers, identical leave, opposite answers.
2. If it triggers COBRA, TR-01 §5’s 60-day fork applies in full — the election is retroactive with 45 days to pay, so the window is free optionality rather than a decision ✅. Do not elect on day three.
3. The marketplace is a live alternative during a leave, and for a household whose MAGI is briefly low it may be much cheaper than COBRA (HC-01, HC-05 §3). A trial is one of the few moments where a low-income year and a coverage gap coincide, which is the profile the subsidy schedule rewards.

Beyond coverage, three things a leave can quietly cost: vesting and pension service credit, employer match true-ups, and bonus eligibility ◻️ — all plan-specific, all worth confirming alongside the coverage question, and all covered by TR-01 §4’s vesting-calendar work.

7. Designing a trial that answers something

• Write the question down first. “Is my $80,000 right?” is a trial. “Will I like retirement?” is not — nothing in §5 answers it in six months.
• Twelve months if you can get it, six if you cannot, for the §4 reason. Below three months you are measuring a holiday.
• Reserve the lumpy items monthly and add them back from records at the end.
• Add the modelled bridge premium to the result, since the trial almost certainly omits it (§5).
• Do not restructure the portfolio for it. A trial is a spending experiment; changing the allocation at the same time confounds it and imports sequence risk you were not testing (ER-06).
• Set the decision rule before you start — the TR-03 §6 threshold, updated with whatever the trial finds. A trial with no pre-committed rule produces a story rather than a number.

Three Scenarios

A. 58, 90.2% funded, arguing about it for the third year. The gap under discussion is nine points; §2 says his spending estimate carries eighteen. He is negotiating a number he has never measured. A six-month leave costs 4.0 points and tells him whether the argument was ever about the right thing. If spending comes back at $76,000 he is 95% funded and the argument is nearly over; at $88,000 he is 82% and it was never close.

B. 61, 142.1% funded, still going (TR-03 scenario B). A trial will not change her arithmetic and §3 says so — above ~110% the spending error is absorbed. For her the trial is not a measurement, it is the decision itself, taken in the only reversible form available. That is a legitimate use of a leave and a poor use of a spreadsheet.

C. 56, took an unpaid three-month leave, came back reassured. He spent $16,000 over three months, annualised it to $64,000, and concluded he was comfortably funded. Both corrections in this page run against him: no lumpy items fell in the window (§4, ~15%), and he was on employer coverage throughout, so the bridge premium is missing entirely (§5). His real number is materially higher than $64,000, and the trial made him more confident and less accurate — the specific failure this page exists to prevent.

💡 Pro-Tips

1. Price the trial in the same units as the extra year before deciding between them — 45% of a marginal year, on this household’s numbers (§1).
2. Get the coverage answer in writing first. FMLA does not cover this purpose ✅, so everything about your coverage during a leave is discretionary (§6).
3. Assume the leave might be a COBRA event even though you stay employed, and check ✅. If it is, use the 60-day window as an option, not a decision (TR-01 §5).
4. Reserve the lumpy items monthly and add them back from records. A six-month trial that skips this understates by about 15% (§4).
5. Add the modelled bridge premium to whatever the trial reports — the trial almost never pays it (§5, HC-01).
6. Take twelve months if the employer will wear it. The lumpy items are annual; an annual window is the first that samples them.
7. Change one variable. Spending is the experiment; leave the portfolio alone (ER-06).
8. Write the decision rule before you start, and update TR-03 §6’s threshold with the result rather than re-opening the whole question.

⚠️ Common Pitfalls

1. Treating another year of work as free and a trial as expensive. The year costs twice the time (§3).
2. Annualising a six-month trial without correcting for lumpiness — a 15% understatement delivered with the authority of a measurement (§4).
3. Believing the trial priced your healthcare. It usually did not, because you were still on the employer plan (§5).
4. Assuming FMLA covers a sabbatical. It does not ✅ — wrong statute, wrong purpose (§6).
5. Being surprised by a COBRA notice while still employed. A reduction of hours is a qualifying event ✅.
6. Electing COBRA on day three of the leave instead of using the retroactive window (TR-01 §5).
7. Running the trial as a holiday — travel-heavy months measure a trip, not a retirement.
8. Economising during the trial to make it pass. A test you have tuned to succeed has told you nothing.
9. Restructuring the portfolio at the same time, confounding the experiment with a sequence-risk change (ER-06).
10. Expecting a trial to answer the questions in §5’s “no” column — sequence risk, longevity, and irreversibility are not six-month questions.
11. Skipping the vesting and match check before taking unpaid leave ◻️ (TR-01 §4).

✅ Actionable Checklist

Before asking

• Write down the single question the trial is meant to answer
• Price the trial against one more working year in funding-ratio points (§1, §3)
• Confirm your funding ratio is in the 90–110% band where a trial is decisive
• Pull five years of statements and identify the lumpy items (§4)

Negotiating the leave

• Get in writing whether active coverage continues or the leave triggers COBRA ✅
• Confirm the effect on vesting, pension service credit, match true-ups and bonus ◻️
• Confirm the return date and whether the role is held
• Price marketplace coverage against COBRA for the leave months (HC-01, HC-05)

Running it

• Reserve the lumpy items monthly rather than skipping them
• Track spending at the same granularity you will use in retirement
• Leave the portfolio and the allocation untouched
• Do not economise to make the number pass

Reading the result

• Add the lumpy items back from records, not from the window
• Add the modelled bridge premium (HC-01)
• Re-run the funding ratio on the corrected number
• Update the TR-03 §6 threshold and its failure branch, then decide

Sources & further reading (verified August 2026)

• tools/tr04_worked_examples.py — the trial’s cost, the spending-error band, and the lumpiness correction, all reproducible by running the file and checked against this article by CI. The 5% real growth, the guaranteed 3% benchmark, the COBRA premium and the lumpy share are stated inputs, not findings.
• Family and Medical Leave Act, 29 U.S.C. § 2601 et seq., and U.S. Department of Labor Wage and Hour Division guidance (Fact Sheet #28A; FMLA FAQ) — 12 workweeks of unpaid job-protected leave, maintenance of group health coverage on the same terms as if working, the 12-month/1,250-hour/50-employees-within-75-miles eligibility test, and the closed list of qualifying reasons that does not include personal or sabbatical leave.
• COBRA, 29 U.S.C. § 1161 et seq., and DOL/CMS continuation-coverage guidance — termination or reduction of hours as the qualifying event for a covered employee, and the 18-month continuation period. TR-01 §5 owns the 60-day election window, its retroactivity, and the rule that only exhaustion reopens a marketplace special enrollment period.
• TR-03 owns the funding ratio, the threshold, and the marginal-year arithmetic this page prices against · TR-01 §3 owns the budget rehearsal as a runway task and §5 the COBRA fork · ER-01 §3 owns the horizon benchmark · ER-06 owns the sequence risk a trial cannot test · HC-01 and HC-05 own the bridge cost a trial usually omits · ER-07 owns the permanent downshift a trial is not · TX-07 owns the spending rule the corrected number should feed.