Ask how big the reserve should be and the internet answers in months. Three to six of them. That answer is not wrong — this site already shows its whole arithmetic, the multiples and the 2026 dates on every row, in the sizing article. What that answer is, though, is a months answer. A month is a tidy shape for a salaried life: the money arrives in twelves, the rent leaves in twelves, and if the money stops, the arithmetic of the shock comes out in whole months, so the whole months of reserve fit over the whole months of shock like a lid on a pot.
Variable pay has no such lid. A commission cheque that slips from the fifth of the month to the twentieth is not a month of missing income; it is three weeks of late money. A gig season that thins is not one bad month; it is a run of half-full ones. An invoice moving from net-30 to net-90 is a gap that arrives in weeks and sits down for a quarter. Which is why I size cushions in the unit the gap actually moves in — and why, for three of the four income shapes below, the honest answer is not inside the flat range at all. It is past it.
Eleven weeks of money is not a plan, and I have been saying it since 2020. The table below is what I say alongside it now, because eleven weeks sits in three different places depending on the shape of the pay: on a salaried household it is two weeks short of that household’s floor; on a commission earner it is short of its own floor by a full quarter of a year; on a self-employed project worker it is not even a third of the way. Same pile of money. Three different households. The flat advice cannot tell them apart — the shape of the income can.
(And in case you keep your own books in pounds like I do: the cost set below is the site’s dated American one, dollars and all. The arithmetic is the same on both sides of the Atlantic; only the symbol changes.)
Why months is the wrong unit for unsteady pay
Three reasons, and the first one is just arithmetic.
A month is not four weeks. The calendar hands you 365 ÷ 7 = 52.14 weeks a year, which is 4.345 weeks to the month (365 ÷ 7 ÷ 12). Size in months, round loosely, and you drift: three months is 13.0 weeks, six months is 26.1 weeks, and by the time you quote “six weeks” as half of “three months” you are already a week light. The flat range, written in weeks, reads 13 to 26 — say that to a freelancer and watch how differently it lands from “three to six”, because weeks are the unit their money moves in.
Second, months hide the shape of a gap while weeks count it. The site’s sizing article leans the flat range “to the long end” when the income can go quiet, and that is the door. This article walks through it and counts the rooms beyond it: for an hourly or gig household, and for a project self-employed one, the arithmetic lands past the 26-week end of the standard range. That extends the months article rather than contradicting it — the range it dates is right, it is simply only the floor, and only for the paycheque shape it was measured against.
Third — and this is the part the flat advice skips — how many weeks is only half the row. What good is a number of weeks that your own month could not survive the first bill? So the table’s rows carry their arithmetic, and every arithmetic rests on a dated input: the site’s dated cost set, the dated median pay, this run’s dated price and labour-market series. Rules with their receipts, not vibes.
What a week of income gap costs, dated 2026
The decision table multiplies one number, so here it is. The site’s dated essential-outgo set is $3,784 a month — $2,000 national average rent (Zillow, March 2026), $884 for the medical exposure (the 2026 individual out-of-pocket maximum of $10,600, spread over twelve months), and a stated $900 line for food, utilities, transport and phone — all built, dated and linked in the sizing article. Convert it on the calendar frame: $3,784 ÷ 4.345 = $870.84, so about $871 of survival per week.
One plain word about frames, because my own last article quotes a nearby number and there is a reason for it: the 50/30/20 stress test works a week out at $873, which is the payroll frame ($3,784 × 12 ÷ 52 — the frame a payslip thinks in). The buffer spends in calendar weeks, so this table sizes in the calendar frame at $871. The difference is $2.39 a week, about $124 a year — one cost set, two honest frames, stated rather than hidden.
Three dated rows sit under the $871, and I pulled two of them fresh for this article:
- The floor bill is dated. The 2026 individual out-of-pocket maximum is $10,600 — the most a health plan can make you pay yourself in a year, and the biggest single shock on this site’s cost ledger. Divided by the weekly burn it is 12.2 weeks. Whatever your row turns out to be, a cushion that cannot swallow the medical maximum is not finished.
- The burn re-dates, slowly and unevenly. Prices ran 3.35% over the twelve months to August 2026 (consumer price index, Bureau of Labor Statistics series, pulled from the series this morning — the site’s dated “about 3.4%” line, re-verified). But the shelter component — the rent inside the cost set — ran +2.0% on the same pull. The biggest row of your outgo is the one drifting slowest; when you re-run the table next year, inflate the $900 running-costs line before you inflate the lease.
- The labour market arrives in weeks before it arrives in months. The broad underutilisation measure (U-6 — unemployed people plus the discouraged, the marginally attached, and everyone working part-time because they cannot find full-time) sat at 7.6% in September 2026 (BLS series, pulled this morning). For a salaried household the bad news is usually a date — a layoff, a start line, a whole month. For an hourly or gig household, the bad news is the U-6 news: thin hours, a light booking month, a quarter that never fills. The gap comes as weather, not as an earthquake — and weeks are the forecast horizon weather moves in.
And what it costs to not have the weeks. Take the site’s own long end: a 26-week gap at $871 of burn is $22,642 of outgo. Bridge that on the card the site dates at 22.15% average APR (Federal Reserve G.19, Q2 2026) and six months of carrying it costs about $2,508 — which is $418 a month you chose not to spend, plus 2.9 more weeks of the buffer burned just standing still. (Which card, and the order to kill one if you already have it, is the debt lane’s work — it is counted there, not here.) The reserve does not make shocks impossible. It makes them ordinary-sized.
The decision table: weeks by income stability
Read your row by the shape of your pay, not by your job title. Plenty of salaried managers have a commission shape; plenty of “freelancers” actually live on one big retainer, which is row A with a costume.
| Income shape | Weeks to hold | Why — the arithmetic | What re-dates your row |
|---|---|---|---|
| A. Salaried, steady pay | 13 (the flat range’s floor, in weeks) | 12.2 weeks covers the $10,600 medical maximum; 13 covers it plus two years of the dated 3.35% price drift on top ($10,600 × 1.0335² = $11,322 against 13 × $871 = $11,321) | A second household member relying on you; any move to variable pay — that is a row change, not a raise |
| B. Commission / bonus-weighted | 22–26, floor at 23 on a 30% variable share | At the dated $5,027 median (BLS, Q2 2026, ages 25–34), a 30% variable share is $1,508 a month; two missed commission quarters is $9,049 = 10.4 weeks; plus the 12.2-week medical floor = 22.6, round up to 23, lean to 26 if one quarter is more than half your year | Your variable share moving ±10 points — re-run the two-quarter math, it moves the floor about ±1.7 weeks per 5 points |
| C. Hourly / gig / schedule-volatile | 30 (26 only with a second, genuinely unrelated income) | Half the dated median ($2,514 a month) missing across the site’s own long end — six months, 26 weeks — is $15,081 = 17.3 weeks; plus the 12.2-week medical floor = 29.5, round to 30 | Two thin months in a row is an early-warning trigger, not a surprise; a season you already know is soft should move the target, not the calendar |
| D. Self-employed / project-based | 39 (52 if one client is over half your income) | One lost trading quarter (13 weeks) + one paid-late quarter riding on top of it (13) + the medical maximum (12.2) = 38.2, round to 39 — $33,963 at $871 | Client concentration crossing 50% — then you are your own payroll, and a payroll does not owe you a quarter’s float: hold 52 |
The stated planning lines are labelled as such, because that is the deal: the 30% variable share in row B is my planning assumption, not a dated statistic — run your own share through row B’s two-quarter arithmetic (at 25% the floor comes down to about 21 weeks; at 35% it climbs to 24). Everything else in the table rests on a dated input: the $871 burn and the $10,600 maximum from the site’s dated cost set, the $5,027 median from the BLS Q2 2026 earnings the site carries, the 26-week horizon the sizing article’s own six-month long end, the 3.35% and the 7.6% pulled from the BLS series this morning.
Two things the table says out loud that the flat range does not. First, the long end past 26 weeks is not exotic — rows C and D are a large share of the working world, and their floors sit at 30 and 39 weeks by plain arithmetic on dated inputs, not by caution piled on caution. Second, the rows are not a ranking. Row A at 13 weeks is not the lazy row; it is the row whose shock really does arrive in months, with the employer inside the shock absorber. Row D at 39 is not reckless for needing 39 — the invoice float is a structural cost, the same cost a business carries as receivables, and the self-employed household is a business that pays itself last.
The self-check on your own income
Three questions, ten minutes, your own payslips. This is the diagnostic, not the table — the table is what the answers look like when everyone’s run it.
- Count your variable share. Of the money you actually receive, what share could legitimately arrive late, thin or not at all in a given quarter? Call it v. (Fixed salary: 0. Commission over half your year: 0.5 and counting. All of it: 1.)
- Count your plausible quiet horizon, in weeks. Not the nightmare, the bad-but-real season you have already seen once: two missed commission quarters is 26 weeks of thin money; a booked-out summer that books out twice is 13. Hourly and gig households with no second, unrelated income line start at the site’s own long end, 26.
- Do the two sums. The gap fund is v × (your monthly pay ÷ your weekly burn) × (quiet weeks ÷ 4.345)… said less like an accountant: your pay, times your variable share, times the number of calendar months of quiet — that is the money missing; divide it by your $-per-week burn and it is in weeks. Then add the 12.2 weeks of the medical maximum, because that shock rides on top of a bad-income year rather than instead of one.
Worked on the dated median, at the table’s stated 30% share and one quiet quarter: $5,027 × 0.30 × 3 months = $4,524 missing ÷ $871 = 5.2 weeks, plus the medical’s 12.2 = 17.4 — call it 18 weeks before you even have a floor; row B’s 23, with two quiet quarters on the table, is what you save toward from there. Run it on your own three numbers and you will not need my table any more; you will have your row.
What you are checking for is the honest number, and this is the site’s whole diagnostic in one sentence: weeks of money is the number a household can state, verify and improve; “comfortable” is not. In, out, the gap — then the gap’s shape.
How fast the row fills, honestly
A target nobody can reach is a guilt machine, so finish the arithmetic in the other direction — and this run’s two dated series both say the same uncomfortable thing.
The Bureau of Economic Analysis put the personal saving rate at 4.1% of disposable income in August 2026 (Personal Income and Outlays, released September 30) — the country, as a whole, keeps about four cents of every after-tax dollar. Your row is not built at country speed, though, and the site already dates the two household engines that decide it, both from the stress test: the floor month’s kept 20% — $839.83 — banks 11.6 weeks of buffer a year on the calendar frame (11.5 on the payslip’s, the same money either side of the $2.39); a shock month’s $415.17 banks 5.7, same in either frame:
- 13 weeks — row A’s floor — is about 13.5 months of good months. Everyone’s first foundation; sequence it before anything clever.
- 23 weeks — row B’s floor — is roughly two years of good months.
- 30 weeks — row C — is about two and a half years of good months, or two years of good months plus a side line that is genuinely someone else’s demand.
- 39 weeks — row D — is about three and a half years. Which is why, for self-employed households, the reserve is not a savings goal; it is a standing position, like working capital, funded every quarter the invoice clears, the way a business funds it.
This is the sequence, and it is the same one this site runs everywhere: auto the reserve before the discretionary money exists to spend, build to 13 first — the floor every shape shares — then run toward your row, and only then look at the clever stuff. If you have not yet routed a first paycheck into any order at all, the engine comes from the five-step sequence; the same arithmetic runs on your own rent, your own variable share and your own horizon in the reserve calculator, which does months — convert at 4.345 and the answer lands on the row above.
And the close is the one the whole site keeps: work out the gap before you touch the rest; the cushion is the foundation the clever stuff gets built on. The flat advice says three to six months to everyone and fits almost no one exactly. Find your shape, run your three numbers, and keep the honest number you actually need — thirteen weeks if your paycheque never lies to you, thirty-nine if your invoices do. The foundation first, the clever stuff after.