UK Lotto · 6 from 59 · two rounds per draw

Same odds.
Fewer people
to share with.

Nothing changes your chance of winning the lottery. Two things do change how much you take home: which draws are worth entering, and which numbers you put on the slip. This toolkit works out both, from real draw data, and checks its own answers against real money.

Next draw · 26 September 2026

SKIP

Jackpot
£5,840,169
A £2 line returns
−£1.04
Worth playing above
£32,290,465

Panel A · your slip

Five lines nobody else is playing

The bar on each line is how many people play it, against an average line at 100%. Shorter is better: fewer people to split a jackpot with.

  1. Your first line32%
    1. 32
    2. 33
    3. 39
    4. 41
    5. 52
    6. 53
  2. 0232%
    1. 32
    2. 38
    3. 39
    4. 43
    5. 48
    6. 58
  3. 0332%
    1. 35
    2. 36
    3. 40
    4. 42
    5. 49
    6. 58
  4. 0432%
    1. 34
    2. 36
    3. 37
    4. 44
    5. 50
    6. 54
  5. 0532%
    1. 32
    2. 37
    3. 43
    4. 45
    5. 47
    6. 49

Built in your browser from the same model the toolkit runs, calibrated on 1,147 draw-rounds of real winner counts. Nothing is sent anywhere.

Six numbers from 59, drawn to avoid the dates, the lucky sevens and the diagonal patterns most tickets carry. Exactly the same chance of coming up as any other line — but if one does come up, you are sharing it with far fewer people.

In short

How this works

Pick numbers nobody else picks. Prizes below the jackpot are fixed, but the jackpot is split between everyone holding the winning line — and people overwhelmingly play dates. A line avoiding 1–31 is picked by about 32% as many players as an average one, measured across 1,147 draw-rounds of real winner counts. Same odds, a much bigger share.

Then play only the draws that pay. An ordinary draw needs a jackpot of £32,290,465 before a £2 line is worth its price. A Must-Be-Won draw — where an unclaimed jackpot rolls down into the lower tiers — needs £9.2m on a Wednesday and £13.2m on a Saturday — Saturdays sell half again as many lines to share it between. The cap forces one every few weeks, which is not the same as an opportunity. Since June 2026 the jackpot takes 8.88% of sales and restarts at £2m, so the 4 forced draws of this era (3 on Saturdays and 1 on a Wednesday) carried only £7.8m–£9.6m. The next is Saturday 3 October 2026, projected at £8.6m.

The draws that can pay are the operator’s specials — a guaranteed £12m, £15m or £20m with a campaign behind it. They sell more than any other draw, so one needs about £14.3m on a Saturday and £13.9m on a Wednesday to be worth playing, and £16.5m to stay worth it at the busy end of what past specials sold. A £20m special clears that; a £15m one is a coin toss on sales; a £12m one does not.

And do not believe anyone who claims more. This project tested four prediction methods over 930 real draws and none of them beat random picking. That finding is at the bottom of this page, with the numbers behind it, because a tool that hides its own negative result is not worth trusting with the positive ones.

Figures price the draw of 26 September 2026, from data collected through 23 September 2026.

Panel B · why those numbers

Everyone plays their birthdays

Numbers up to 31 are days of the month, and up to 12 they are months too. You can see it in the winner counts: draws made of low numbers produce far more small winners per ticket than draws of high ones. Nobody is picking 53.

Pick six

least played (0.60)  →  most played (1.36)

Your line is played

×3.83

as often as an average line

Share of a £10,000,000 jackpot
£6,593,339
after splitting with everyone else holding it — in both rounds. Only round one’s co-winners hold your numbers; round two’s are drawn independently of your slip, so half the sharing is not yours to avoid.
Match-3 winners per ticket, by how many of the six drawn numbers were 31 or below. Across 1,147 draw-rounds.
  1. 00.75×
  2. 10.78×
  3. 20.92×
  4. 31.00×
  5. 41.20×
  6. 51.52×
  7. 61.84×

Panel C · when to play

And which draws are worth it

A line pays back the fixed tiers whatever happens, plus a share of the jackpot if it wins. Add those up, subtract the £2 you paid, and you have what the ticket is actually worth. Move the jackpot and watch for the crossing.

£5,840,169
Ordinary draw
−£1.04
8,401,561 lines sold · breaks even at £32,290,465
Must-Be-Won roll-down
−£0.71
10,669,982 lines sold · breaks even at £13,170,825

Both below zero. Every pound spent here is a donation.

Two things carry most of the difference. A Must-Be-Won draw must pay its jackpot out, so an unclaimed pool rolls down into the lower tiers and every ticket takes a slice — worth far more than a share of a jackpot almost nobody wins. And more people play those draws, which dilutes the slice: the figures here assume 10,669,982 lines against an ordinary draw’s 8,401,561.

Fixed prizes are observed (20 draws), and both regimes are priced for the draw of 26 September 2026.

Panel D · the one exception

When the jackpot has to be paid out

After the jackpot rolls over 5 times it becomes Must-Be-Won: if nobody matches six, the whole pool cascades into the lower tiers instead of rolling again. Suddenly a £2.00 line is claiming a share of eight figures for matching two numbers.

£5 to every Match 2 winner first; whatever remains is shared among the Match 3 winners. Ribbons sized by pounds, at the median historical roll-down pool of £12,011,099 — so a Match 2 pays £5.00 instead of £1, and a Match 3 about £26.63 instead of £10.
Roll-downs found
89
Driven by the cap
53
Actually worth playing
42%
Median line
−£0.32

89 roll-downs between 2019-01-30 and 2026-07-18, about 11.9 a year. Only 22 of the 53 cap-driven ones clear break-even — the rest sold too many tickets for the pool to be worth chasing. Counterfactual: historical roll-downs re-priced under today's two-round rules, each at its own measured ticket sales. An order of magnitude, not a record of what was actually paid.

Panel E · the wheel

6 lines that cover each other

Pick the 12 least-played numbers and arrange 6 lines across them so that any four coming up guarantees at least 3 of your lines match three. Not a better chance of winning — a promise about what happens when you do.

The pool

  1. 32
  2. 34
  3. 36
  4. 39
  5. 41
  6. 43
  7. 46
  8. 48
  9. 50
  10. 53
  11. 55
  12. 57
323436394143464850535557
01
02
03
04
05
06

Every line is played by 32% as many people as an average one, and every line has the same expected value as every other.

Panel F · real money

What it cost to find out

The model said SKIP for the draw of 8 August 2026. A ticket went in anyway — ten lines, because a claim about expected value that has never been tested with money is just arithmetic.

Staked
£20
Returned
£5
Net
−£15.00
Return
−75%

Of 10 lines, 5 matched 2, 1 matched 1, 4 matched 0. Nothing above two numbers, in either round. That is the ordinary outcome, and the reason the verdict is almost always SKIP.

10combinations so far

Interlude · the scale of it

Six numbers from fifty‑nine

Pick six. There are 45,057,474 ways to do it, and the draw picks one.

That is the whole game. Everything on this page is a fraction with that number underneath it.

You cannot buy your way out

A line costs £2. Covering every combination would cost £90,114,948 — far more than any jackpot this game has ever paid.

Almost every ticket is a loser

The jackpot is 1 in 45,057,474. The likeliest thing that happens to a ticket, other than nothing, is matching two numbers for £1 — and that is still only 1 in 10.

Which is why the sharing matters

Against a number this size, nothing you do moves your chance of winning. That is also why the two levers above are the whole game: they are not about winning more often, they are about what the win is worth and whether the ticket was priced fairly in the first place.

Running average of numbers matched, 2017-09-09 to 2026-08-05. Four prediction methods and a random baseline, drawn identically because that is the finding.

Panel G · the check

We tested prediction too. It does not work.

Every lottery system claims an edge. This one tested its own, on 930 real draws, and found none.

Walk-forward: each draw is predicted using only the 200 draws before it, so nothing is fitted on the answer.

4 methods, one picture

Frequency counting, weighted sampling, a probability map, and a consensus of all three. Their running averages converge on 0.610 — which is just 36/59, what six random numbers match by arithmetic — and stay there for nine years.

Inside the noise, every one

Against 10,000 Monte-Carlo replicates of the no-skill model, the best method scores 0.6194 against a random baseline’s 0.6151. Every confidence interval straddles the no-skill mean.

Which leaves the arithmetic

Draws are independent and uniform, and no amount of history changes what comes out of the machine. So the rest of this page is not about predicting anything. It is about the two decisions that remain — and those turn out to be worth real money, just not often.

Panel H · how it runs

Nobody presses anything

Every figure on this page was collected, checked and published without a human in the loop. 452 tests across 32 files gate every change, and the alerts stay silent unless something is worth reading.

No compute on AWS. It stores files and serves them; the lottery toolkit runs on GitHub Actions and a laptop, and the data lives in the repository.
Collect draw data
fetch the official feed, assert it is fresh, score a Must-Be-Won, refresh this snapshot, commit the CSVs, email only on a PLAY verdict
Wed/Sat evening · Thu/Sun morning retry
Collection watchdog
an independent check that the draw actually landed
Thu/Sun midday
CI
the Python suite, plus a drift check on this snapshot
every push and pull request
Deploy site
build, assume an AWS role over OIDC, upload, invalidate, smoke test
every push to main touching the site · dispatched by the collector when a draw moves this file

Things that are not true of it

It is not punctual. GitHub's cron is best-effort: on this repo it ran 13-58 minutes late through August 2026 and has drifted to 2-11 hours since. Two runs per draw plus an independent watchdog is the answer to that, not punctuality.

Success was not always success. the collector returns success when it falls back to cached data, so a separate step asserts the expected draw is present - a silent no-op becomes a red build.

A missed run is survivable. the feed serves every draw of the last ~180 days by number, so a missed collection is backfilled on the next run — a window of about 180 days.

Deployment assumes a role over OIDC, so no long-lived AWS key exists to leak. The bucket is private and reachable only through the CDN, and the role that publishes the site cannot change the infrastructure serving it.