When a lottery changes its numbers
UK Lotto drew 6 from 49 for twenty-one years. Since October 2015 it has drawn 6 from 59. Both are called UK Lotto, and any statistic that treats them as one game is wrong — while looking entirely reasonable.
How often this happens
More often than people expect. Among the games we cover, 5 matrix changes have taken effect, and none is currently pending.
| Game | Date | Matrix | Jackpot odds | What moved |
|---|---|---|---|---|
| UK Lotto | Saturday 10 October 2015 | 6/49 → 6/59 | 1 in 13,983,816 → 1 in 45,057,474 | pool grew by 10 |
| Mega Millions | Tuesday 8 April 2025 | 5/70 → 5/70 | 1 in 302,575,350 → 1 in 290,472,336 | the extra ball |
| Lotto Max | Tuesday 14 April 2026 | 7/50 → 7/52 | 1 in 99,884,400 → 1 in 133,784,560 | pool grew by 2 |
| UK Lotto | Wednesday 10 June 2026 | 6/59 → 6/59 | 1 in 45,057,474 → 1 in 45,057,474 | rounds per draw |
| Irish Lotto | Saturday 5 September 2026 | 6/47 → 6/45 | 1 in 10,737,573 → 1 in 8,145,060 | pool shrank by 2 |
Why a bigger pool means a bigger jackpot
This is arithmetic rather than motive. Adding numbers to the drum lengthens the odds: UK Lotto went from 1 in 13,983,816 to 1 in 45,057,474 by adding 10 balls — more than three times harder.
Longer odds mean the jackpot is won less often, and a jackpot that is not won rolls over. Rollovers accumulate into the very large headline figures that sell tickets. Every pool expansion in the table above has that same shape.
Which makes Irish Lotto the interesting case, because it went the other way: 6/47 down to 6/45 on Saturday 5 September 2026, shortening the odds from 1 in 10,737,573 to 1 in 8,145,060. Smaller headline jackpots, won more often. It is a different bet, and it is a legitimate one.
What it does to a frequency table
Here is the failure, and it is not subtle once you see it. Numbers 50 to 59 did not exist in UK Lotto before Saturday 10 October 2015. They could not be drawn. They were not in the drum.
Pool the two eras into one frequency table and every one of those 10 numbers gets nothing at all for the earlier stretch. Take 1,000 draws on each side of the change:
| Number | Expected appearances |
|---|---|
| A number in both drums (say 39) | 224 |
| A number only the new drum has (say 50) | 102 |
50 would show fewer than half the appearances of 39 — a factor of 2.2 — and it would stay that way forever, with nothing whatever wrong with the ball. It would sit at the bottom of every "coldest numbers" list for as long as anyone kept the archive, and the list would be measuring a rule change rather than a drum.
This is the single most common error on lottery statistics sites, and it is invisible. The table looks fine. The numbers look plausible. Nothing about it announces that it is meaningless.
What we hold, and what we have not measured
The figures above are expected appearances, worked from the two matrices and a stated number of draws. They are arithmetic, not a result, and you can redo them from this page: 6 from 49 gives each number 122.4 expected appearances in 1,000 draws, 6 from 59 gives 101.7, and a number present only after the change gets the second figure and nothing for the first.
What we have measured is narrower, and saying so is the point of this section. Our archives are not thirty years deep — several of our sources publish only a rolling window — so here is exactly what we hold either side of each change:
| Game | Changed | Draws before | Draws after |
|---|---|---|---|
| UK Lotto | Saturday 10 October 2015 | 0 | 71 |
| Lotto Max | Tuesday 14 April 2026 | 0 | 27 |
| Irish Lotto | Saturday 5 September 2026 | 25 | 32 |
One archive straddles its boundary. On Irish Lotto we hold 25 draws under 6 from 47 and 32 under 6 from 45. The numbers that left the drum (46 and 47) appear 14 times before the change and 0 times after.
That is the mechanism in miniature, and it is also why the table above is still a calculation. A number present throughout should average about 3.2 appearances before the change and 4.3 after, on these counts — a difference of 1.1. What the archive actually shows, counting each era separately, is a range of 0 to 9 appearances across the 45 numbers in both drums. The noise is roughly 8 times the effect being looked for. 57 draws is enough to show a number stop appearing, because that is structural; it is nowhere near enough to measure a frequency distortion. We have not measured the factor, and we are not going to claim it from a sample this size.
If you want to check any of it, every game's page says what it holds and from when, and the open data page names the operator's own source per game.
Two changes that do not touch the pool at all
Mega Millions, Tuesday 8 April 2025. The main matrix stayed at 5/70 and the jackpot odds still moved, from 1 in 302,575,350 to 1 in 290,472,336, because the extra ball is drawn from its own separate pool and that pool changed size. The main-number statistics are unaffected; the odds are not. They are different questions and a registry that stored one number per game could not tell them apart.
UK Lotto, Wednesday 10 June 2026. The matrix did not move and the jackpot odds did not move. What changed is how many times the drum runs: from 1 round a night to 2. The odds per round are identical, but a "draw" is now two results rather than one — so every count on a statistics page has to be a count of rounds, not of nights, or it halves overnight for no reason. This is the change that catches archives out, because nothing about the numbers themselves looks different.
How we handle it
A game's matrix is stored as a dated record rather than as a property of the game. There
is no game.mainPool anywhere in this codebase to get wrong. Every draw
carries the identifier of the format in force on its own date, and every statistic is
computed inside one window and never across a boundary — which is why our
UK Lotto statistics start where they do
rather than in 1994.
The lookup throws rather than guessing. Asking for a matrix on a date we have no record for is an error, not a default — because a default would be wrong silently, and silently is the whole problem.
That rule has a consequence we accept deliberately. When an operator announces a change but has not yet published the new pool sizes, our collector refuses to write any such draw dated on or after that day rather than score it against the matrix being replaced. A gap in the archive can be filled later, and we say on the page that it is there. A draw filed under the wrong matrix cannot be found again, because nothing about it looks wrong afterwards. A nightly job reads the operator's own published rules and stops the build the moment they change. That is the part worth copying if you keep an archive of your own: the failure you are guarding against is not noticing.
What it means for you
Practically, three things. Any "most drawn numbers" list that covers more than about a decade of a UK or Australian game is probably pooling two matrices. Any figure quoted for a game's odds without a date attached may be describing the previous game. And a number that looks permanently cold is worth checking against the date its drum acquired it.
None of which helps you pick, because nothing does. Generate a line against the matrix that is actually in force, which is the one thing we can promise.