Skip to content
Back to blog
⚔️
Our own study

Do Arthur, Guinevere, Lancelot and Merlin Have Favourite Numbers? UK Lotto's Draw Machines, 1994-2026

October 27, 20269 min read

For three decades Britain's Lotto numbers came out of machines with names - Arthur, Guinevere, Lancelot, Merlin, then gemstones, and since 2025 plain numbers. Some players suspect that one of them, or one set of balls, has favourite numbers. We matched an independent log of every machine and ball set against our own archive and tested each one with enough draws. Here is what 3,258 rounds show, and what they can't.

3,258
draw rounds since 1994
15
machine runs tested separately
0
flagged at the 5% level
2026
last draw by an Arthurian machine

Guinevere Is in a Museum Now

On 19 November 1994 a machine called Guinevere drew the first numbers of Britain's National Lottery: 30, 3, 5, 44, 14 and 22, bonus ball 10. A Guinevere made in 1994 is now in the collection of London's Science Museum, name plaque and all, which lists it in its mathematics gallery. Its colleagues were called Arthur, Lancelot and Merlin, later joined by Galahad and Vyvyan.

With the names came a suspicion that keeps resurfacing: that one machine, or one set of balls, favours certain numbers. Checking it needs two facts per draw that most archives don't keep - which machine ran and which ball set was in it. The operator shows them on its results pages, but only for recent draws. Two private results sites list them all the way back to the first draw. We checked the more complete one against our own data, the operator's pages and the other site, then tested every machine on its own.

Four Generations, Three Naming Schemes

The log belongs to Richard K. Lloyd, who runs an unofficial National Lottery site at lottery.merseyworld.com. It covers all 3,218 Lotto draws up to 24 October 2026. Since June 2026 every draw has two rounds, so that makes 3,258 rounds. In every one of them the date, the six numbers and the bonus ball match our own archive. The machine names fall into a clear sequence:

MachinesFirst roundLast roundRounds
Arthur, Guinevere, Lancelot, Merlin, Galahad, Vyvyan19 Nov 199422 Oct 2003817
Amethyst, Topaz, Sapphire, Moonstone, Opal, Pearl, Garnet25 Oct 200318 Nov 2009634
Arthur, Guinevere, Lancelot, Merlin (names reused)21 Nov 200910 Jan 20261,633
Lotto 1, 2, 4, 5, 69 Jul 202524 Oct 2026174

From the lottery.merseyworld.com log. Lancelot appears there from April 1995, Merlin from July 1998, Galahad from 2000 to 2003 and Vyvyan from 2000 to 2002. Seven rounds from May and June 2026 that the log lists as Lancelot, and one ball set, are corrected from the operator's results pages. The 3rd and 4th generations overlap because Arthur still drew twice in January 2026.

On 25 October 2003 the Arthurian names vanish from the log and gemstones take their place: Amethyst, Topaz, Sapphire and others. On 21 November 2009 Arthur, Guinevere, Lancelot and Merlin return. According to Wikipedia these were new machines that simply took over the old names; we found no official confirmation of that. Because the question is how devices behave, we still treat them separately from the 1994 machines. Anyone who thinks of Guinevere as one machine with 30 years' service may be comparing two different pieces of equipment.

Lancelot drew for the last time on 5 July 2025. Four days later, on 9 July, a machine called "Lotto 2" ran for the first time. The results site lottery.co.uk noticed a transparent rather than solid background in the area where the drawn balls collect, and asked the operator about it without getting an answer. Since February 2024 the National Lottery has been run by Allwyn instead of Camelot, a name that went naturally with the Arthurian set. A connection seems likely, but as far as we know the change was never officially explained. After its debut, Lotto 2 drew almost every draw in the log. Only on 7 and 10 January 2026 did Arthur stand in again, which lottery.co.uk confirms - the last Arthurian draws we found. Lotto 4 joined in February 2026 and Lotto 5 in March. Since May 2026 the operator's results pages have named Lotto 1, Lotto 4, Lotto 5 and Lotto 6. Since 10 June 2026 two of them have been used at every draw, one for each round.

What the two-round format pays

The Test: Each Machine on Its Own

A fair machine draws every number equally often in the long run. In a 6 from 59 game any given number is among the six winners in just over 10 per cent of rounds, so after 256 rounds it should have turned up about 26 times. It never comes out exactly: some numbers run ahead, others lag. The question is whether the gaps are bigger than chance produces on its own.

So we counted the main numbers each machine drew, split by generation and by format - the game was 6 from 49 until October 2015 and has been 6 from 59 since. We left out machines with fewer than 100 rounds, which are too short to show much of anything. That left 15 machine runs with 2,814 rounds between them. For each we worked out a chi-square score, which rolls every deviation from the fair expectation into one number. Rather than read off a table how often a fair machine produces such a score, we tried it: 20,000 simulated runs of the same length, drawn without replacement just as the real machine does. The share of simulations with a score at least as high is the p-value.

MachineFormatRoundsMost drawnExpectedp
Arthur (1st generation)6/4923838 (42×)29.10.41
Guinevere (1st generation)6/4919044 (34×)23.30.90
Lancelot (1st generation)6/4918138 (35×)22.20.12
Merlin (1st generation)6/4910847 (22×)13.20.40
Amethyst (2nd generation)6/4917239 (30×)21.10.83
Topaz (2nd generation)6/491579 (31×)19.20.13
Sapphire (2nd generation)6/4913524 (25×)16.50.81
Arthur (3rd generation)6/4914417, 33 (25×)17.60.91
Guinevere (3rd generation)6/4917129 (28×)20.90.85
Lancelot (3rd generation)6/4914235 (26×)17.40.15
Merlin (3rd generation)6/4915724 (29×)19.20.71
Arthur (3rd generation)6/5925752 (38×)26.10.65
Guinevere (3rd generation)6/5925649, 54 (39×)26.00.11
Lancelot (3rd generation)6/5925734 (35×)26.10.79
Merlin (3rd generation)6/5924937 (34×)25.30.97

Main numbers only, no bonus ball. "Expected": how often each number would be drawn on average by a fair machine. p-values from 20,000 simulations, rounded. Where numbers tie for most drawn, all are shown.

None of the 15 machines stands out. The lowest p-value is 0.11, meaning roughly one fair machine in nine would look at least this uneven. The usual threshold for "suspicious" is 0.05, and with 15 tests you would expect close to one to dip below it by chance alone. Not even that happened here.

We checked the ball sets the same way: sets 1 to 8, each for 6/49 and 6/59, which makes 16 tests. The lowest p-value, about 0.05, sat right at the threshold, for set 5 in the 6/49 years. Across 16 tries, one result like that is about what chance alone would deliver. Besides, a set number in the log is a label, not proof that the same physical balls were in use for 21 years.

Why Guinevere's 49 and 54 Aren't a Signal

Scan the table for outliers and one row jumps out. In the 6/59 years the 2009 Guinevere drew 49 and 54 no fewer than 39 times each, against an expected 26.0 - half as often again as they should. Those look like favourite numbers. It isn't, because with 59 candidates some number is always the most frequent. We simulated 256 fair 6/59 draws 20,000 times: the most frequent number came up 38 times in the typical run, and 39 times or more in 35 per cent of them. Two numbers both reaching that mark is rarer, about 6 per cent of runs. But look through 15 machines and something that rare is likely to turn up somewhere. The test that weighs all 59 numbers together puts Guinevere at 0.11 - nothing unusual.

It's the same trap as with "overdue" numbers: pick the most striking value after the fact and forget how many you looked through to find it. Our UK Lotto statistics page shows how often every number has come up since 2015, with the same warning.

What This Test Can't Do

Finding nothing doesn't prove every machine was perfect. It means any tilt would have had to be too small to show in a few hundred rounds. How small? After 256 rounds of 6/59, a single number's count wanders by about 4.8 either side of its expected 26.0 through chance alone. For a test at the 5 per cent level to notice it at all, it would have to come up about 36 per cent more often than fair - and since you are checking 59 numbers at once, more than that. A gross fault would have shown; a subtle one would not.

The second limit is the data itself. For the 51 draws from 2 May to 24 October 2026 (91 rounds) we could compare the log with the operator's results pages. The numbers matched throughout. But in seven rounds from May and June 2026 the log says "Lancelot" where the operator says Lotto 6 or Lotto 1, and for 22 August 2026 it gives ball set 4 for both machines, while the operator lists set 3 for one of them. We corrected those eight entries before testing. For the decades before, there is no official list, but there is a second site: for 80 randomly chosen draws from 1995 to 2026, lottery.co.uk names the same machine and the same ball set every time, and it doesn't share the 2026 errors. Occasional wrong labels would blur a real tilt a little, but they wouldn't make a large one disappear.

What It Means for Players

Even if a machine had favoured a number, it wouldn't get you far. Suppose one number came up 20 per cent more often than fair - a tilt that could slip through this data unnoticed. A line containing it would then have a jackpot chance of about 1 in 37,500,000 instead of 1 in 45,057,474. That's no edge worth the name, and you'd still pay full price for the line.

The more useful lesson is a different one. The safeguards that keep a lottery fair - such as rotating several machines and several ball sets - are meant to stop gross faults and tampering. Thirty-two years of draw data are consistent with them having done that job. A winning strategy can't be squeezed out of them. If you notice yourself hunting for patterns to win back losses, our responsible gambling notes list places to get help.

How Would Your Numbers Have Done?

Our checker runs your numbers against every UK Lotto draw since 1994 and shows you what they would have won - and what you would have paid to play them.

Sources: machines and ball sets for every round from 19 November 1994 to 24 October 2026 from Richard K. Lloyd's log (lottery.merseyworld.com, unofficial), numbers checked against the LottoROI archive (all 3,258 rounds identical); numbers, machines and ball sets for draws 3168 to 3218 from the National Lottery's results pages (national-lottery.co.uk); machines and ball sets for 80 sample draws, the first draw and the January 2026 draws from lottery.co.uk's results pages; Science Museum Group Collection, object 2006-207 ("Guinevere, Random Number Selector and name plaque"); lottery.co.uk, "New Ball Machines for Lotto Draws, And A New Studio in London!" (15 July 2025); iGaming Business, "Allwyn takes control of UK National Lottery" (1 February 2024); Wikipedia, "National Lottery (United Kingdom)", on the origin of the 2009 machines (not officially confirmed). Tests and simulations: our own calculation.

Disclaimer: Gambling can be addictive. Play responsibly. Participation from 18 years. This article is for informational purposes only.

More on this topic

Back to blog