Published · 2026-10-25 · Last reviewed · 2026-10-25 · 7 min read

How a lottery draw is kept fair

Doubts about the fairness of lottery draws have always existed, and they usually rest on a feeling: "this cannot be chance". This page answers the question in two parts. One is what procedures a draw uses to try to guarantee fairness; the other is how far, holding the published results, statistics can check. The description of procedures rests on press reports, and the statistical checks are our own calculations. We say at the outset that neither proves "there was no manipulation"; what can be shown is narrower and more specific.

What a fair draw needs

A lottery draw needs four things.

Randomness: every combination must come up with equal probability.
Unpredictability: nobody can know the result before the draw.
Tamper resistance: equipment and balls cannot be manipulated.
Verifiability: outsiders can watch the procedure or check the results.

Unlike the computer random numbers of the earlier article, How computers make random numbers, a real Lotto draw is a physical device using balls. The advantage of a physical draw is that you can see it. The drawback is that physical bias in the equipment, such as differences in ball weight, can affect results, which is why procedures that guard against bias matter.

The Korean Lotto draw procedure (as reported)

Donghaeng Lottery states that Lotto 6/45 is drawn at about 20:35 every Saturday, subject to change by the broadcaster. The detailed preparations are described most fully in a news report from the field at the round 1147 draw on 23 November 2024. We quote that report and mark the points press report, since we could not open the operator's own written procedure.

Ball size and weight checks: the draw balls are measured with gauges for each round. The reported standard weight is 4 g with a tolerance of ±5% (3.8–4.2 g) and the standard circumference 44.5 mm with ±2.5% (43.4–45.6 mm).
The observers' role: observers witness the procedures. Of five prepared ball sets, the observers themselves choose one for the broadcast draw and two as backups, check that the balls are properly registered in the system, test them in the machine, and watch the balls being taken out and displayed after the final rehearsal. Police officers are also said to attend the preparation.
Equipment tests and spares: before broadcast, the main machine and two backup machines are each tested three times, nine tests in all, and after the sales cut-off (8 p.m.) there are two rehearsals under live conditions. If a machine fails during the broadcast, a backup replaces it.
Storage and security: the storage room has CCTV, double locks and seals; temperature and humidity are checked before opening, and a security log records checks such as door locking and patrols.
Equipment and live broadcast: the machine is reportedly a product of the French company AKANIS TECHNOLOGIES, which says it is used in about 40 countries (the company's statement, unverified by us). The draw is broadcast live.

The structure maps onto the four requirements. Checking ball dimensions and testing equipment repeatedly target physical bias; observers choosing the ball set target manipulation in advance; CCTV and seals target tampering in storage; the live broadcast targets verifiability. But this list is "what was confirmed in one report", and we could not confirm that every round follows the same procedure, or which body supervises it. It was also reported that after suspicions of manipulation in 2023, public live draws with recruited observers began.

What statistics can check

With published results, anyone can check some things. Here are the checks this site ran on rounds 1 to 1247.

Are number frequencies even? The chi-square of per-number counts was 28.8. Among 300 fake histories simulating a fair draw the 5th–95th percentile range was 27.7–51.8, and the real value was near its lower end. There is no trace of some numbers appearing especially often; the real draw is, if anything, slightly more even (hot and cold numbers article).
Do earlier results affect the next? Numbers that appeared three or more times in the previous 10 rounds came up next at 13.62%, equal to the expected 13.33%.
Do winner counts match sales? Across rounds 88–1247, actual winners in prizes 1 to 5 were 0.995 to 1.003 times the expectation computed from sales. If there were an odd relationship between the numbers drawn and the games sold, this ratio would have been off.
The shape of the numbers: the shares of odd/even splits, sums and consecutive numbers matched theory.

What these checks say is "the statistical properties of the drawn numbers do not conflict with a fair draw". What they cannot say is a lot. For example, secretly changing one particular round, or inserting a bias so small as to be statistically invisible, is not caught by such tests. Statistical tests are tools for finding defects, not for guaranteeing there are none.

A case: 664 second-prize winners in round 1057

In round 1057 on 4 March 2023 there were 664 second-prize winners. By this site's calculation from that round's sales, the expected number was about 82.9, so the actual was more than eight times higher. The standardised residual (actual − expected divided by the square root of expected) is 63.8, the largest of the 1,160 rounds we analysed. A second prize means five numbers plus the bonus number matched; the numbers in this round were 8, 13, 19, 27, 40 and 45 with bonus 12. The same round's first-prize count, meanwhile, was 17 against about 13.8 expected, unremarkable.

Such a result makes suspicion easy. How should it be read? What this site can say:

The number alone neither proves nor rules out manipulation. The second-prize count is a matter not of "the numbers drawn" but of "how many people chose them". That people do not choose evenly we saw earlier in the overdispersion of first-prize counts (variance 1.73 times Poisson, and 6.2 times for second prizes). If some combinations were heavily chosen, second prizes can pile up when numbers close to them are drawn.
The government's explanation points the same way. The press, citing the Ministry of Economy and Finance's explanatory statement, reported that 609 of the 664 second-prize games were manually chosen combinations and that if buyers chose evenly about 83 second-prize winners would be expected, the same as this site's sales-based 82.9. The reports also said that of 103 games from one retailer, 100 of 102 manual games were sold on the same date and time and were presumed to be bought by one person. These are reported explanations and not something this page verified independently.
Which numbers were heavily chosen is not published. So this page cannot check whether the explanation is right. Doing so would need the distribution of numbers on the games sold.

What this case gives is not a conclusion but the form of the question. "What is the probability that this is chance?" is the wrong question if it assumes people choose uniformly and ignores real clustering; once clustering is allowed, extreme values become more common. That does not mean the suspicion is mistaken. Dispelling it needs both transparent procedures (the observers above) and publication of data that outsiders can check.

Checks anyone can do

Download the published round results and compare each number's frequency and the sums with the range expected of a uniform draw. The benchmark should be "the scatter that a uniform draw would produce", not "exactly equal".
Winner counts mix in clustering of choices, so for judging the draw's fairness use the aggregate comparison with sales-based expectation, and read individual unusual rounds separately.
For procedure, the most direct method is to see it yourself through an observer programme or the public broadcast.

Other countries

This page covers only Korean Lotto. Oversight of draws elsewhere (independent auditors attending, pre-draw tests and so on) could not be confirmed from official sources and is not described. We note only that it may differ by country.

Limits

The procedure description rests on one field report (round 1147) and is not the operator's official procedure document. It does not guarantee that other rounds follow the same steps.
The government's explanation is as reported and was not independently verified.
Statistical checks can find only a few known defects and are not a proof of no manipulation.
Without the distribution of numbers chosen in round 1057, we could not establish the cause of the second-prize surge.

The conclusion is two sentences. The statistics of the drawn numbers did not conflict with a fair draw. But what makes people believe in fairness is not statistics but the combination of procedures one can see and data one can check. Whatever numbers you choose, the chance of winning is the same, and this page has no information on how to choose numbers to win.

LottoFlick is independent and not an official lottery operator. Numbers are random and cannot be predicted.

Sources and verification

Donghaeng Lottery: Lotto 6/45 page (draw about 20:35 every Saturday, subject to broadcaster) — verified
The Fact, 2024-11-24: field report on the round 1147 draw (ball gauges, observers, five ball sets, nine machine tests, two rehearsals, storage security, machine maker) — press report · a news report of one draw; the operator's written procedure was not opened
Newsis, 2023-03-06: the Ministry of Economy and Finance's explanation of round 1057 (609 of 664 second-prize games were manual picks; about 83 expected under uniform picking; live broadcast, police and public observers, seals) — press report
The Fact, 2023-06-10: public observers at a draw after the March controversy (664 second-prize winners; one person reportedly won with 100 tickets) — press report
LottoFlick aggregate output, including second-prize counts by round — computed by LottoFlick
LottoFlick aggregate output: number-frequency checks — computed by LottoFlick