A draw is not a set of numbers.

Stars and bars

Lay out the 36 positions of the pool as a row. Choosing 6 of them leaves 30 empty positions, and those empties fall into 7 runs: one before the first chosen number, one between each neighbouring pair, and one after the last.

Call those runs $g_0, g_1, \dots, g_k$. They always add to the same total:

$$g_0 + g_1 + \cdots + g_k = N - k$$

So a draw is not really a set of numbers. It is a way of sharing 30 identical empty positions among 7 labelled runs — and the classic count of those sharings is

$$\binom{(N-k) + k}{k} = \binom{N}{k}$$

which is exactly the number of tickets. The same object, viewed sideways.

Why small gaps dominate

Fix one run at $g$ and the remaining $N-k-g$ empties are shared freely among the other $k$ runs. Counting those arrangements gives the marginal directly:

$$P(g_i = g) = \frac{\binom{N-g-1}{\,k-1\,}}{\binom{N}{k}}$$

A delta — the distance between two consecutive numbers — is just a run plus one, so $P(\text{delta} = d) = P(g = d-1)$. That binomial shrinks as $g$ grows, which is the entire fuel of the Delta System: there really are more ways to build a ticket out of small gaps. Small gaps are not favoured by the machine. They are favoured by arithmetic.

The mean follows:

$$E[g_i] = \frac{N-k}{k+1}, \qquad E[\text{delta}] = \frac{N+1}{k+1}$$

which for Powerbucks Wheel of Fortune is 5.29.

The twins, without a calculation

Here is the part that surprises people. Nothing in that count distinguishes $g_0$ from $g_3$. The runs are labelled, but the counting treats them identically — swap any two and you get another valid sharing of the same total. They are exchangeable: all 7 of them, boundary runs included, have one and the same distribution.

The space before the first number is therefore a distribution-twin of the space between any two numbers. And since the first number is that leading space plus one,

$$E[\,n_1\,] = E[\text{delta}] = \frac{N+1}{k+1}$$

The first number and the gaps live in the same place because they are the same kind of object. No calculation is needed — only the symmetry.

A transform cannot create probability

The delta map is a bijection. Given a ticket you can compute its deltas; given the deltas and the first number you can rebuild the ticket, uniquely. Two descriptions, one object.

That settles the system without any statistics. If every delta pattern corresponds to exactly one ticket, then choosing a delta pattern is choosing a ticket, and it carries that ticket's odds — 1 in 1,947,792. This holds for any invertible re-description: deltas, digit sums, mirror numbers, birth dates rearranged. Re-labelling the tickets never changes which one comes out of the machine.

The honest twin

None of this makes spacings useless — quite the opposite. Comparing observed gaps against exactly these expectations is a genuine randomness test. Knuth sets it out as the gap test in The Art of Computer Programming, Vol. 2, and it is still used to certify random number generators.

Identical mathematics; opposite purpose. The gap test asks whether the machine is honest. The Delta System asks how to beat it, using a fact that applies equally to every ticket, and so answers a question it has not asked.

The system that was half right

Through the 1990s and long after, a method circulated among lottery players called the Delta System, spread by forum posts and a piece of software called Analysis Lotto. The pitch was unusually concrete: stop looking at the numbers, look at the distances between them. Do that, and a pattern jumps out — winning tickets are built overwhelmingly from small gaps.

That is a real observation. It is also completely correct. Pull any list of past winning draws, compute the gaps, and they will indeed cluster low.

What the method never did was compute the same thing for the tickets that lost. Had it done so, the pattern would have been there too, in precisely the same proportion, because it is a fact about how many ways there are to arrange numbers on a line and not a fact about lotteries at all. The system found the structure of counting and mistook it for the structure of luck.

It is worth being fair about why it convinced. Anyone who checks winners and finds a strong, stable, reproducible pattern has found something — the instinct to look was right, the observation was accurate, and the pattern is genuinely there. The single missing step was the control group.

One lemma, many worlds

The counting argument behind all of this — stars and bars — is a small classic, set out in Feller's An Introduction to Probability Theory and Its Applications. It answers a question of the form: in how many ways can you share $n$ identical things among $r$ labelled boxes?

That question turns up everywhere. It counts the ways a lottery draw can cut the number line. It counts the ways to split a pile of identical coins among several people. And in physics it counts the states available to indistinguishable particles — the reason a boson gas behaves as it does. The lottery spacings and the quantum statistics of light are, at the level of the counting, the same problem wearing different clothes.

The compliment that spacings deserve

There is a version of this idea that professionals take seriously. Knuth's gap test — the same expectations, the same arithmetic — is a standard tool for checking whether a random number generator is telling the truth. Spacings are rigorous enough that computer science trusts them to certify randomness.

That matters here, because rigged machines do leave statistical fingerprints. When Eddie Tipton compromised the Hot Lotto generator, what eventually made the case legible was that the numbers a tampered generator produces are not distributed the way honest ones are. Spacing-style tests belong to the family of tools that catch such ghosts.

So the spacings really are informative. They are just informative about the machine, not about the next draw.

What they actually found

The Delta System's believers were right that there is structure hiding in the gaps. There is — a beautiful, exact, entirely knowable structure, and this page is largely a tour of it.

They simply found the structure of counting, not the structure of luck. It was never a signal from the machine. It was the shape of the question.