A night sky made of lottery numbers

Why the sky is a solid web at zero

Turn the dial all the way down and every pair is an edge. That is not a discovery about Bank a Million — it is arithmetic.

The chance two particular numbers land in the same draw is $p_2 = k(k-1)/\big(N(N-1)\big)$, and the chance they have never met across $D$ draws is:

$$P(\text{never together}) = (1 - p_2)^D$$

For a 6/49 game $p_2 \approx 0.01276$, and that probability falls below 1% after about 360 draws. Here is the decay: — by the time a lottery has a few hundred draws behind it, essentially every pair has "a connection". A fact true of all 780 pairs tells you nothing about any of them.

What the dial is actually doing

Each edge carries the same deviation as the Harmony Matrix:

$$E = D \cdot \frac{k(k-1)}{N(N-1)}, \qquad z = \frac{O - E}{\sqrt{E}}$$

For this era $E =$ 22.2 per pair. The dial sets the minimum $|z|$ an edge needs to stay on screen, and the counter beside it shows how many should survive under pure chance.

The textbook figure for that is $\binom{N}{2}\cdot 2\Phi(-t)$ — the normal tail. We do not use it. These counts are small, and at that size the normal curve is a poor stand-in for a discrete, non-negative count: for a 5/69 game it predicts 29 surviving edges at $t = 2.5$ where a fair machine actually leaves about 19. Overstating the expectation would make a real lottery look quieter than chance, so the counter uses the exact Poisson tail instead.

Watch the two numbers as you turn the dial. They fall together, all the way down.

A graph is thousands of hypothesis tests wearing a trench coat

This sky shows 780 pairs at once. Every edge is a separate question — "do these two go together?" — and asking that many questions guarantees striking-looking answers.

That is the same trap as the Harmony Matrix, but a network hides it better. A grid of cells looks like a lot of data. A constellation looks like a finding.

Triplets, and why they are computed on demand

Select two stars and the drawer asks which third number joined them. The baseline is:

$$E_3 = D \cdot \frac{k(k-1)(k-2)}{N(N-1)(N-2)}$$

which for this game works out at 2.338 appearances for any specific triplet across the whole era. For 6/49 it is $D \cdot 120/110{,}544 \approx 0.001086 \cdot D$ — about one appearance per thousand draws.

There are 9,880 possible triplets in a pool of 40. Storing a table of them for every game and every rule era would be a combinatorial explosion in service of a curiosity, so the drawer computes its answer live from the draws themselves.

Constellations are the original apophenia

The stars in Orion are not near each other. Betelgeuse is around 550 light years away; Rigel is closer to 860. They share no history, no motion, no connection of any kind. What they share is a line of sight from one small planet, and a species that cannot look at scattered points without joining them up.

Every culture did it, and every culture drew something different. The seven stars Europeans called the Plough were a bear to some peoples, a wagon to others, a coffin followed by mourners elsewhere. The Chinese saw a government minister's chariot. The same photons, arriving at the same eyes, resolved into entirely different pictures depending on who was looking.

The sky never contained a hunter. We did.

This page does to lottery numbers exactly what our ancestors did to stars — and then hands you the one thing they never had: a dial that asks how much evidence you would need before you believed the shape was real.

Why randomness looks clumpy

Kahneman and Tversky named the underlying error in 1972: the representativeness heuristic. We judge whether something is random by whether it looks like our idea of random — and our idea of random is an even sprinkle.

Real randomness is not an even sprinkle. It clusters, streaks and leaves holes. So genuine randomness reads to us as suspicious, and a suspiciously tidy arrangement reads as fair. It is precisely backwards, and it is why the sky at low thresholds looks so much like it means something.

Thomas Gilovich spent a career on the consequences, including the hot hand in basketball: players, coaches and fans all saw streaks in shooting records that statistically were not there.

Complete disorder is impossible

There is a mathematical version of this, and it is stronger than a warning about perception.

Ramsey theory proves that in any sufficiently large structure, order must appear. Colour the connections between six people by whether they are friends or strangers, and you cannot avoid creating three mutual friends or three mutual strangers. It is not likely — it is forced. Theodore Motzkin summarised the field as: complete disorder is impossible.

So when the constellation finds shapes at a low threshold, that is not evidence of anything. A large enough random graph is required to contain them. Finding structure proves the graph is big, not that the lottery is crooked.

If an edge ever survived

Somewhere, eventually, even whole draws repeat — Bulgaria's lottery produced the same six numbers two draws running in September 2009, and the country briefly lost its mind before the mathematicians pointed out how many lottery draws happen worldwide.

But if a single edge here held at 3.5 across multiple eras, that would be different. It would not be a lucky pair. It would be a machine with something wrong inside it — and that is a story for the fraud files.