The machine made no pattern.

Counting your way to a lane

What is the chance the i-th smallest number in a draw is exactly $x$?

Put $x$ on the table. For it to be the $i$-th smallest, exactly $i-1$ of the other numbers must come from below it, and the remaining $k-i$ from above. There are $x-1$ numbers below and $N-x$ above, so:

$$P(n_{(i)} = x) = \frac{\binom{x-1}{\,i-1\,}\binom{N-x}{\,k-i\,}}{\binom{N}{k}}$$

That single line generates every ribbon on this page. No simulation, no fitting — the lanes were fixed the day the rules were written.

The elegance drop

Take the mean of that distribution and something unreasonably tidy happens:

$$E[n_{(i)}] = \frac{i(N+1)}{k+1}$$

The lane means are evenly spaced. For Canada Lotto 6/49 they land at 7.1, 14.3, 21.4, 28.6, 35.7, 42.9 — each one exactly 7.14 further along than the last.

There is a reason, and it belongs to the spacings between numbers. A sorted draw's positions are the partial sums of the spacings between them, and all $k+1$ spacings are exchangeable — identical in distribution, boundary gaps included. Add up equal expected steps and you get evenly spaced expected positions. One counting idea has now produced neighbours, delta patterns, collection times and these lanes: four features, four costumes.

As a rough intuition, position $i$ lives near the fraction $i/(k+1)$ of the pool — formally $n_{(i)}/N$ approaches a Beta$(i, k+1-i)$ distribution for large , which is an approximation and labelled as one. The exact PMF above is what the ribbons draw.

Sorted lanes, flat slots

Now the part that matters.

Ask instead which number came out of the machine first. The answer is uniform: any of the $k$ numbers in the draw is equally likely to have been drawn first, so slot 1's distribution is flat on $1 \ldots N$. So is slot 2. So is every slot.

Same draws. Same numbers. One bookkeeping produces sharply stacked lanes and the other produces flat lines, and the machine has no idea which one you are using.

Sorting is a function of the set, and a function of a uniform set inherits its shape from the function, not from the source. The lanes are real, lawful, and entirely ours.

The trap this retires

"Position 1 favours small numbers, so pick small numbers for position 1."

Positions are not picked. They are assigned, afterwards, by sorting whatever you picked. There is no slot on the ticket labelled smallest; there is only the number you chose, and where it happens to fall once the six are put in order. Every positional system dissolves on contact with that definition.

How many tanks was Germany building?

In 1943 the Allies needed a number. Conventional intelligence — agent reports, aerial reconnaissance, prisoner interrogation — put German tank production at around 1,400 a month, and the invasion planning that followed depended on it.

A small group of statisticians had a different idea. Captured and destroyed German tanks carried serial numbers, stamped sequentially on gearboxes, chassis and engines. If the Germans numbered their tanks 1, 2, 3, … then the serials recovered from the battlefield were a sample, and the largest one seen carried information about how many existed.

This is the distribution in the top ribbon on this page. The largest number in a sample sits below the true maximum, and by a predictable amount: with $n$ items sampled, the gap above the largest observed value is on average about the same as the average gap between the sampled values. That gives the estimator

$$\hat{N} = m\left(1 + \tfrac{1}{n}\right) - 1$$

where $m$ is the largest serial seen and $n$ the number of tanks examined. It is the minimum-variance unbiased estimate, and it needs no assumptions about German industry at all — only that the tanks were numbered in order and that the ones the Allies met were not chosen for their serial numbers.

The statisticians' answer was about 246 tanks a month.

Intelligence said 1,400. The statisticians were off by a factor of five in the other direction, which is not a comfortable place to be during a war.

After Germany surrendered, the production records were recovered. The true figure was 245.

It remains one of the great verified triumphs of statistics — and it is powered by exactly the order-statistics mathematics that draws the top lane of these ribbons.

The archive that makes this page possible

Almost no lottery publishes the order the balls came out. The draw is broadcast, the numbers are recorded, and by the time they reach a results feed they have been sorted — because sorted is how a ticket is checked.

France is one of the exceptions: its Loto is published in the order the balls actually came out. A handful of other countries do the same. It is a small, unglamorous act of open data, and it is the only reason the comparison on this page can be made with real records instead of a simulation.

It is worth adding what that comparison cost to build. Most feeds in this collection appear to supply a draw order — the field is there, and populated — and for the great majority of games it is simply the sorted numbers copied across. Believing the field would have produced a page showing sorted structure while claiming to show ball order. Every lottery here is therefore tested before it is trusted, and most fail.

Structure from bookkeeping

This is the third time the playground has run into the same lesson.

The constellations were an artefact of where a threshold was drawn. The Delta System's discovery was an artefact of counting arrangements. The lanes on this page are an artefact of sorting.

None of the three is fake — the constellations are really there, the small deltas really do dominate, the lanes really are lawful. They are simply not facts about the lottery. They are the shape of the questions, and the machine has never heard of any of them.

The ribbons are real, the lanes are lawful, and the machine has never heard of any of it.