Home → Guides → How scrambles work

How WCA scrambles work

Guide · about 8 minutes

A competition scramble is not somebody shuffling a cube. It is a generated object with specific fairness properties, and the way it is generated has consequences for what your practice should look like.

Why not just shuffle it

The obvious approach — turn random faces twenty-five times — is what most people imagine, and it has a subtle problem.

Applying random moves does not sample the cube's states evenly. Some states become much more likely than others, and, more importantly, the results are biased in ways that correlate with solving. Short random sequences leave recognisable structure behind: pieces that started adjacent tend to stay near each other, and a solver who has internalised that pattern gets an edge that has nothing to do with skill.

For a competition that ranks people worldwide, that is not acceptable. Two solvers should face the same difficulty distribution, and a scrambling process should not quietly favour anyone.

Random state

The alternative inverts the process. Instead of picking moves and seeing where you land, you pick the destination first.

1. Choose a state uniformly at random from all the puzzle's legal states.
2. Compute a move sequence that reaches it from solved.
3. That sequence is the scramble.

Every reachable state is exactly as likely as every other. There is no residual structure, because the moves were derived from the state rather than the other way round.

This is why 3x3 competition scrambles are short — around twenty moves, sometimes fewer. A short scramble sequence feels wrong to people expecting a long shuffle, but it is a consequence of the method, not a shortcut: the solver computes an efficient route to a genuinely random state. A 19-move scramble from a random-state generator is considerably more scrambled than 40 random turns.

The hard part is step 2. Finding a short sequence to an arbitrary state is a substantial search problem, which is why random-state scrambling arrived late and is still not used for every event.

Which events get which

WCA scrambles are generated by TNoodle, the official scrambling software. Random-state generation is used where it is computationally feasible — which covers the 3x3x3, 2x2x2, 4x4x4, Pyraminx, Skewb, Square-1 and Clock.

The larger puzzles — 5x5x5, 6x6x6, 7x7x7 — and Megaminx use random moves instead. Their state spaces are far too large to search for a short solution in reasonable time. Instead they use long move sequences: many dozens of turns, enough that the residual bias is negligible in practice.

This is the reason for the length difference you have already noticed. A 3x3 scramble is around twenty moves and a 7x7 scramble is a wall of text. It is not that the bigger puzzle needs more scrambling in some intuitive sense — it is two different algorithms solving two different problems.

Everyone gets the same scramble

Within a round, every competitor solves the same set of scrambles. Attempt one is the same scramble for everyone in the round; so is attempt two.

This is the core fairness property, and it is why the scrambling procedure at a competition is as controlled as it is. Scrambles are generated in advance and kept secret. A designated scrambler applies them out of your sight. You do not see the scramble until you sit down, and you are not permitted to see other people's attempts of a scramble you have not yet done.

It also means a lucky scramble is not unfair. If attempt three happens to give an easy cross, it gives everyone an easy cross. The luck is in the round, not in your copy of it — which is exactly why "I got a good scramble" is not a caveat anyone needs to add to a competition result.

What "lucky" actually means

Since states are sampled uniformly, every scramble is equally likely. But "equally likely" is not "equally difficult", and this is where intuition misleads people.

Some states genuinely are easier to solve. An OLL skip is a real thing that will occasionally happen to you. So is a scramble with three F2L pairs nearly built. These are not the generator being generous — they are simply rare states, and rare states occur at their natural rate.

The useful consequence: you cannot practise your way out of scramble variance, and you should not try. Over five attempts you will get a spread. The average of five format exists precisely because of this — dropping the best and worst removes both the lucky scramble and the disaster, leaving something closer to your actual ability.

Which is also why one great single says much less about a solver than a consistent average does.

What this means for practice

Do not filter your scrambles

Skipping a scramble because it "looks awkward" corrupts the thing you are measuring. Your average over scrambles you liked is not an average, and it will not survive contact with a competition where you get what you are given.

This is the most common self-inflicted wound in speedcubing practice, and it is entirely avoidable: solve every scramble you generate.

Use scrambles from the same distribution

If your practice scrambles come from a generator with different properties than TNoodle's, you are practising a slightly different puzzle. In particular, hand-shuffling or short random-move sequences produce systematically easier and more structured states than competition scrambles.

Practise the whole distribution, including the bad end

People remember lucky scrambles and forget hard ones, then are surprised when a competition round contains two awkward ones. It did not: your practice contained them too, and you did not count them.

Why scramble handling is so strict

Everything a competition does around scrambles follows from the same requirement. The scramble must be secret until you sit down, identical for everyone, and applied correctly.

Hence: scrambles generated in advance and kept out of sight, a separate scrambler rather than the competitor, puzzles covered between scrambling and inspection, and the convention that you do not discuss a scramble with someone who has not attempted it yet.

None of that is ceremony. Each rule closes a specific way the fairness property could be broken.

Practising it properly

The competition simulator generates its scrambles with the same random-state approach TNoodle uses, gives you the round's scrambles in order, and does not let you skip one you dislike. Combined with the enforced inspection and the scorecard, that gets practice conditions genuinely close to the day.

The timer uses the same generator for ordinary sessions. If you take one thing from this page, take the discipline: generate it, solve it, record it. Every time.