On paper, a colour sudoku is a reskin. Take the nine digits, swap them for nine colours, keep every rule identical. The solver logic doesn't change. The generator doesn't change. You could ship it in a weekend.
In practice it changes almost everything about how the puzzle plays, because digits and colours are read by different parts of the brain. Digits are symbols — you read them serially, one cell at a time, and you can hold "7" in your head while you scan. Colours are perceptual — you take in a whole region at once, and you compare rather than recall. That difference makes a colour puzzle easier in some places, harder in others, and inaccessible to a meaningful slice of players unless you design for it.
Scanning gets faster, recall gets worse
The first thing playtesters notice is that finding a duplicate in a row is quicker with colour. You don't read nine cells; you glance and the repeat pops. That's genuine — the visual system is very good at detecting a repeated hue in a small field.
The second thing they notice, usually a few minutes later, is that the harder deduction techniques get worse. Anything that requires holding a candidate set in working memory — "this cell is a 3 or a 7, and if it's 3 then that one must be 9" — is meaningfully harder when the candidates are "teal or magenta" rather than "3 or 7". Named symbols are easier to reason about than remembered hues.
The design consequence is that a colour board and a number board with the same generated difficulty rating do not feel the same. Easy puzzles feel easier. Hard puzzles feel harder and more frustrating. If you ship a colour variant using a difficulty curve tuned for digits, you get a game that's boring for the first hour and a wall after it.
What we did about it
We stopped treating the generator's difficulty score as the difficulty. The generator rates a puzzle by which solving techniques are required to crack it — that's a property of the grid, and it's still useful. But the curve players actually experience got retuned separately: more gentle progression in the middle band, and a slower introduction of puzzles that need multi-step candidate chaining, because those are the ones colour makes disproportionately harder.
Nine colours is more than you think
Choosing nine colours that are all mutually distinguishable, at small sizes, on phone screens of wildly varying calibration, in bright sunlight, is a real constraint and not a styling exercise.
A few things that made the palette work:
- Vary lightness, not just hue. If all nine colours sit at the same lightness, they collapse into mush for anyone with reduced colour discrimination and also for anyone outdoors. Spreading lightness across the set means the board is still readable in greyscale — which is the fastest test you can run.
- Avoid adjacent hues at similar saturation. Two greens is one green too many. We ended up with fewer, more separated hue families and used lightness to get the count up.
- Test at real cell size. Colours that are obviously different on a swatch card can be indistinguishable in a 32-point square with a border around it.
Colour alone is never the whole answer
Somewhere between four and eight percent of men have some form of colour vision deficiency, depending on the population studied — most commonly red-green. For a game whose entire mechanic is colour identity, that isn't an edge case to handle later. It's a design input.
The approach that worked was redundant encoding: colour is the primary channel, but it is never the only channel available. Options that let a player turn on a small symbol, a pattern, or a numeral inside each cell mean the same board is solvable through a second channel without changing the puzzle at all. Crucially these are settings, not a separate mode — a separate "accessibility mode" tends to feel like a lesser version of the game, and people don't turn it on.
The greyscale test is the cheapest version of this discipline. Screenshot the board, desaturate it, and try to solve it. If you can't tell two cells apart, neither can a real player on a bad screen.
Input is where colour puzzles get lost
With digits, input is solved: you tap a cell, you tap a number, done. With colours you have to invent the palette control, and it's the single biggest source of playtest complaints.
Two models compete. Cell-first — select a cell, then pick a colour — matches the digit convention and is easy to explain. Colour-first — pick a colour, then paint every cell that takes it — is faster for experienced players because it matches how people actually solve: you hunt for one value at a time.
We support both, with colour-first as the mode that survives contact with expert players. The important part isn't which one you choose; it's that the currently selected colour is unmistakably visible at all times. Every "I don't know what's happening" playtest moment traced back to a player who had lost track of what was selected.
Two small things that mattered more than expected
- Highlight all matching cells on selection. Selecting a colour dims everything else and makes the existing placements glow. This turns a scanning task into a glance and is the single biggest quality-of-life feature in the game.
- Undo, unlimited and instant. Colour puzzles invite experimentation in a way digit puzzles don't, because painting feels casual. If experimentation is punished, players stop experimenting and the puzzle gets less fun.
The onboarding problem
Everyone knows sudoku's rules. Almost nobody knows that they know them well enough to recognise the same rules in an unfamiliar coat. In early tests, players who had solved hundreds of number sudokus stalled on the first colour board — not because it was hard, but because they didn't immediately trust that it was the same game.
The fix was to say so. A single line — this is sudoku with colours instead of numbers, same rules — recovered more first-session completions than any tutorial we built. The lesson generalises: when your game is a familiar thing in unfamiliar clothes, the most valuable onboarding is naming the familiar thing out loud.
What we'd tell someone starting a colour variant
- Assume the difficulty curve needs its own tuning. Don't inherit it.
- Design the palette in lightness space first, hue second.
- Ship redundant encoding as settings, never as a separate mode.
- Solve input before you solve art. Colour-first selection with a loud, obvious current selection.
- Tell the player what game they're playing in one sentence.
None of that is exotic. It's just the work that a "simple reskin" quietly contains, and the reason a colour puzzle that's been through it feels completely different from one that hasn't.