Find the code from how wrong you were
Each guess comes back with two numbers: how many symbols are right and in the right place, and how many are the right symbol somewhere else. Nothing says which is which. Two modes — probe a hidden code, or work one out from a table where exactly one answer fits.
The misplaced count, and why it is the thing that goes wrong
The second number counts overlaps, not symbols present. Guess A A B B against the code A B C D and you are told one exact and one misplaced — not two, even though both A and B appear in the code, because the code contains only one of each and your guess has two.
The tempting shortcut is to ask, for each symbol in the guess, whether it appears anywhere in the code. That over-credits repeats and produces feedback that no code can satisfy. It matters more than it sounds: a player deducing carefully will reach a contradiction, and will be right to conclude the game is broken rather than that they made a mistake. The scorer here counts each symbol the smaller of the two tallies and subtracts the exact matches, and the test suite checks precisely the cases where the two versions disagree.
What each mode actually promises
Probe a hidden code is the ordinary game. What is guaranteed is that the feedback is always truthful and consistent — not that every guess narrows things to one answer, which is not how the game works and would not be worth playing if it did.
Work one out is a constructed puzzle: a table of guesses already scored, where exactly one code fits all of them. That is proven by enumerating every possible code and counting how many survive — and then every row is removed in turn and put back only if the answer stops being unique, so nothing in the table is padding. A clue you could delete without changing the answer was never doing any work.
That proof needs the whole code space enumerated, so the mode is offered only where it can be. At six places and ten symbols there are a million codes; the promise could not be checked there, so it is not made, and the button says so rather than quietly producing something weaker.
How to get better at it
Open with a guess you know is wrong. A first guess of A A B B tells you more than a plausible-looking one, because it splits the field of codes into large distinct groups rather than confirming a hunch. The measure of a good guess is not how likely it is to be right — it is how small the largest remaining group is once the answer comes back, whatever that answer turns out to be.
Count first, place second. On the standard setting, working out how many of each symbol the code contains takes about four guesses and reduces what is left to a handful of arrangements. Trying to pin down positions before you know the counts means every new fact has to be re-checked against arrangements you had no business considering.
When a guess comes back with the same two numbers as an earlier one, that is information, not a wasted turn — the two guesses differ somewhere, and the code is indifferent to the difference.
What the hint does
Not hand you a symbol, at least not until the third press. In a game this short that ends the puzzle rather than helping with it. The first press counts how many codes still fit everything you have been told. The second suggests a probe and states what it achieves in the worst case — often a guess that cannot possibly be the answer, which is usually the most useful thing to play.
If the rows you have been given fit no code at all, it says that instead. On a generated puzzle that means a row was misread rather than that the puzzle is wrong, and knowing which of the two you are looking at saves a great deal of time.
Colour is never the only difference
Every symbol carries its letter as well as its tint, and each has a name that a screen reader announces. A code game distinguished only by hue is unplayable for a large number of people and prints as a row of identical grey circles.
Related
Logic Grids for deduction from written clues, Minesweeper for deduction under a counting rule, or the puzzle hub for everything else.
Questions people actually ask
What exactly does "misplaced" count?
Overlap between the two multisets, minus the exact matches. Guess A A B B against the code A B C D and you get one exact and one misplaced — not two — because the code holds only one A and one B while your guess holds two of each. The shortcut of asking whether each guessed symbol appears anywhere in the code over-credits repeats and produces feedback that no code satisfies.
Does every guess narrow it down to one answer?
No, and a game where it did would not be worth playing. What is guaranteed in the probing mode is that the feedback is always truthful and consistent with the hidden code. The "work one out" mode is different: there, exactly one code fits the table, and that is proven by enumerating every possible code and counting the survivors.
Why is "work one out" unavailable on the largest setting?
Because the promise could not be checked there. Six places and ten symbols is a million codes, and proving that exactly one of them fits means examining all of them. Rather than quietly offering a weaker version, the mode says so and stays off.
Are the rows in a constructed puzzle padded out?
No. After the table is built, every row is removed in turn and put back only if the answer stops being unique. A clue you could delete without changing the answer was never doing any work, and a table full of them reads as a wall of numbers rather than a puzzle.
Does the hint just tell me a symbol?
Only on the third press, and then only one place. The first counts how many codes still fit everything you know. The second suggests a probe and says what it achieves in the worst case — frequently a guess that cannot possibly be right, which is often the most informative thing to play.
Can I play without seeing colour?
Yes. Every symbol carries its letter as well as its tint and has a name a screen reader announces, and the whole game can be played from the keyboard by typing letters. A code game told apart only by hue is unplayable for a lot of people and prints as identical grey circles.
Is anything sent anywhere?
No. The code is generated by your own browser and stays there, along with your guesses. There is no account and no server.