What it looks for
Identifying a cipher by eye is a process of elimination, and the tool follows the same steps a person would.
1. The characters
A message made only of dots and dashes is Morse. Only ones and zeros: binary. Numbers no higher than 26: A1Z26. A solid block of mixed-case letters and digits: Base64. These formats give themselves away and are simply decoded.
2. The letter statistics
For a message made of ordinary letters, the first question is whether some letters are much more common than others. The index of coincidence answers it with one number, shown on the scale beneath the result.
- Near 0.067, like English: the letters have been relabelled or reordered, but each one still stands for a single letter. That means a simple substitution (including Caesar and Atbash) or a transposition such as rail fence.
- Between about 0.045 and 0.055: flatter than English. Typical of Vigenère with a short keyword.
- Near 0.038: flat. A long-keyed polyalphabetic cipher, a machine cipher, or modern encryption.
3. Trial by breaking
Statistics narrow the field; they don't name the cipher. So the tool runs its breaker for each candidate and scores what comes out. If several ciphers produce a clean reading, the simplest explanation is preferred: a Caesar shift is reported as a Caesar shift, even though a general substitution solver could also have cracked it.
Questions people ask
How does the cipher identifier work?
It first looks at which characters the message uses, which is enough to recognise Morse, binary, Base64 and number codes. For messages made of letters it measures how uneven the letter frequencies are, then actually attempts to break the message with each plausible cipher and reports whichever attempt reads most like English.
Which ciphers can it identify?
Caesar and ROT13, Atbash, affine, simple substitution (cryptograms), Vigenère, Beaufort, autokey, rail fence, scytale, columnar transposition, Morse code, tap code, binary, hexadecimal, decimal character codes, Base64, A1Z26, the Polybius square and the Baconian cipher.
How long does the message need to be?
Encodings such as Morse and binary are recognised at any length. Caesar shifts need only a few words. Vigenère and substitution ciphers need more to work with: aim for at least a hundred letters, and more for long keywords.
Why did it get my message wrong?
The usual reasons are that the message is too short for statistics to work, that it is not in English, that it uses a cipher outside the list above (Playfair, Enigma and modern encryption are not covered), or that two ciphers have been layered on top of each other.
What is the index of coincidence?
The probability that two letters picked at random from the text are the same. For English it is about 0.067, because some letters are much commoner than others. For evenly random letters it is 0.038. Simple substitutions and transpositions leave it at the English value; polyalphabetic ciphers pull it down toward random.