That solver indeed works really well. I have played around with it before. It is very likely also using the chi-squared statistic to determine the keyword. That is what you are looking for and it only works on plaintext ciphers.

Nope, the solver does not use any chi-squared statistic method. It solely relies on ngrams. This is true for determining the key length as well as finding the best key.

I see. How does it find the key length on transposed ciphers as Largo states? I guess it was not transposed thoroughly.

https://www.guballa.de/implementierung- … re-solvers

The solver works in a way that it starts using 1-grams, and if required, it is switching to 2-grams, or – in a subsequent step – to 3-grams. Now for 1-grams each column of the cipher is looked at independent from the other columns, i.e. the order of the columns does not matter. That’s why the correct solution is found, even if the columns are transposed. Using 2-grams would probably not work, as neighboring characters are teared apart by the transposition.

Now the solver simply tries different key length, and when the correct key length is tried, the corresponding clear text (still transposed) scores exactly like an "untransposed" (i.e. original) text when 1-grams are used.