CipherClerk's Applet derives the arrangement of the cipher discs from the given key word: This is repeated until it's length is 25 letters. This alphabetic key is converted into a numerical key by using the "standard" procedure. The position where the cipher text is read is obtained from a random number generator initialized using the date and time of the encryption.
When deciphering, CipherClerk's Applet must determine where to read the plain text. It assumes that the plain text is english, exactly speaking: it's letter frequencies are those of an english text. Since this analysis can only be done on groups of 25 letter, it is far from being perfect. CipherClerk's Applet would most likely not be able to decipher a final block of just a few letters. For this reason, it assumes for decrypting - and assures when encrypting - that such groups will use the offset of the group before.
I've written another simulation of this device, offering a look & feel close to the original. Definitely worth a try!