When all this failed, other ideas emerged. The numbers were between 3 and 215, which made us interested in the crossword shown in The Code Book. The crossword has 15 x 15 = 225 squares; surely this cannot be a coincidence! However, no matter how we counted, the 4 times repeated index 109 always became a filled square, and no message emerged.
At this point we decided to stop being intelligent, and start using brute force. Several programs were written. A first version of our program tested all possible offsets in a text. For each offset the program extracted the first, last, or kth (for all possible choices of k) letter in the magic words. We also built a similar program which instead of counting words counted letters (including or excluding whitespace and separators). The possible plaintexts were then written on a file which was scanned with the Unix command grep for probable keywords like ``stage'' and ``word''. What we learned from this was that given enough amount of keytext, seemingly improbable amounts of candidate plaintexts may emerge. Some of the messages we found scanning the King James version of the Bible would have made Michael Drosnin jump with joy.
Our next step was to make a fully automatic version of the program. To obtain a fully automatic program, we created an evaluation function that evaluated each potential plaintext. The evaluation function uses a large English dictionary. Candidate plaintexts with high enough evaluations were output by the program for visual inspection.
The two programs (the word counter and the letter counter) were then run on various texts. The Project Gutenberg provided us with several hundred MB of text. A small subset of the texts that were tested is:
Not only electronic texts were examined. One group member hunted the Stockholm libraries for the French version of ``A void''. We also had advanced plans for scanning The Code Book, and ``Fermat's last theorem'', also by Simon Singh. Singh's PhD thesis was also searched for. After all, the two first solvers that we heard of had connections to Cambridge and might therefore have had access to this thesis.
Alternative solution methods were also discussed. We actually have to admit that during a beer drinking session, we came up with a scam that could be used to get the solution. The CipherChallenge email list was now buzzing with people announcing their success with Stage 5. What if we said that we had solved Stage 5, and volunteered to confirm the solutions of others... Luckily, we never needed to resort to getting the scam, since Lars announced the successful solution some days later.