The "Decipher Puzzle" paper is still on Alan Sherman's website, an amusing read. (Google does index the postscript file, but the browsers I use don't usually have PS viewers integrated.)

The puzzle was a book cipher, like the Beale ciphers, Thouless's "life after death" test, or Simon Singh's "Code Book" problem #5. It's possible problem #5 was the most difficult of the 10 "Code Book" problems.

The eight original clues were

"1. The key is in the public domain - you have easy access to it.
2. 3,19.
3. The first place to start may not be the first.
4. A cube is a clue if you are geometrically inclined.
5. 300 is closer to 'A' than 'Z'.
6. One source will do but what will you do with it?
7. Some numbers are repeated but do not be fooled.
8. If you knew it began with 'C', would it help you?"

(my comment: these clues seem pretty useless for finding the book - I would have taken "public domain" to mean "out of copyright")

The puzzle ($12 each) started in 1983 sometime, and the solution was in a safe-deposit box which was to be opened if no one solved the puzzle by 30 June 1989. Everyone who submitted a correct solution by 1 March 1984 split the $100,000.

"Holland's decision to put the ciphertext on a puzzle was clever - even if the contestant cannot discover the plaintext, he can almost certainly put together the jigsaw puzzle which lets him feel as if he has made significant progress."

"As for the key-extraction strategy, we reasoned that it was probably something that could be easily carried out by hand without the use of a computer - many people would consider it unfair to require contestants to use a computer."

Relevant paras for Kryptos like challenges ...

"We speculate that Holland intentionally designed the puzzle to be controlled by the clues. Doing so would accomplish the following three objectives. First: by forcing each contestant to solve the puzzle by guessing the keytext, Holland can control the difficulty of the puzzle by judiciously choosing the clues. To maximize public interest in this and future puzzles: the puzzle should be solvable yet the puzzle must not be too easy nor too hard.  A puzzle that is too easy would be solved too soon: thereby reducing sales; a puzzle that is too hard might discourage contestants from buying future puzzles. Second: Holland can ensure that the puzzle will eventually be solved by giving away sufficiently many additional clues. Third: by forcing contestants to guess the keytext, lay people and mathematicians are put on a roughly equal footing."

The creator Holland started releasing clues - first of all saying the keytext was a "popular novel" written by one of twenty-one specified authors, then eliminating two authors from the list each month - but the only one that was any use was the very last clue in February 1985, after which 36 people solved it in March 1985. (i.e. all of those people depended on that clue - reminds of "Masquerade", but there only 2 people understood the clue)

"When no one solved the puzzle by February 28, 1985, Holland announced that the key was a sequence of "first letters" from Cosmos [by Carl Sagan] chapter 6." (beginning "first ages of the world...", first letters of all words, treating hyphenated words as single words and deleting all numerals, abbreviations, footnotes and figure captions. "Cosmos" is certainly popular, but it is not a novel!!)

"Holland explained that he designed the clues to have multiple levels (heh!) of interpretation and to challenge people to think more deeply; for most clues the "obvious" interpretation is incorrect. For example, clue 5 means that codenumber 300 is closer to 'A' than 'Z' not in alphabetical order but in the sense that codenumber 300 represents a vowel rather than a consonant. Clue 4, which mentions a cube, refers to chapter 6 since a cube has 6 sides. Holland said that he had used the world 'novel' in the spirit of the clues to mean both 'book' and "new and unusual." The picture of a cube in Cosmos, chapter 10 and the references to the country Holland in chapter 6 were simply coincidences."

"Holland also said that he was surprised the puzzle was not solved within the first year." (now that sounds familiar)

"There are many lessons to be learned from our experience. One of the obvious lessons is "Read instructions carefully!" But beyond this obvious lesson: we would like to reflect on a number of points. 

First: throughout our cryptanalysis we found ourselves caught in the following dilemma: how can we tell how close we are to the solution? Although the windowing technique detects some near misses: and although interpretations of the clues provide some confirmation: for the most part it was virtually impossible to know whether or not we were near solving the puzzle. This made it difficult for us to decide how much time to dedicate to the puzzle."

Second: the computer's facility for doing tedious calculations is a double-edged resource. On the one hand: the computer was a tremendous aid in calculations such as trying all possible starting positions in a keytext. On the other hand: the computer posed a danger in inviting us immediately to follow the straightforward approach rather than to think about a possibly more effective clever approach. For example: to check a candidate key  the straightforward approach is to use it to decipher the entire ciphertext and to check the resulting candidate plaintext for English. But if the entire candidate plaintext is checked for English, poor results are obtained because this checking procedure is too sensitive to errors in the keytext. Someone solving the puzzle by hand would almost certainly check a candidate keytext without generating the entire key and without deciphering the entire ciphertext. By virtue of  his  limited  computational powers: the hand-solver is practically forced to use a superior method that detects when part of a candidate key is correct! We do not wish to imply that straightforward methods are necessarily bad; nor do we wish to imply that computers cannot perform clever feats. We simply point out that computers can make it convenient for their users not to think deeply.

Third: we found ourselves more interested in the process of our solution than in the solution itself. For example: we preferred to spend our time analyzing our test for English or adding features to our cryptanalysis engine rather than to run our engine on candidate keytexts. This situation contributed to our missing the deadline.

Cryptanalysis is a complex activity involving probability, statistics, algorithms, computer systems, software and hardware engineering, and general problem solving. We hope the reader can learn from and enjoy our experiences in this fascinating area."