In early May 2001 I had the opportunity to visit CIA Headquarters and see Kryptos up close. My CIA escort said that only two people had cracked the code, one person in the CIA and a "computer guy" in California. Apparently both had solved three of the four parts of the message but no one had decrypted the last 97 characters.

Always looking for any project that could divert me from accomplishing anything productive at my real job, I jumped at the challenge.

The first three sections were easy. A simple Google search turned up hundreds of links to further information on Kryptos. I was able to quickly find the encrypted text, hints from the CIA, and the decryption for the first 3 parts of the 4 part message. (It doesn't get much easier than that!)

However, even knowing the types of encryption, the keywords, and the plain text, it still took the better part of a weekend to figure out how it all worked, and to understand how it could be possible for someone to crack the code from scratch.

To prove to myself that I understood how it worked, I spent the next few weeks writing a Basic computer program that, provided with only the raw text, could indeed solve the first three parts of the message.

In the "small world" department, some of the documents I found in my Google search pointed out that the "computer guy" my CIA escort referred to was someone that I knew (or at least knew of), Jim Gillogly. Our paths had crossed a year ago when we were both working on the Cipher Challenge in Simon Singh's The Code Book. Jim's advice during the Cipher Challenge was invaluable. Details on how he cracked Kryptos can be found in an article he wrote for the September-October 1999 issue of The Cryptogram.

BTW, I highly recommend Singh's The Code Book for anyone interested in learning more about cracking codes. Attacking the problems at the back of the book is a great way to waste about a year of your life.

EMUFPHZLRFAXYUSDJKZLDKRNSHGNFIVJ
YQTQUXQBQVYUVLLTREVJYQTMKYRDMFD
VFPJUDEEHZWETZYVGWHKKQETGFQJNCE
GGWHKK?DQMCPFQZDQMMIAGPFXHQRLG
TIMVMZJANQLVKQEDAGDVFRPJUNGEUNA
QZGZLECGYUXUEENJTBJLBQCRTBJDFHRR
YIZETKZEMVDUFKSJHKFWHKUWQLSZFTI
HHDDDUVH?DWKBFUFPWNTDFIYCUQZERE
EVLDKFEZMOQQJLTTUGSYQPFEUNLAVIDX
FLGGTEZ?FKZBSFDQVGOGIPUFXHHDRKF
FHQNTGPUAECNUVPDJMQCLQUMUNEDFQ
ELZZVRRGKFFVOEEXBDMVPNFQXEZLGRE
DNQFMPNZGLFLPMRJQYALMGNUVPDXVKP
DQUMEBEDMHDAFMJGZNUPLGEWJLLAETG
ENDYAHROHNLSRHEOCPTEOIBIDYSHNAIA
CHTNREYULDSLLSLLNOHSNOSMRWXMNE
TPRNGATIHNRARPESLNNELEBLPIIACAE
WMTWNDITEENRAHCTENEUDRETNHAEOE
TFOLSEDTIWENHAEIOYTEYQHEENCTAYCR
EIFTBRSPAMHHEWENATAMATEGYEERLB
TEEFOASFIOTUETUAEOTOARMAEERTNRTI
BSEDDNIAAHTTMSTEWPIEROAGRIEWFEB
AECTDDHILCEIHSITEGOEAOSDDRYDLORIT
RKLMLEHAGTDHARDPNEOHMGFMFEUHE
ECDMRIPFEIMEHNLSSTTRTVDOHW?OBKR
UOXOGHULBSOLIFBBWFLRVQQPRNGKSSO
TWTQSJQSSEKZZWATJKLUDIAWINFBNYP
VTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR

[Transcriber's Note: According to Sanborn, ID BY ROWS is an error that should read LAYER TWO when X is inserted after WEST.]

There were several things about the puzzle so far that I found intriguing:

• There were a number of highly visible clues that made it easier to crack the first parts of the message. The Vigenere table not only gave away the type of cipher (for parts one and two), it supplied one of the key words - KRYPTOS. At the end of message two is the phrase "ID BY ROWS." The text from that point on used columnar transpositions. Could that have been a clue? Could there be other clues?
• PALIMPSEST, the other key word to part one, seemed to be a very unusual word to have selected at random. A palimpsest is a document in which an original text was erased and overwritten with a new message. Could the Kryptos sculpture itself contains just such an overwritten text?
• Was there any significance to the three spelling errors? Some had speculated that Sanborn (or Scheidt) had placed them there just to throw people off course. Since they aren't significant enough to prevent one from cracking the message, what would be the point? They could just be errors. However, if they were intentional, surely they have a purpose beyond that of simply adding an insignificant additional level of complexity to the decoding process.

Having no immediate answers to these questions, I went on with exploring the last part of the message. It was then that I noticed something interesting.

Below you will find the text cut into the left panel of the Kryptos sculpture. The 97 unsolved characters at the bottom of the panel are highlighted in blue. The red K & R represent the location of two of the "errors" that appear in the decrypted message.

Note that throughout the first three parts of the message the number of characters per line vary, presumably to make each line exactly fill the width of the panel. So, when set in a fixed width font, the ends of the lines formed a jagged edge, that is until you get to section four. The last four lines that make up the unsolved portion all contain exactly the same number of characters.. My first thought was, "What types of encryption would require equal line lengths?"

But then I noticed the KR and YP at the end of the first and third lines. In this sculpture anytime you see anything even remotely resembling the word KRYPTOS it's got to be worth a second look.

I began looking for patterns and found that the displacement of the YP relative to the KR is consistent with that of the alphabets used in the Kryptos Vigenere table. Then I recalled PALIMPSEST, the key word for part one. Could the KR and YP be part of an underlying table peeking through? The message looking like this:

ECDMRIPFEIMEHNLSSTTRTVDOHW?OBKR
UOXOGHULBSOLIFBBWFLRVQQPRNGKSSO
TWTQSJQSSEKZZWATJKLUDIAWINFBNYP
VTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR

VWZKRYPTOSABCDEFGHIJLMNQUVWXZKR
WZKRYPTOSABCDEFGHIJLMNQUVWXZKRY
ZKRYPTOSABCDEFGHIJLMNQUVWXZKRYP
KRYPTOSABCDEFGHIJLMNQUVWXZKRYPT

It seemed rather unlikely at this point. However, it also seemed unlikely that there would be a KR and a YP at the end of the lines that exactly fit the pattern of the known Vigenere table. Was there any supporting evidence - beyond that of the key word PALIMPSEST?

What about the misspellings? The DESPARATLY was easy to write off. It is a very common misspelling of desperately. Just type "desparatly" into a search engine and you will find hundreds of people making the same mistake. And, as it contains two errors, one an error of omission, it was not consistent with the other two errors. Both of the other errors, IQLUSIONS and UNDERGRUUND, were obvious. And the encrypted letter associated with the Q and U were... K and R. There's those two interesting letters again. Instead of "errors" could they actually be parts of the underlying table peeking through?

Because of the variable character widths it is hard to tell exactly where that K and R would fall on the primary text and in the underlying table. For example, when looking at the encrypted text the R is on a longer line than is the K. Thus the R would shift slightly to the left relative to the K as the lines are fit onto the actual sculpture. The table below is my best guess at what an underlying table that is consistent with the positioning of the K, R, KR and YP would look like.

.ABCDEFGHIJKLMNOPQRSTUVWXYZABCDE
AKRYPTOSABCDEFGHIJLMNQUVWXZKRYPT
BRYPTOSABCDEFGHIJLMNQUVWXZKRYPTO
CYPTOSABCDEFGHIJLMNQUVWXZKRYPTOS
DPTOSABCDEFGHIJLMNQUVWXZKRYPTOSA
ETOSABCDEFGHIJLMNQUVWXZKRYPTOSAB
FOSABCDEFGHIJLMNQUVWXZKRYPTOSABC
GSABCDEFGHIJLMNQUVWXZKRYPTOSABCD
HABCDEFGHIJLMNQUVWXZKRYPTOSABCDE
IBCDEFGHIJLMNQUVWXZKRYPTOSABCDEF
JCDEFGHIJLMNQUVWXZKRYPTOSABCDEFG
KDEFGHIJLMNQUVWXZKRYPTOSABCDEFGH
LEFGHIJLMNQUVWXZKRYPTOSABCDEFGHI
MFGHIJLMNQUVWXZKRYPTOSABCDEFGHIJ
NGHIJLMNQUVWXZKRYPTOSABCDEFGHIJL
OHIJLMNQUVWXZKRYPTOSABCDEFGHIJLM
PIJLMNQUVWXZKRYPTOSABCDEFGHIJLMN
QJLMNQUVWXZKRYPTOSABCDEFGHIJLMNQ
RLMNQUVWXZKRYPTOSABCDEFGHIJLMNQU
SMNQUVWXZKRYPTOSABCDEFGHIJLMNQUV
TNQUVWXZKRYPTOSABCDEFGHIJLMNQUVW
UQUVWXZKRYPTOSABCDEFGHIJLMNQUVWX
VUVWXZKRYPTOSABCDEFGHIJLMNQUVWXZ
WVWXZKRYPTOSABCDEFGHIJLMNQUVWXZK
XWXZKRYPTOSABCDEFGHIJLMNQUVWXZKR
YXZKRYPTOSABCDEFGHIJLMNQUVWXZKRY
ZZKRYPTOSABCDEFGHIJLMNQUVWXZKRYP
.ABCDEFGHIJKLMNOPQRSTUVWXYZABCDE

The compelling thing about the table is that the positioning the the K error fits reasonably well with a line that starts with K (looking at the second line in from the left, not the coordinate line on the extreme left). Likewise, R with fits with T, KR with W, and YP with Z, giving us a totally consistent Vigenere table.

So, what's it all mean? There are a number of possibilities.

1. My favorite interpretation is that the KR and YP are parts of the underlying table peeking through at the end of two shorter lines of text.. That could mean the the real message is only 93 characters long instead of the 97 everyone has been trying to decrypt. It would look like this:

OBUOX OGHUL BSOLI FBBWF LRVQQ
PRNGK SSOTW TQSJQ SSEKZ ZWATJ
KLUDI AWINF BNVTT MZFPK WGDKZ
XTJCD IGKUH UAUEK CAR

2. The alternate extreme position would be that it's all random and just happens to fit by chance. I have to admit to being biased against this. Maybe it's because of the process of discovery I went through, first noticing just the even line lengths and the KR and YP. When everything else started to fit, PALIMPSEST, the displacement of the YP relative to the KR, the alignment of the K & R errors, it started to look as if the odds against all this happening by chance were astronomical. Even in the cases where flaws in my thinking were pointed out (see the discussion in sci.crypt), each correction made the underlying table more and more closely resemble a perfect Vigenere.

3. The KR and YP may indeed be part of an underlying table. But, they might not indicate that the message is only 93 characters long. It could be that the KR and YP have replaced other characters in the message, just as the K and R errors appear to have replace characters in the original encrypted text. This leaves the message at 97 characters with four errors. Personally I don't find this very satisfying. Why place the errors at the end of a line? Why use double letter errors that would be easier to spot?

The net result is that we still know nothing for certain about the last 97 characters. If anyone has gone down this path before, I'd love to hear from you. If you find any of this intriguing and begin trying to crack the 93 character version of this message, I'd be interesting in hear about your progress. Please feel free to email me at the address below.

The error I found in the original hypothesis above is that, even if correct, the "KR" & "YP" showing through from an underlying table does not necessarily make the message 93 characters long. As in the earlier "errors" in the cipher text these letters could be replacing other letters. While I had earlier ruled this out as keeping the number of letters the same didn't add any difficulty to the decryption. It now occurs to me that these letters showing through from underneath could just as easily have replaced anywhere from 0 to all 4 characters of the original message. As long as the length changes it adds enough extra difficulty to explain why the last part of the message has so far stood up to all attempts to crack it. It doesn't really matter if the change in length is 1, 2, 3, or 4 characters; any change is enough to make cracking many types of encryptions much more difficult, if not impossible.

Following this reasoning, it is possible that the original plain text was 96, 95, 94, or 93 characters long (instead of the original 97). Based on that idea I did some more thinking. Here is my train of thought:

• Even though Sanborn did the sculpture, the method of encryption was devised by Edward Scheidt, former chairman of the CIA's Cryptographic Center. He has been quoted as saying the Kryptos, in addition to being a work of art, was intended to be a learning experience. Presumably his target "students" would have been members of the CIA, as there was some chance that no one outside the agency would ever have access to the message. Thus it is reasonable to assume that he had some plan in mind as to how someone might solve the puzzle.
• There have been just enough clues provided along the way to allow a competent code breaker to take a crack at the solution. (And just enough "tricks" to keep someone from easily finding it.)
  
  
  a. The second panel gives us the method used to encrypt part of the message, while dividing the message into several parts using different keys (and, beyond the first 2 parts, different methods of encryption) makes finding the solution, even given the method for part of it, significantly more difficult.
  
  
  b. At the end of the second part the "ID BY ROWS" gives us a hint that the next part involves a columnar transposition, which is written in rows and read off in columns. (It turned out to be three such transpositions.)
  
  
  c. The key word PALIMPSEST might provide some explanation for the rather obvious "errors". (This is still speculation, but it is hard to believe that there isn't a specific purpose for these errors.)
• Are there clues to the method used in part 4? There is no reason to believe that doesn't include a columnar transposition. Parts 1 and 2 used the same method. It is just as likely as not that Parts 3 and 4 use the same method. The frequency distribution of the letters is consistent with a simple substitution. So my guess would be it is a combination of the two. The substitution making it just enough harder so that the techniques used to crack part 3 might not suffice on part 4 – especially when coupled with the changed length of the message. This "change in alphabets" could also have been an intentional device to allow the remaining part (3 & 4 combined) to be more easily divided into two separate messages.
• In section 3 Scheidt was uncommonly kind. The message was 336 characters long – evenly divisible by all 3 periods for the columnar transpositions. And, in all three transpositions the columns were simply read off in reverse order – no key words were used to rearrange the columns. This 336 length was obviously deliberate as Scheidt had to add a "Q" null character to fill out the message. Common practice would have been to add enough nulls to end in number divisible by 5, giving consistent 5 letter code groups. In this case no nulls would have been necessary. The message was already 335 characters long. Not only is this divisible by 5, it would result in a partial last column. Columnar transpositions in which the last column is a partial are more difficult to crack. So, it would seem that the only possible motive for adding the null was to make the triple columnar transposition "just" easy enough to solve via the methods Scheidt had in mind.
• If Scheidt was to take a similar approach in section 4 the obvious message length would be 96, with prime factors of 2 x 2 x 2 x 2 x 2 x 3. This would provide the maximum possible number of evenly divisible periods for the triple columnar transposition. None of the others lengths are as attractive – 97 prime, 95 (5 x 19), 94 (2 x 47), 93 (3 x 31).

If the method used in part 4 is a simple substitution with a triple columnar transposition using a 96 character message length, the cipher text would be one of the following (where the ? is an unknown character):

...........................OB??
UOXOGHULBSOLIFBBWFLRVQQPRNGKSSO
TWTQSJQSSEKZZWATJKLUDIAWINFBN?
VTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR

...........................OB?
UOXOGHULBSOLIFBBWFLRVQQPRNGKSSO
TWTQSJQSSEKZZWATJKLUDIAWINFBN??
VTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR

Visualize this as if parts of the original cipher text document have been damaged exposing the KR and YP below. Since this "damage" to the message occurs at the end of a line there is no requirement for it to be a character for character replacement. One of the underlying characters could just as easily be replacing bank space at the end of a line.

I've had the mathematical definition, a measurement along the x-axis, locked in my mind – as I think we all have. After all, cryptology is highly mathematical. I always thought that somewhere, somehow in solving this puzzle there would be a requirement to use a measurement along an x-axis.

In a moment of boredom I typed ABSCISSA on the Goggle search page just to see if anything inspirational would pop up. Plenty of mathematical sites came up. But, much to my surprise, there were also a large number of references to the Latin definition, which is to cut away or shed. This, I would argue, fits perfectly with PALIMPSEST and the cipher text "errors," and is yet one more bit of evidence that the idea of the "KR" and "YP" in the final message showing through from below has some merit.

Comments welcome. See contact info below.