“If the key were running English text: do the key letters forced at EASTNORTHEAST and BERLINCLOCK read as English under any of the 208,062 dictionary-word alphabets, better than on scrambled coppers?”
F1.3. 0009 and 0010 read the forced key stretches as English under the standard, KRYPTOS and 61 puzzle-word alphabets (best -3.81, the same as shuffled K4). Here every keyword alphabet from the project dictionary (208,062 distinct, as in 0109) is used as the plain alphabet, the cipher alphabet or both, under three table rules and two key alphabets (standard, or the same keyword alphabet): 18 readings per alphabet, 3.7 million per text. The best English score of the two forced stretches together is compared with 50 scrambled coppers searched identically, and 20 planted running keys measure power.
revertedstatus
a new rootparents
5 mintime, all phases
0candidates
0/24known letters
1
The experiment method
What the method does with the 24 known letters.
A running keya running key A secret word as long as the message, taken from a book or another text. uses the letters of a book as the keythe secret word The word or string a cipher needs besides the method., one book letter per message letter. If K4the carved letters The 97 letters cut into the copper at the CIA, shown at the top of every experiment page. were made that way, then where we know the messagethe message The 97 letters of English that Jim Sanborn wrote and enciphered. (EASTNORTHEAST and BERLINCLOCK), working backwardsa mirror Reading or writing the text from the end to the start. gives 24 letters of the book itself, which would read as English. We try this with tables built from every word in an English dictionary, and compare the most English-looking result with what the same search finds on scrambled copies of the carved text, where no book is hidden.
Words in the margin are defined on the glossary page.
1.1
The method at work
The carved letters go in at the top; the reading comes out at the bottom.
the idea
With a running key, the key is a book's text. Where the message is known, working backwards gives 24 letters of that book, which would read as English. Tried with tables built from every dictionary word, three ways, three rules, two key alphabets.
the check
The best of 3.7 million readings scores -4.94 on K4; the same search on 50 scrambled coppers gives -5.28 to -4.30. Twenty hidden running keys from Carter's book score -4.78 to -3.82.
the bound
K4 sits in the middle of the coppers and below every hidden running key: a running key from an English text on such a table would have shown. Closed as sampled at stated power.
2
Best candidate against the known letters
This experiment produced no candidate: the search ruled everything out before a full text was built. Its product is the bounded negative in section 7.
3
Walkthrough, by hand
Redo the method on a few letters with pencil and paper.
Pick a dictionary word, say WINTER, and build its table alphabet: WINTER then the unused letters in order.
Pick a rule, say Vigenere with WINTER on both sides.
At place 22 the message has E and the copper F: key value = place of F minus place of E in that alphabet.
Read the key value as a letter of the key alphabet.
Do this for places 22-34 and 64-74: two strings of 13 and 11 key letters.
Score both strings for English four-letter groups.
Repeat for every word, side, rule and key alphabet; keep the best.
Do the same with scrambled copies of K4 and compare.
the worksheet: the first steps by hand, on squared paper
4
Worked example, generated
The same steps run by the code on the real letters.
alphabet from WINTER: WINTERABCDFGHJKLMOPQSUVXYZ; Vigenere, WINTER on both sides, key letters on the standard alphabet
place known carved key value key letter
22 E F 6 G
23 A L 9 J
24 S R 11 L
25 T V 19 T
26 N Q 17 R
27 O Q 2 C
28 R P 13 N
29 T R 2 C
30 H N 16 Q
31 E G 7 H
32 A K 8 I
33 S S 0 A
34 T S 17 R
64 B N 21 V
65 E Y 20 U
66 R P 13 N
67 L V 7 H
68 I T 2 C
69 N T 1 B
70 C M 8 I
71 L Z 10 K
72 O F 19 T
73 C P 10 K
74 K K 0 A
5
Hypothesis, for a cryptanalyst
Family, key space, the constraint under test, and the clues it reads.
C1(b) (the table alone), C27(c) (a different tableau), C5 ('key word/s', a running key from a named text). A running key from an English book on a keyword-mixed table: at each known place the key letter is forced, k = C(c) - P(p) (Vigenere), P(p) - C(c) (variant) or C(c) + P(p) (Beaufort), read through the key alphabet (standard or the keyword alphabet). The 13 forced letters at 22-34 and the 11 at 64-74 would then be two windows of English. Score: mean log10 quadgram over both windows (10 + 8 quadgrams, the harness's table). Alphabets: every word of four or more letters in kryptos/data/words.txt, keyword-mixed (repeats dropped, rest of the alphabet in order), standard and KRYPTOS excluded (0009); configurations (W, std), (std, W), (W, W). Exhaustive over this space. Controls: 50 shuffles of K4 (seed 125), identical search. Power: 20 planted texts, each a K4-length ciphertext made from a random English plaintext with the known letters in place, a running key from Carter's Tomb of Tut-ankh-Amen vol. 1 at a random offset, a random dictionary alphabet and configuration; a plant counts as found if its best score beats every copper. Tests reading C5 'a named text' for any text, without the text.
6
Prediction
Written before the run; the result decides it.
K4's best combined score lies inside the range of the 50 coppers (the space is 3.7 million readings per text, and the best of that much noise reaches near-English on 24 letters). Power is low: fewer than half of the 20 planted running keys beat every copper, because a true 24-letter English window scores about -4.2 while the best of 3.7 million noise readings is expected near -4.0. If so, F1.3 is recorded blocked (the forced letters are too few to tell a running key from chance), not closed.
7
Method
Build the 208,062 keyword alphabets.
For K4, 50 shuffled coppers and 20 plants: for every alphabet, configuration, rule and key alphabet, compute the 24 forced key letters and score the two windows.
Keep the best combined score and the best per window.
Compare K4 with the coppers; count the plants found.
8
Result and bounds
The harness's record: candidates, gates, loss, and the negative result this experiment leaves behind.
0candidates returned
0passed every gate
0/24known letters, best
–lowest loss
nonechampion before
K4 best -4.936 (('IREPABLCDFGHJKMNOQSTUVWXYZ', 'W-W', 'vig', 'W', 'CEFEEREDADFIZGWENTACTTWI')); coppers -5.281 to -4.300; plants found 11/20 (true windows above every copper: 11).
What was searched, as the harness recorded it.
family
running key from an unknown English text on keyword-mixed tables from the dictionary, no move
alphabets
208,062
readings per alphabet
18
readings per text
3,745,116
exhaustive
yes
windows
22-34 (13 letters) and 64-74 (11), scored together (18 quadgrams)