Code
Reading a text without its key: naming the cipher, peeling off encodings, counting letters, measuring an unknown script, breaking nomenclators, reading messages in depth, replaying the Enigma breaks, checking anagrams.
Messages in Depth
Sets several ciphertexts one over another and counts the places where they agree, as Kullback’s manual does, to say which share a key. Those that do are read together, column by column: by frequency, by a word tried at every place of one message, and for letters by a search of the whole key, set beside the same search on random letters. The examples are Unveil’s own messages.
Terms
History, how it works, how it was checked, limits
History
Messages enciphered with the same key from the same starting point are said to be in depth. Solomon Kullback’s Statistical Methods in Cryptanalysis, issued by the War Department, gives the reason they can be found: two texts enciphered alike and set one over the other show the same letter in the same place as often as two plain texts do, 0.066 of the places in English, where unrelated text agrees in one place in 26.
How it works
Each line is a message: its letters, or its bytes in hex. Every pair’s coincidences are counted; then each message is counted against all the others, and one that falls clearly under what a shared key gives is set apart. A count is taken for a shared key only where chance gives as many less than once in a hundred times. In the rest every column stands under one key value, first guessed from the column’s commonest letters. A word typed for one message gives the key under it, which reads the other messages there; tried at every place, the places where the others read best come first. For letters, a search (run unasked on a small depth) then changes one key letter at a time until the messages’ quadgrams read best, and runs again on random letters of the same lengths.
How it was checked
Kullback’s Examples 14 and 15 are recomputed: φ is 42 and 18, beside the 39.6 and 22.8 he expects. Six of Unveil’s own messages under one random key show 69 coincidences in 1,006 pairs of letters, where a shared key gives 66 and chance 39. Frequency alone reads 246 of their 411 letters right, the search 406 of the 410 in depth; REPORT placed in the second message opens the other five there. With two messages the search reaches a fluency of 100 with 42 of 142 letters right, and random letters reach 100. A ninth message under another key is set apart, as it is under ten of sixteen keys tried for it.
The tests hold 31 checks of this tool; 2 rest on outside material:
- A print’s own example Kullback’s Example 14, p. 38: φ of IBMQO PBIUO MBBGA JCZOF MUUQB is 42, beside 39.6 expected of one alphabet in English (0.066 × 25 × 24) and 22.8 of random text (0.038 × 25 × 24)
- A print’s own example Kullback’s Example 15, p. 38: φ of HKWZA RRPBQ BIVYS MPDMQ MVUDC is 18, nearer random text’s 22.8 than one alphabet’s 39.6
Limits
The messages must begin at the same place of the key: no offset between them is searched. The search reads letters in the six languages; bytes are read by their columns and by words placed, and for bytes only chance is known, not what a shared key gives. A reading that fits is not proved.