A (hopefully) considerably copious collection of classical cipher crackers
Written in Python3 (and run with PyPy)
- Investigate using
__slots__for time/memory optimization - Add methods to the cipher objects so that keys can be more quickly mutated
- Write a word splitter which separates plaintext into words
- Investigate using AI to identify what cipher a ciphertext might be encrypted with
- Investigate training a simple AI-like model which can evaluate the english-ness of text
- Plaintext: a message which is readable and unencrypted
- Ciphertext: a message which is encrypted
- Decoder/Decrypter: a program which will decrypt a ciphertext when given the key
- Cracker: a program that can decrypt a ciphertext without the key, usually by making use of the decrypter
A cracker has three major components: the key-finding algorithm, the decrypter, and the text evaluator.
The key-finding algorithm will generate keys which will be used by the decrypter to find the plaintext that the key would generate. This plaintext will be tossed into the text evaluator. If it gets a high score, the current key and plaintext will be saved as a possible answer. As more keys get tested, the best key-plaintext pair will be continuously updated.
If the keyspace is small, then every key will be tested. If brute force isn't feasible, a hill-climbing algorithm will be used, where random changes to the key are either kept or dropped if they create a better text evaluation. The goal of the algorithm is to reach a "hill", where the plaintext is as likely to be the real message as possible. To avoid running into local minima (which is a common problem with hill-climbing), random-restart hill climbing (also known as shotgun hill climbing) is used. Simulated annealing may also be used in the future.
Code should be written to prioritize speed when run under PyPy, though it should still be reasonably readable. In particular, decrypters must be optimized to run quickly, as they are often called thousands of times in order to crack a cipher. Different versions of decrypters should be tested and timed to determine which is most performant. The same applies to the text fitness evaluator.
http://practicalcryptography.com/cryptanalysis/text-characterisation/quadgrams/
https://medium.com/analytics-vidhya/how-to-distinguish-between-gibberish-and-valid-english-text-4975078c5688
https://planetcalc.com/7959/, https://planetcalc.com/8045/
https://gitlab.com/guballa/SubstitutionBreaker/-/blob/development/subbreaker/breaker.py