[split] Verbose cipher?

[MarcoP]

Code:

Alphabet:    '+0123456789acdefghijklmnopqrstuxy  uses 35 characters

Thank you for the examples!

This is a suggestion for a specific experiment inspired by your results. Of course, there is a huge (infinite?) number of different scenarios. I may be misunderstanding something, so take what I write with a big grain of salt.

I think that treating space as a character is anachronistic. Instead, one could simply remove all spaces and start from there.

Since the alphabets you get have many symbols, one could tackle this as a verbose homophonic cipher: some plain-text characters are represented by more than one cipher sequence ("sequence" because the cipher is also verbose).
As a first experiment, I would exclude the idea that a single cipher symbol corresponds to something longer than a single plain-text character (BTW, this would probably result in lower entropy again).

For instance, the alphabet above could be interpreted as (top cipher symbols, bottom plain-text - dots only added for alignment):

Code:

'+0 1 2 3 456 7 8 9 acd e f g h i jkl m n o p q r s t u x y .A. B C D .E. F G H .I. J K L M N .O. P Q R S T U V W X Y Z

Meaning that plain text 'O' can be encoded as any of j,k,l. Of course, a problem is that you have a huge number of mappings like this. I don't think that an homophonic cipher can be addressed by brute-force, I guess there is a literature about algorithms to tackle the problem.

For each verboseAlphabet+homophonicMapping combination, you decode the VMS and compare with candidate plain texts (where spaces were also removed). Rather than comparing entropies, you could directly compare bigram or trigram frequencies. I am afraid that, if alphabet sizes are different, comparing entropies does not really make sense.

Things would be somehow simpler if the verbose step produced an alphabet with the same size as the plain-text alphabet: in that case, entropies could be more significant. But Rene recently pointed out that similar entropies can result from very different character-frequency distributions: similar entropy is necessary but not at all sufficient for a decent decoding candidate.