Another practice cipher | Noita | Page 1

cosmic hare Mar 24, 2025, 7:38 PM

#

📎 message.txt

ionic trail Mar 24, 2025, 8:41 PM

#

ASCII+32 again

📎 sdlwdr_practice_v2_ascii32.txt

#

Looks like this one has a lot less isomorphs?

cosmic hare Mar 24, 2025, 8:47 PM

#

i would guess this one looks closer to random, but i've done zero analysis on either

ionic trail Mar 24, 2025, 8:59 PM

#

It's weird that this one has so few isomorphs compared to the other if this is more CTAK-based

cosmic hare Mar 24, 2025, 9:00 PM

#

there's another bit of complexity to this one that kind of makes that make sense...
i'll see if i can make a v3 that preserves more isomorphs

low spade Apr 5, 2025, 10:06 PM

#

This time there is only one letter J, huh.
Shared starts up to position 10 and fairly weak isomorphs with direct conflicts. Lots of ciphertext repeats.
Kasiski gcd(512, 92, 122, 299, 247, 424, 429, 55, 177, ...) = 1, convincingly not periodic.
Overall IoC is around 0.0128, cross-message IoC from position 10 to 200 is a bit higher at 0.0167 with no aligned repeats to attribute it to (UPD: actually there are some aligned repeats, longest being A*g<\g*, >)(FZ, TE=<#, and many shorter ones).

low spade Apr 5, 2025, 10:22 PM

#

Periodic IoC has high values around 42 and 84.

#

low spade Apr 24, 2025, 1:55 PM

#

Tallying up distances between positions of the same n-gram, we find them bunching around multiples of 85/2. The effect appears to be a bit more pronounced (higher peak-to-valley ratio) when we only count distances within the same message vs when we only count distances between occurrences in different messages.

low spade Apr 24, 2025, 2:14 PM

#

Tallying up letter occurrences at positions mod 42, it looks like every letter has a preferred range of positions mod 42 it likes to occur around. This can be seen more clearly if letters are ordered by the "mean position mod 42".

#

Plotting letters (ordered as above) vs their positions (red lines are mod 42), looks like messages start synchronized, but very slowly drift apart. E.g. message 3 appears to pull ahead, while message 6 appears to lag behind (or the opposite, depending on how you look at it).

#

Message 10 is interesting, because it's the only one that has letter J (code 42 at position 91), and it appears to "change direction" exactly at the position of this letter J.

low spade Apr 30, 2025, 1:00 PM

#

low spade Message 10 is interesting, because it's the only one that has letter J (code 42 ...

Given this is the only section that moves in the opposite direction, it is helpful to ignore it in further statistical analysis.

low spade Apr 30, 2025, 1:50 PM

#

low spade Tallying up letter occurrences at positions mod 42, it looks like every letter h...

Looking at standard deviations of position mod 42 for every letter (i.e. how wide the range of positions it likes to appear at), there appear to be two separate clusters of letters. This becomes especially apparent if we plot them against total letter frequencies.

#

I want to postpone using the meta-knowledge from the previous cipher as much as possible, but I will note that ||the clusters exactly match the cycles found there. The internal order of the cycles also seems to very convincingly correlate with the order inferred from statistics here, but obviously not exactly because statistics is noisy.||

low spade May 9, 2025, 4:06 PM

#

Using the inferred orders within the clusters (even though they are fuzzy), we can find probable plaintext repeats by finding long sequences that share the same "pattern of cluster belonging" and looking at alphabetic distances between them.

Here are 4 fragments with pattern 101101111001111110011:

A1. #L:<YA#%12W^T>e_CLmF0
A2. C"O4kiF>ePXZB13B>R~A4
A3. Q=[TjNf3>nkA5#Q#CKIeF
A4. hrF[=BQ#>)(Fl%1h&*9d[

Looking at distances between each pair we can speculatively say that A1, A2 and A4 are likely to at least partially share plaintext (A4 seems to have slightly different beginning than A1 and A2).

#

Another likely triplet:

    010110101100111101100111101
B1. (iX3BqG8oQX]3d1O~#_U7Cc0cj>
B2. ":p45]38c&m;4F[e;AF2bVT_TPG
B3. kZLa&@-"+<Ir.Si:'d4D!#.o.=g

#

And the longest pairs:

    1010111100110100111110011001010111
C1. i*BjA+45*pC&H3nqel1O1/Da0)qM$hrdAo
C2. 0?1n[45[?6+ZmB'jSCc0cjYd#`\ZkT7Gl%

    1011101001111001010110101111100110
D1. SP^1F`,kpV.G5IR+Ic6o0?[/fN[e1U'dh$
D2. l\<ZB*12W^,VMLbObaD+.RG@[S-GVDm_c6

    01011001111011001111010111010010
E1. @-"+<Ir.Si:'d4D!#.o.=gjS&^*dbnC*
E2. KC`AePXZaFMpc.?]41O1/TE+N0US2YQ=

    1100111101010110110011100101011
F1. _hrj#_hQKC~#j4&H:%7WOcd(p^*>?[T
F2. g#j!AogN7SP^ES-R%:W@Bfepr<rV9ai

low spade May 9, 2025, 10:45 PM

#

Peeking into how these distances look under almost certainly correct order (meta-knowledge), and I'm not sure how one would make the jump from fuzzy order to correct order. Perhaps knowing the entire mechanism could eventually help. So I think I will have to continue with the meta-knowledge, and maybe then circle back.

With meta-knowledge (and spoiler images) we find that ||even under the correct alphabet the distances are not constant (sorta expected because CT fragments were not isomorphic). But notably they appear to change only by ±1.|| This reads to me as ||drift that only happens in one direction, i.e. that two cipher states, when encrypting a letter of the same plaintext, either drift in parallel (both drift or both don't drift), causing distance to be preserved, or one drifts and the other does not, causing distance to increase or decrease by 1.|| With meta-knowledge and this property in mind, one could easily find much shorter likely instances of plaintext repeats.

#Another practice cipher