Grid reading orders plus a repeating key, on 73 letters left by a column-band filler model

More than ten ways of reading a 28-by-31 layout of the cipher text, each with a repeating key of length 1 to 26, on the 73 letters left by a column-band filler model. The key was tried in Vigenere, Beaufort and Variant Beaufort form. Reading orders: column by column in either direction, back and forth by rows or columns, diagonals, a clockwise spiral, paired mirror columns, and the whole text read backwards. The 73 letters are what remains after treating 24 K4 letters in columns 8 to 16 as filler (2,262 settings). The best, 22 of 24 known letters (back and forth by rows, Beaufort, key length 24), is noise: with a 24-letter key there are as many key letters as known letters, so high scores come easily by chance. At key lengths of 7 or less, where the test can separate signal from chance, the best was about 12 of 24. Column reading of all 97 carved letters gave at best 20 of 24 at key length 25, also too long a key to mean anything.

In plain English: More than ten ways of reading a 28-by-31 layout of the cipher text, each with a repeating key of length 1 to 26, on the 73 letters left by a column-band filler model, using a combined approach: scramble the letter order AND replace each letter. Every combination within the stated scope was tested (2 thousand configurations); none produced a valid solution. Best match: 22/24 known letters. That is not a real signal: searches of this kind can reach scores this high without the right key. This record is negative within the tested scope.

NOISE Tier 2

How to read this record

Verdict – ELIMINATED (ruled out within the tested scope), NOISE (nothing beyond what chance produces for a search of this size; also shown when a record stored no verdict of its own), INCONCLUSIVE (the test left candidates or options standing, or could not decide, so it does not rule the method out), INTERESTING (slightly above random, almost certainly coincidence), SIGNAL (statistically unusual, warrants investigation), or FULL MATCH (all 24 known letters correct).
Confidence Tier – Tier 1 = mathematical proof, valid while its stated assumptions hold. Tier 2 = every possibility within the stated scope tested. Tier 3 = partially tested or sampled. Tier 4 = not yet properly tested.
Configs Tested – How many different key/parameter combinations were tried.
Best Score – How many of the 24 known plaintext letters the best attempt matched (out of 24). N/A means no such score could be read from this record.
Keystream Consistency (Bean) – Whether the key values at different positions are mathematically consistent with each other. The check only holds when the known letters stay in their carved positions (no letters rearranged or removed), and the full set used here assumes the standard A to Z alphabet.
Scope Limitations – What this elimination does not rule out.

Configs Tested: 2,262
Best Score: 22 / 24 known letters matched · high, but judged chance for this search
Confidence: Tier 2: Exhaustive search under stated assumptions. Does not eliminate multi-layer variants.
Truth Tag: [INTERNAL RESULT]
Date Tested: 2026-03-12

Scope Limitations

This elimination does not rule out:

Keys that do not repeat (self-keying, running keys and others) combined with these reading orders
Reading orders not in the list
Other ways of choosing the 24 filler letters

Assumptions

The cipher text laid out in 28 rows of 31 letters
Filler letters are the K4 letters in columns 8 to 16, minus one kept column
A reading order of the grid is the letter rearrangement
Repeating key of length 1 to 26

Reproduce

PYTHONPATH=src python3 -u scripts/grille/e_native_grid_reading_01.py

Requires the kryptos repo, Python 3.11+, PYTHONPATH=src. Some recent scripts may not be in the public repository yet.

Found an error? Report on GitHub