Workbench

At the 1990 dedication Sanborn said there are “two systems of enciphering the bottom text”. The lower half of the panel holds both K3 and K4, so how that applies to K4 alone is an interpretation, not a fact. Pure transposition is impossible (CT has 2 E’s, cribs need 3), so at least one layer is substitution. Beyond that, the architecture is open: the live questions are about ordering, segmentation, and whether any procedural layer is present.

Eliminated proven impossible (algebraic)

• Repeating-key substitution on the raw 97 characters (Vigenère, Beaufort and Variant Beaufort, key lengths 1–26, on both the A–Z and the KRYPTOS alphabet)
• Pure transposition (CT has 2 E’s, cribs need 3)
• Filler letters removed (any 24 of the 97, none of them known-plaintext letters) + repeating-key substitution on the 73 letters left (key lengths 1–23, standard or KRYPTOS alphabet)
• Self-keying (autokey), plaintext- and ciphertext-keyed, on the raw 97 characters with a starting key of up to 25 letters. With an unrestricted rearrangement layer added, it is not ruled out.

Eliminated tested exhaustively

• Columnar transposition (widths 4–9, every column order) × repeating-key substitution (key lengths 1–24, A–Z alphabet): no combination fits (exhaustive re-run, August 2026)
• Single-letter and 1M-word dictionary self-keying keys on the raw 97 characters

Open territory viable

• Non-standard transpositions (serpentine, spiral, Myszkowski) with a longer or non-repeating key (with repeating keys of length 2 to 7 they were tested, no signal)
• Self-keying (autokey) or running key after undoing a rearrangement
• Custom tableaux, including repeating keys that use a different alphabet for plaintext than for ciphertext (for example two different keyword-mixed alphabets), or one keyword-mixed alphabet used the K1/K2 way with a key of 23 letters or of 26 or more
• Filler-letter hypotheses (for example 24 fillers leaving 73 letters) with a non-repeating key: not ruled out, but no independent evidence supports them
• W-delimiter hypothesis: the 5 W characters at positions 20, 36, 48, 58, 74 divide K4 into 6 segments of 20|15|11|9|15|22 chars. That makes them a bounded layout to test, not evidence of a delimiter: the old width-21 repeated-bigram effect that vanishes when the Ws are removed also weakens or vanishes when other letters spread through the text are deleted, so it says nothing about the Ws (Finding 1). Use the “W-segment” transposition and the “W positions only” filler setting to explore this.

About near misses (23/24)

A 23/24 is not a sign of a hidden delimiter. Long keys and flexible methods reach 23/24 by chance, and none of the 24 known-plaintext positions holds a carved W. The idea that all five carved Ws decrypt to one separator letter such as X, Q or Z does not work with any repeating key of up to 26 letters on the standard or KRYPTOS alphabet: a key with no repeated letters would need a length of 1 or 2 to fit the gaps between the Ws, and the known plaintext rules out every repeating key of up to 26 letters on those alphabets anyway. Some keys of 27 letters or more do fit the known plaintext, and this argument does not rule them out. If only some of the Ws do, or the key never repeats, the idea makes no prediction that can be tested (Finding 1).

Full methodology and elimination landscape →