721.7B+ configurations evaluated across recorded experiments

Kryptos K4 remains unsolved.
Here’s everything we’ve ruled out.

A public research archive documenting 721.7B+ tested configurations across 597 recorded experiments, with reproduction commands and scope notes where the record has them. Built by a human and an AI who find this problem genuinely fascinating.

  1. 721.7B+ configurations evaluated summed from the counts each run reports
  2. 597 recorded experiments repro commands and scope notes where recorded
  3. 597 published eliminations written up and browsable
  4. 0 solutions K4 is still unsolved
Research

What’s been eliminated

Detailed records of what we tested, what failed within scope, and which claims were later retired after better controls.

Explore

Try it yourself

Interactive tools for experimenting with K4: test cipher combinations, visualize the cylinder, or submit your own theory.

What we’ve learned
Open lead, heavily tested

The W-Delimiter Hypothesis

Five of K4's carved letters are Ws, and one idea is that they act as dividers. Deleting them makes a statistical oddity vanish, which once looked like support, but a 2026 audit showed that the oddity cannot tell one letter from another: deleting other letters breaks it too, so it says nothing about the Ws. Every simple version of the divider idea has now been tested without success; it stays alive only as one ingredient of a more complex scheme.

Read more →
Key result

Most Textbook Ciphers Are Ruled Out

Most standard classical ciphers, applied in the ordinary way, have been tested against K4 and failed: repeating keywords on the standard and KRYPTOS alphabets, letter rearrangements, and large bounded slices of two- and three-step combinations. A few variants built on keyword-mixed alphabets are not yet fully closed. Whatever K4 is, it is probably not a textbook cipher used the textbook way.

Read more →

Why this exists

Kryptos K4 has been unsolved since 1990. In that time, thousands of people have independently tried the same approaches (Vigenère with every keyword, columnar transposition at every width, Beaufort with every alphabet) and hit the same walls, alone, with no way to know someone else already tried it.

This site is an attempt to fix that. Each recorded approach is documented with its verdict and best score and, where the record has them, its parameters, scope limits, and a command you can run to verify the result yourself. If an idea has already failed inside a bounded test envelope, you’ll know that before you spend a week on it.

The project started as one person’s obsession with a sculpture and grew into a collaboration between a human researcher and Claude, an AI that turned out to be an excellent partner for exhaustive cryptanalysis. The codebase is public on GitHub. Most recorded experiments include a command to re-run them. The failures are documented as carefully as the successes.

We haven’t solved K4. But we’ve mapped a lot of the territory where the answer isn’t.

What is Kryptos K4?

Kryptos is an encrypted sculpture at CIA headquarters carrying four messages. The first three were solved in the 1990s. This is the fourth, still unsolved since the sculpture was dedicated in 1990: scrambled strips of its plaintext turned up in Sanborn’s archived papers in 2025 but have never been published, and the method has never been made public. Learn more →

K4: 97 characters, 24 of them known
O B K R U O X O G H U L B S O L I F B B W F L R V Q Q P R N G K S
· · · · · · · · · · · · · · · · · · · · · E A S T N O R T H E A S
S O T W T Q S J Q S S E K Z Z W A T J K L U D I A W I N F B N Y P
T · · · · · · · · · · · · · · · · · · · · · · · · · · · · · B E R
V T T M Z F P K W G D K Z X T J C D I G K U H U A U E K C A R
L I N C L O C K · · · · · · · · · · · · · · · · · · · · · · ·
Sanborn released these plaintext words between 2010 and 2020, and a 2025 layout image from him shows them in these slots. Drawing them under the carved letters assumes each carved letter decrypts to the plaintext letter in the same position: the best-supported reading, but it rests on relayed conversations, not a direct quote from Sanborn. The rest of the plaintext has not been published. Positions are 0-indexed, as they are throughout this site; most published sources count from 1 (22 to 34 and 64 to 74).
Categories

Bespoke

133 eliminations

Non-standard methods inspired by the physical sculpture or military cipher systems. These approaches don't fit neatly into classical categories. Includes DRYAD charts, Morse code analysis, and coordinate-based approaches.

Fractionation

19 eliminations

Methods that break each letter into smaller pieces (like grid coordinates), scramble those pieces, then reassemble them into new letters. Includes Bifid, Playfair, Four-Square, and ADFGVX.

Key Models

128 eliminations

Different ways to generate the secret key: from a passage in a book, from a date, from a mathematical formula, or from the sculpture itself. Includes running keys, self-keying (autokey) ciphers, and keyword-derived approaches.

Multi Layer

73 eliminations

Combined approaches that stack multiple encryption steps. For example, replacing letters first, then scrambling their order. Includes filler-letter removal, cascaded layers, and joint optimization.

Substitution

80 eliminations

Methods that replace each letter with a different letter using a key or pattern, like a secret alphabet. Includes Vigenère, Beaufort, Quagmire, Hill, and more.

Transposition

68 eliminations

Methods that scramble the order of letters without changing them, like writing a message into a grid and reading it back in a different order. Includes columnar, rail fence, route, and grille ciphers.

Uncategorized

96 eliminations

Eliminations not yet assigned to a specific category.