SIDE QUEST 01 / CODE-BREAKING / CHECKING THE EVIDENCE
K4 is the fourth coded message on the Kryptos sculpture. In the project's latest check, it was still listed as unsolved, and K5, a related fifth message, still hadn't been released. So this project's best result isn't a claimed answer. It's a more and more exact map: which methods are ruled out, what's still unclear, and the tests any real solution has to pass.
We know 24 of the hidden letters. The words EAST, NORTHEAST, BERLIN and CLOCK sit at known spots. Code-breakers call known words like these cribs. Any proposed method has to explain them. But it can't sneak in so many adjustable tables, masks or settings that it could produce any 97-letter message at all. A decoder only counts as evidence if its settings were worked out without peeking at the message it outputs.
The project's latest scheduled check found that the official Paradigm page still lists K4 as unsolved. No K5 text is out yet. That matters, because a second related message could add fresh, independent clues. Many very flexible methods can't be pinned down with K4 alone, and those clues are exactly what they're missing.
An independent rebuild of a public, SolveKryptos-style decoder found 97 equations but 108 unknown table entries in use. That's more unknowns than equations, and the equations never loop back on each other (the equation graph is a “forest”). With the tables left free, it fit the proposed answer. It also fit two other English messages, and 200 out of 200 random strings that contain the known words. So getting one chosen answer out of it isn't evidence for that answer. It only counts if the tables were worked out independently.
Two exact tests rule out large, clearly limited groups of methods. One asks whether a method could produce the known words at all. The other runs through every setting in a controlled way. Ruled out so far: tied Quagmire III (a classic cipher type) with every period, or key length, up to 26. Also repeating tableaus (grids of shifted alphabets) built on the KRYPTOS alphabet, several chart routes, and stepping-rotor variants. Also key streams from two Berlin clocks, the Weltzeituhr and the Mengenlehreuhr, and several two-clock or tied-key families. Each one is ruled out only in the exact form that was checked.
The Paradigm archive has a photo of a copper test plate made in 1988 by Sanborn, the sculptor. Read straight from the photo, the rows are CDEFGHIJ / DEFGHIJK / EFGHIJKL. Those are plain A–Z rows, each shifted by one, not the KRYPTOS-keyed alphabet. It's weak but real evidence that a standard Vigenère-style grid showed up somewhere in the design process.
This picture sums up two tests from the research notes. Can it fit the known words? Does it pass the letter-match test? It's a quick screen, not a solution.
Group K4's positions the way a proposed cipher would. Inside one group, every letter goes through the same one-to-one swap. A swap can rename letters, but it can't change which ones match: two equal letters stay equal. So compare how often letters match inside each group with how often they match in normal English. This tests every possible alphabet, without having to guess one.
Several early rounds were built around period 8. It's the smallest “weak” repeating chart that can fit the known words. Later, this test showed that, statistically, the tempting small model doesn't fit normal English with one fixed letter swap per state. Looking back, that one “no” made many expensive period-8 searches unnecessary.
Follow each branch as it gets tested, narrowed, fixed, ruled out or kept alive.
This sums up experiments recorded in my Kryptos research project. It doesn't claim to solve K4. Each status applies only to the exact group of methods that was tested.
Change the period to regroup all 97 spots by their remainder when divided by n. At two spots, 32 and 73, a known letter encrypts to itself. Those two tell us a lot about Quagmire III.
Treat each W in K4 as a blank to skip (a null), and the key pattern shifts. With n=19, this version passes the letter-match test. It's a really interesting lead, and one that can be settled for sure.
Before reading meaning into a proposed answer, ask: could this method fit any random string that contains the known words? A decoder with more free settings than real clues can cook up convincing output. That doesn't mean it found the real method behind K4.
Sometimes a group of methods is small enough to list in full, or to simplify with algebra. Then a “no” can remove a whole method, not just one bad key. The project keeps those exact, limited results, even while bigger story-driven theories get thrown out.
Letter-match and English-likeness scores can push a method way down the list. But they don't prove that every key made of real words, or every complicated variant, is impossible. So I label each piece of evidence by type: exact check, statistics, or rule of thumb.
K5, original coding charts from the archive, or new statements from the creator could add clues that K4's 97 letters can't. Until one of those shows up, more open-ended searching has shown little payoff.
The plan is narrower on purpose. Keep doing cheap, exact checks and small proofs. Watch for K5, or sharper original material from the archive. Test any outside “solution” for too much wiggle room before spending lots of computer time on it. Very flexible methods that use K4 alone are on hold, rather than squeezed until they spit out a candidate answer.