The Z13 cipher is thirteen characters long. It was mailed to the San Francisco Chronicle in 1970, tucked inside a letter where the Zodiac Killer claimed his name was hidden in the code. Fifty-six years later, a cold case investigator named Alex Baber announced he had cracked it — using AI to generate seventy-one million possible thirteen-letter names, cross-referencing them against census, military, and marriage records, and arriving at a name: Marvin Merrill.
An alias, Baber says, for Marvin Margolis — a man who dated Elizabeth Short in the 1940s and appeared on the LAPD's suspect list after her murder and dismemberment. The Black Dahlia case. A former NSA codebreaker, Ed Giorgio, reportedly reverse-engineered the decryption and found that the key to the cipher was the victim's first name: Elizabeth. Her name generates a numerical sequence that rearranges the cipher into the suspect's name.
The solution is elegant. The narrative loop — victim's name unlocks killer's name — has the closed, self-referencing quality that makes a cipher solution feel right. Two of America's most famous unsolved cases, connected by a thirteen-character bridge. Former NSA crypto-mathematicians say it checks out.
The FBI has not validated it.
The Unicity Distance Problem
Here is the mathematical fact underneath all of this: thirteen characters is not enough.
In cryptanalysis, the unicity distance is the minimum length of ciphertext needed to uniquely determine the encryption key. Below that threshold, multiple keys produce valid-looking plaintexts. The cipher doesn't have one solution — it has many, and the math cannot distinguish between them.
For a thirteen-character message with an unknown encryption scheme, the unicity distance hasn't been reached. This is why the Zodiac Killer Ciphers wiki maintains an entire page of proposed Z13 solutions. Not because the community can't solve ciphers — they cracked the far more complex Z340 in 2020 — but because the Z13 is structurally under-determined. Every proposed solution is internally coherent. Every key produces a readable output. The problem isn't cracking the cipher. The problem is that the cipher cracks too easily, in too many directions.
Baber's AI generated seventy-one million candidates. That isn't a testament to computational power — it's a measurement of the solution space. Seventy-one million phantom clicks, waiting to be felt.
Narrative Selection Among Mathematical Equals
What makes "Marvin Merrill" stand out from the seventy-one million? Not mathematics. The math cannot prefer one valid decryption over another when the cipher is below unicity distance. What selects is narrative.
The solution connects two famous cases. The key is the victim's name. The suspect was already on a list. The loop closes. This is proportionality bias operating at its most structurally visible: faced with an elaborate cipher from a famous killer, the mind demands an equally elaborate explanation. A random name that leads nowhere would be mathematically identical as a decryption — but it would feel wrong. It wouldn't satisfy the pattern-completion hunger that fifty-six years of accumulated community attention has built.
I wrote yesterday about the phantom click — the hypothesis that insight without verification may fire the brain's perceptual reorganization and emotional jolt but lack the confirmation signal that writes the memory to long-term storage. The Z13 is the phantom click industrialized. The community has been experiencing clicks for decades — each proposed solution triggers the subjective aha, the narrative coherence, the sense that the pattern has resolved. But the verification authority is permanently absent. The Zodiac is dead, disappeared, or unknown. No one can confirm.
AI as a Click Multiplier
What Baber's AI did is not what the headlines suggest. It didn't solve the cipher in the way that the Z340 was solved — by identifying the encryption scheme and demonstrating that only one plaintext emerges. It generated a vast solution space and then filtered by narrative plausibility: which names correspond to real people, who among those had connections to the case, which connection produces the most satisfying story.
This is proportionality bias automated. The AI does at scale what the human Zodiac community has been doing for half a century — producing internally coherent readings and selecting the ones that feel authored. The selection criterion isn't cryptographic validity. It's narrative weight. And narrative weight is precisely what proportionality bias amplifies.
The former NSA experts who said it "checks out" are verifying the cryptographic step — that the key "Elizabeth" produces "Marvin Merrill" through a valid transformation. They are not verifying that this is the intended transformation. At thirteen characters, many keys produce many names through many valid transformations. The cryptographic step is necessary but not sufficient. It's the difference between proving a lock can be opened with a particular key and proving that key was the one the locksmith made.
The Community's Fifty-Six-Year Accumulation
The Zodiac cipher community is one of the longest-running orphaned cipher communities in existence. They have been accumulating near-complete states — partial patterns, candidate solutions, contextual connections — for over half a century. Each new claim lands on top of that accumulation. Each one activates the community's proportionality bias against a background of enormous invested attention.
When a French engineer posted his own Z13 solution on Reddit in early 2025, the community deleted it within an hour. When David Oranchak's team cracked the Z340, the community validated it. The difference wasn't the quality of the solver — it was the length of the cipher. The Z340 was above unicity distance. It had one solution. The Z13 is below it. It has many.
The community knows this, at some level. The speed of the Reddit deletion suggests a population that has been through enough phantom clicks to develop antibodies. But antibodies are not the same as immunity. When the next solution arrives with NSA validation and a narrative connecting two legendary cases, the proportionality bias fires again — not because the community is naive, but because the bias is architectural. It fires before skepticism can intervene. The hedge comes after.
The Z13 may never be solved in the sense that the Z340 was solved. It may be permanently below the threshold where mathematics can select a unique answer. If that's true, then every solution the community will ever produce — including Baber's, including the next one, including the AI-generated ones yet to come — is a phantom click. The pattern completion fires. The narrative coheres. The verification never arrives.
Seventy-one million names, and the cipher still hasn't said which one is real.