Claude Fable 5.1 Solves the 370‑Year‑Old Cyphral Distich

TL;DR – What happened and why it matters

Claude Fable 5.1 cracked the 370‑year‑old Cyphral Distich by treating the 32 Proquiritations that precede the cipher as the key itself, turning the puzzle into a straightforward book‑cipher. The decoded plaintext is a royalist prayer for Charles II, confirming a long‑standing historical mystery and demonstrating that frontier LLMs can solve archival puzzles through persistent, clue‑driven search.

The Cipher and Its History

The Cyphral Distich appears at the end of Sir Thomas Urquhart’s Logopandecteision (1653) as two lines of 32 numbers each. Scholars have listed it among the top‑50 unsolved historical ciphers (Klaus Schmeh) and it has been cited in Notes and Queries (1899) and later cryptography surveys. Prior attempts used frequency analysis, substitution, and homophonic substitution, but none succeeded because they assumed an external key.

Key Insight Discovered by Fable 5.1

Fable 5.1 identified two textual clues that point to a self‑referential key:

32 Proquiritations immediately precede the cipher, and Urquhart emphasizes the number 32 in the prose.
The accompanying poem promises that an honest reader will find “his own heart’s wishes, and the Author’s minde,” echoing the recurring “wish/hope” language of the Proquiritations.

“there can no number like that of two and thirty … be pitched upon” – Urquhart, Logopandecteision

Combining these clues, Fable inferred that the i‑th number in the cipher indexes a word in the i‑th Proquiritation, and the first letter of that word forms the plaintext.

Decoding Procedure

Step	Action
1	Enumerate the 32 Proquiritations in order.
2	For each cipher number n at position i (1‑based), go to Proquiritation i and select the n‑th word.
3	Take the first letter of the selected word.
4	Concatenate the 32 letters to form the first line; repeat for the second line.

The script verify_octastick.py (included in the Vals AI repo) automates this mapping and prints the resulting letters.

Resulting Plaintext

The two decoded lines are:

O GOD UPHOLD KING CHARLS THE SECOND AND
MAKE HIM THE SUPREME RULER OF THIS LAND

Both lines contain exactly 32 letters and end with the rhyming pair and / land, matching the expected distich form. The content aligns with Urquhart’s known Royalist sympathies, making the solution historically plausible.

Extending the Method to the Cyphral Octastich

Fable applied the same principle to a larger cipher in Urquhart’s The Jewel (1652), which contains 285 numbers across eight lines plus a decagram. The rule becomes:

The k‑th number indexes a word on page k of the book.
The first letter of that word is taken.

The decoded octastich (with nine letters still ambiguous) reads as an ottava rima royalist prayer, e.g.:

GREAT LORD, MANTAINE THAT REGAL FAMILIE
WHEREOF KING CHARLS THE SECOND IS THE HEAD,
AND GRANT THAT HE MAY BEARE THE SUPREME SWEIGH
WHERE ENGLISH, SCOTS AND IR[I]SH ARE BORNE AND BRED,
…
AMEN, SO BE IT.

Minor irregularities (e.g., a slip in “IRSH” and a garbled segment “CONERTHTO”) are documented in the original blog and attributed to transcription errors or printing quirks.

How the Model Was Prompted

The author gave Fable 5.1 a simple goal: solve an unsolved cipher. Additional constraints were:

Avoid ciphers with known solutions or many plausible answers.
Exclude the hardest historical ciphers (e.g., Kryptos K4).
Encourage creative, exhaustive analysis.

Fable spent 44 minutes and processed 176 k tokens, iterating over possible interpretations until the Proquiritation‑based book cipher emerged. The model’s persistence, not any novel cryptanalytic algorithm, was the decisive factor.

Community Reactions (Selected Comments)

“The answer was simple in hindsight. It just kept looking until it found it—and that persistence might show up in many other areas.” – author’s take, echoed by several commenters.

“I’m surprised Anthropic didn’t steal that announcement.” – a humorous nod to the competitive AI landscape.

“Models are going to be a boon in the hands of historians.” – highlighting the broader impact on archival research.

“It’s more about low‑hanging fruit than raw capability.” – several users noted that the cipher was tractable once the correct hint was found.

“Could current AI solve the Zodiac killer code?” – a recurring question about the limits of LLM cryptanalysis.

These comments collectively stress that the breakthrough is less about magical reasoning and more about exhaustive search guided by textual clues.

Significance for AI‑Assisted Scholarship

Proof of concept – Frontier LLMs can autonomously identify solvable historical puzzles and apply appropriate cryptographic techniques.
Scalability – The same prompting framework can be applied to other unsolved ciphers, inscriptions, or archival riddles that have been ignored due to limited human bandwidth.
Verification needs – Independent replication (deterministic scripts, source scans) remains essential to confirm that solutions are not artifacts of training data or hallucination.
Long‑tail impact – Solving many modest‑difficulty problems may cumulatively advance knowledge far more than a few high‑profile breakthroughs.

Open Questions and Future Work

Complete the Octastich – Resolve the nine ambiguous letters by consulting a physical copy of The Jewel (1652) to verify page‑offset anomalies.
Benchmark other models – Systematically test Claude Opus, GPT‑4, and emerging open‑source LLMs on a curated list of historical ciphers.
Create a public test suite – Assemble a repository of verified unsolved ciphers (with provenance) to track progress as models improve.
Assess security implications – While these ciphers are historical, the methodology raises questions about future AI‑assisted attacks on modern cryptographic schemes.

References

Urquhart, Logopandecteision (1653), pp. 412‑417, 417.
Klaus Schmeh, “Top 50 Unsolved Encrypted Messages”, 2019.
Vals AI blog post, Fable solves Cyphral Distich, 13 Sep 2026.
DOST entries for “swey” (early‑17th‑century Scots), DSL.
ESTC record R203867, CERL.
Cipherbrain discussion, scienceblogs.de (2019).