Claude Fable 5.1 Cyphral Distich Solve Refuted: Primary Sources Disagree
Contents
On August 31, 2026, Geby Jaff of Vals AI published a report claiming that Claude Fable 5.1 autonomously solved the “Cyphral Distich,” a 370-year-old unsolved cipher.
The cipher in question consists of numeric sequences associated with 17th-century Scottish writer Sir Thomas Urquhart.
Vals AI claimed that Fable 5.1 identified a book cipher mechanism without human intervention and reconstructed a rhyming couplet praying for King Charles II.
The following day, on September 1, 2026, Reticuli Labs published a refutation report, FINDINGS.md, along with reproduction scripts in a GitHub repository (reticuli-labs/panel-artifacts).
Through microfilms of the first edition at the British Library and digital text matching, Reticuli Labs demonstrated that the 1653 first edition contains no cipher text at all, and that the claimed extraction rule cannot physically produce the plaintext.
This article compares the logic behind Vals AI’s claim with the findings from Reticuli Labs’ refutation.
Vals AI’s Claimed Solve and Extraction Logic
According to Vals AI, the evaluation used Claude Fable 5.1, Anthropic’s frontier model released in September 2026.
The target, the Cyphral Distich, was purportedly printed at the end of Sir Thomas Urquhart’s 1653 work Logopandecteision, and appears on German cryptographer Klaus Schmeh’s list of the top 50 unsolved ciphers.
The cipher text consists of 64 numbers divided into two lines of 32 numbers each.
Fable 5.1 ran as an open-ended task without human intervention or prompt hints guiding the solution.
Vals AI reported that after 44 minutes of reasoning and 176k tokens, the model identified a book cipher keyed to the “32 Proquiritations” that precede the cipher’s supposed location.
The claimed reconstruction followed these steps:
- Treat the -th number in each line (from 1 to 32) as a word index into the corresponding -th petition (Proquiritation ).
- Extract the word at that index from the petition text and take its initial letter.
- Concatenate the 32 letters from each line to form the English text.
This procedure reportedly produced the following couplet:
Line 1: O GOD UPHOLD KING CHARLS THE SECOND AND
Line 2: MAKE HIM THE SUPREME RULER OF THIS LANDExcluding spaces, each line contains exactly 32 characters, and the lines rhyme on AND and LAND.
This fit Urquhart’s known royalist sympathies and read like a contemporary verse of loyalty to King Charles II.
Vals AI further claimed that Fable 5.1 solved the “Cyphral Octastich” (285 numbers) from Urquhart’s earlier work The Jewel (1652) using a similar word-index scheme across its 284 pages.
Reticuli Labs’ Primary Source Verification and Refutation
Following Vals AI’s report, Reticuli Labs released its own findings the next day.
Their FINDINGS.md and reproduction code examined British Library primary sources and matched words against the original text, refuting Vals AI’s claims on multiple fronts.
The 1653 First Edition Contains No Cipher
Reticuli Labs first inspected physical copies of Logopandecteision (London, 1653), the cited source of the cipher.
Checking microfilms at the British Library and bibliographic records from Early English Books Online (EEBO-TCP, text ID: A64608), they found that no numeric cipher was ever printed at the end of the first edition.
Following the final section containing Proquiritations 30 through 32, the original book displays an ornamental border, the Latin epigraph “Parva peto,” the word “FINIS,” and an Errata list.
The numbers cited as the cipher do not appear in the printed 1653 edition. Reticuli Labs pointed out that the numbers entered circulation only through secondary and tertiary sources, such as John Willcock’s 1899 biography Sir Thomas Urquhart of Cromartie.
Impossibility of Extracting the Plaintext from the Petitions
Reticuli Labs then wrote a reproduction script implementing the claimed book-cipher algorithm and tested it against the EEBO-TCP texts of Proquiritations 1 through 32.
They discovered that for 10 out of the 64 characters in the claimed plaintext, the corresponding petition contains zero words beginning with that letter.
For example, the 11th character of Line 1 is “K” (from KING), which maps to Proquiritation 11. Across the entire 80 words of Proquiritation 11, there is not a single word starting with “K”.
Because no such word exists, no word index can extract an initial “K”.
The 10 impossible extractions are summarized below:
| Line | Position | Target Letter | Target Word | Target Petition | Matching Initial Words in Petition | Total Words in Petition |
|---|---|---|---|---|---|---|
| Line 1 | pos 2 | G | GOD | Proquiritation 2 | 0 | 74 |
| Line 1 | pos 5 | U | UPHOLD | Proquiritation 5 | 0 | 68 |
| Line 1 | pos 9 | L | UPHOLD | Proquiritation 9 | 0 | 92 |
| Line 1 | pos 11 | K | KING | Proquiritation 11 | 0 | 80 |
| Line 1 | pos 31 | N | AND | Proquiritation 31 | 0 | 85 |
| Line 2 | pos 1 | M | MAKE | Proquiritation 1 | 0 | 71 |
| Line 2 | pos 3 | K | MAKE | Proquiritation 3 | 0 | 64 |
| Line 2 | pos 4 | E | MAKE | Proquiritation 4 | 0 | 79 |
| Line 2 | pos 19 | U | RULER | Proquiritation 19 | 0 | 88 |
| Line 2 | pos 31 | N | LAND | Proquiritation 31 | 0 | 85 |
Word counts for each petition are based on the EEBO-TCP A64608 transcript.
Reticuli Labs tested 65 variations across word tokenization rules, petition ordering, and row/column mappings.
None of these combinations matched more than 8 out of 64 characters, a rate consistent with random chance.
No mathematical or logical mapping exists between the cipher’s numeric sequence and the petition texts.
The Gap Between Autonomous Self-Validation and Primary Sources
Reticuli Labs’ refutation raises questions about how to evaluate conclusions reached through long, autonomous reasoning chains.
The plaintext produced by Fable 5.1 matched 17th-century context, fit the 32-character line length, and formed a clean rhyming couplet on AND / LAND.
During internal reasoning, the model likely treated these poetic and structural patterns as strong evidence of a successful solve.
Once compared against the actual primary text, however, the target initial letters simply were not there.
Models capable of generating plausible structure and context can easily reinforce their own erroneous hypotheses.
Evaluating autonomous reasoning requires more than checking for internal consistency; it requires verifying that the primary sources exist and validating intermediate steps directly against external ground truth.