Historical tokenization notice: this report uses the earlier punctuation-split word grouping. Session 42 corrects three apostrophe splits for lexical word analysis. This word-length experiment has not been rerun under the revised grouping. Read Session 42.
Result: the longest solved-to-unsolved match is seven consecutive word lengths. Both randomized comparisons routinely produce matches that long. No new plaintext or supported quotation. This changes the kind of evidence being tested: the search uses word boundaries and lengths, without fitting a cipher or guessing rune values.
The hypothesis
An unsolved passage might repeat wording from a solved page while encrypting its runes differently. If encryption retains word order, word boundaries and one plaintext rune per ciphertext rune, a copied passage must preserve its sequence of rune counts per word.
We therefore compare runes per word, not Latin letters in a transliteration. For example, THE occupies two runes and THINGS occupies three. Matching ordinary English character counts would test a different and often incorrect pattern.
The reference includes all 15 parsed LP1 text pages and the two solved LP2 pages, 56 and 57: 740 rune words, 2,977 runes and 24 eligible blocks. The target is LP2 pages 0–55: 2,899 words, 12,956 runes and 21 blocks. Headings and rune-bearing labels are included. No new decryption of the reference pages is needed to measure their original word lengths.
Words can span transcribed line and page wraps. Clauses may be crossed, but paragraph, segment and chapter marks end a block. Numeric and Latin-character material also creates a hard break, preventing phrases from being stitched across a non-runic data block. LP1 and LP2 references are not joined to one another.
The two longest matches
Every position in every eligible reference and target block was searched. Exactly two location pairs reach the maximum of seven words.
| Unsolved location | Rune lengths | Matching solved wording |
|---|---|---|
| LP2 page 25, runes 215–243 | 2, 4, 4, 2, 9, 4, 4 | BE TRUE. TEST THE KNOWLEDGE. FIND YOUR |
| LP2 page 55, runes 46–75 | 2, 7, 7, 2, 6, 3, 3 | THE TOTIENT FUNCTION IS SACRED. ALL THINGS |
Page and rune offsets are zero-based; endpoints here are inclusive. The reference wording is displayed with readable English spelling, but counts come from source runes. The second reference's direct rune rendering uses FUNKTION and SAKRED.
The first source is the solved warning, at parsed LP1 page 0, runes 47–75. The second is the solved wisdom text at parsed LP1 page 3, runes 27–56. These parsed indices should not be confused with combined-archive image numbers.
These are length matches, not decoded text. The first unsolved window has ordinary word separators where its source has clause breaks after TRUE and KNOWLEDGE. The second has an ordinary word separator where its source has a clause break after SACRED. Neither match preserves that internal clause punctuation.
There is no exact eight-word-or-longer length match under the stated block rules. This rules out literal reuse of an eligible reference passage of that length under the preservation assumptions. It does not address rewording, altered word divisions, inserted symbols, other encodings or matches crossing the excluded boundaries.
The tempting page-55 coincidence
The page-55 window ends with ᛞᛄᚢ ᛒᛖᛁ, the repeated two-word ciphertext phrase confirmed in Session 32. Its 3,3 lengths can accommodate ALL THINGS.
That is an interesting location, but it adds no verified plaintext. Many different words share those lengths. The source contains a clause break before ALL THINGS; the target does not. More decisively, the full search's longest match is entirely ordinary under the comparisons below. We do not select ALL THINGS as a crib or claim that the other occurrence on page 27 means the same thing.
Compare the full search with shuffled data
The measured statistic is the longest match anywhere against the entire fixed reference set. Each randomized trial repeats that same search. This accounts for scanning many possible quotations and target positions, rather than calculating a probability for one appealing match selected afterward.
| Target randomization | Trials | Mean longest match | Central 95% range | Trials at least seven words | Upper-tail comparison |
|---|---|---|---|---|---|
| Shuffle lengths within each block | 9,999 | 7.564 words | 6–9 | 9,691 | 0.9692 |
| Preserve adjacent length-pair counts | 999 | 7.574 words | 7–9 | 982 | 0.9830 |
The first comparison preserves each block's word count and exact inventory of word lengths. The stricter second comparison also preserves every ordered adjacent pair of lengths and both endpoints. It uses the existing last-exit-tree sampler with word lengths mapped to compact labels. Every generated block is checked for those invariants.
The two comparisons are dependent. A conservative two-comparison Bonferroni adjustment gives 1.0 for both. The raw upper-tail values use the standard plus-one Monte Carlo calculation. They are probabilities under these specific randomized comparisons, not probabilities that a quotation is genuine or that a cipher theory is true.
The stricter sampler changed 19–20 of the 21 target blocks per trial. Its outcome is not caused by every trial simply reproducing the source. Neither comparison preserves all higher-order properties of language or deliberate composition.
Verification
Two constructed rune passages retain 12- and 16-word reference length patterns while changing their rune identities. The complete parsing-and-search process locates the planted matches at their exact lengths. This verifies length-pattern detection, not plaintext or key recovery.
The longest-match implementation is also checked against a brute-force search on 100 small fixtures. Separate fixtures check words spanning line/page wraps, numeric-block barriers and accurate end coordinates for words immediately before a page boundary.
No source image was newly transcribed or authenticated in this session. The analysis uses the retained master transcription and its recorded hash. This is exploratory work on a previously examined corpus, not an independent replication.
Decision
The solved pages do not currently provide a distinctive word-length fingerprint for the unsolved text. The page-55 resemblance is too weak to justify building a cipher around ALL THINGS.
This method can test an independently motivated source passage without choosing a cipher first. Expanding it to arbitrary books would substantially increase opportunities for accidental matches, so any such extension needs a defined reference set and a full-search comparison of its own.
Reproduce
Run python -B word_length_fingerprint_audit.py beside the retained corpus helpers and transition_surrogates.py. Requires Python and NumPy. No external text, network calls or language-model fitting is used.