Result: no verified new plaintext. Completed the proposed section-marker and unencrypted-rune tests, and a further test of the base-60 block as a key. This narrows only the explicitly tested models. It does not eliminate these broad cipher families.
1. Boundary resets and key advancement
Used the community transcription's explicit section markers instead of assuming a page equals a section. Visually inspected additional archived images corresponding to LP2 pages 2, 3 and 55: section-ending red marks and the within-page paragraph division on page 3 agree with the chosen parsing examples. Not every marker was checked manually.
Eight substantial marked sections contain 12,947 runes in total; a separate nine-rune segment is too short for reliable language ranking and was excluded. The eight sections span the rune-bearing unsolved corpus, including sections that cross page boundaries. Non-runic blocks remain outside the rune cipher tests.
Tested six counter-reset policies: continuous, line, page, paragraph, clause and word. Also tested consuming or ignoring selected punctuation delimiters, consuming or skipping all ciphertext F symbols, and including or removing the first clause of each paragraph. Removing the first clause is a hypothesis, not a claim that every such clause is a title. Eight arithmetic sequences and offsets 0-64 were tested, with both signs and normal/reversed rune values.
798,720 candidate settings evaluated. Counts include overlapping transforms.
Each candidate was ranked using only the first 60% of the target's runes. Its remaining 40% was then scored without selecting parameters on that suffix. The best prefix-ranked output for every section remained unreadable in the continuation. Lower-ranked saved outputs are also failed candidates, not proposed translations.
Validation: encrypted the held-out known plaintext of LP2 page 57 using a prime-minus-one stream, offset 7, resetting at clauses. The search recovered the exact plaintext and offset among its leading candidates. This verifies the tested implementation path, not its power against every real cipher.
2. Selectively unencrypted F runes
The previous search handled no skips, all ciphertext F symbols, or one chosen F. This pass allowed multiple individually selected F skips using a beam search. At an F the search branches: either decode normally and advance the key, or leave F unchanged and do not advance. It retains eight candidate histories per parameter setting. All recovered paths were independently replayed to check counter handling.
For each target, tested 528 combinations of stream, starting offset and signs under three reset policies. Streams: primes minus one, Fibonacci, integer totients and cumulative primes. Offsets: 0-32. Resets: continuous, paragraph and clause. Beam pruning is approximate and can discard a correct history.
Positive control: recovered the real LP2 page-56 plaintext and its single skipped F at zero-based rune position 56 automatically. The plaintext was not part of the LP1 scoring-model training data.
Because choosing skips can manufacture English-like fragments, repeated the complete model/reset selection on 32 shuffled versions of each target. Shuffling preserves the rune inventory and text layout, while moving runes among their positions.
| Target | Best original score | Shuffled texts scoring at least as high | Empirical comparison p-value |
|---|---|---|---|
| LP2 pages 0-2 as one stream | -7.5037 | 16 of 32 | 0.515 |
| LP2 pages 54-55 as one stream | -7.4765 | 12 of 32 | 0.394 |
These are coarse calibration results, not probabilities that a cipher hypothesis is correct. Neither target stands out from the shuffled comparisons. No coherent passage was recovered. The apparent fragments are not useful evidence of decryption.
3. Verify the base-60 data and test a byte-key hypothesis
The non-rune blocks on LP2 pages 49, 50 and 51 contain 80, 104 and 72 two-character tokens: 256 bytes in total under the alphabet 0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwx.
An older archive's decimal conversion tables contain inconsistencies, so the experiment did not use those tables. Instead it independently converted the master transcription, then compared all 256 bytes against a separate hexadecimal listing in the community archive. They match exactly. The page images were visually inspected, although visually similar I/l glyphs were not independently settled at every location. Agreement between these archive files may reflect shared ancestry and is not PGP authentication.
Verified byte-block SHA-256: 3b9b07d9a26e6d55c432d94d2661fdff3c2b348daed06821f2bdb23184a4b290.
Tested the block as a cyclic additive rune key modulo 29, using all 256 starting positions, eight row/column reading orders of its 32-by-8 matrix, three reset policies, both signs and normal/reversed rune values, and no F skips versus skipping all ciphertext F symbols.
393,216 candidate settings evaluated. Parameters were selected using the first 60% of each substantial section, with the remaining 40% reserved for continuation assessment. No readable continuation emerged.
Validation: a separately constructed test encrypted known LP2 page-57 plaintext with the byte key at offset 17. The search recovered it exactly. The 256-byte size does not identify the block as a key, RSA ciphertext, a hash collection or any other specific format.
Interpretation
This session adds 1,191,936 direct parameter settings plus the adaptive skip experiments. More settings are not evidence of progress by themselves. The useful outcome is that these specific models were tested reproducibly, with positive controls and safeguards against scoring random fragments as a discovery.
No original Cicada plaintext, hidden webpage, creator identity or novel historical clue was found. The section marks, selective-F behavior and base-60 block are established source material.
Untested directions
- Key schedules combining prime indices and Fibonacci recurrences, rather than the simple streams tested here.
- Keys derived from authenticated puzzle texts, with fixed choices tested on independent continuations.
- Cipher rules that act at word or other structural levels rather than one rune at a time.
- Other interpretations of the 256-byte block; its role remains unproven.
Before expanding those searches, stronger independently trained rune-language scoring and more known-cipher validation would help measure false negatives. Keep each new hypothesis bounded and retain negative results.
Reproduction
Requirements: Python 3 and NumPy. Run from the extracted directory:
python boundary_pass.py
python skip_pass.py
python byte_key_pass.py
lp_core.py supplies shared source parsing and the same small LP1-trained scoring model. Results are in the corresponding JSON files. base60_verified.bin is the independently converted 256-byte block. Scripts use local source snapshots; no live sites or Tor access is required.
Sources and provenance
- Master transcription
- Published hexadecimal byte strings
- Archived page images and numbering
- Older write-up with base-60 and decimal tables
provenance.json records source file hashes and the original transcription revision. The first session report and results are retained unchanged in this package.