# Session 16: Four inscriptions and the reversal hypothesis

Date: 19 September 2026.

**Result: corrected the layout premise; no verified new plaintext.** Full-page inspection of LP2 pages 33–39 found four copies of the numeral motif, on pages **33, 35, 38 and 39**, all in the same visible orientation. The earlier discussion focused on the two copies inspected in Session 9 and incorrectly treated them as if they were the only pair.

This is a correction to our research record, not a claim that we discovered a previously unknown puzzle feature.

## 1. What the pages actually show

| LP2 page | Archive image | Motif position | Other relevant layout |
|---|---|---|---|
| 33 | 50.jpg | Bottom, after the page's text | A new marked section begins partway down the page, before the motif. |
| 34 | 51.jpg | No motif visible | Continuous text. |
| 35 | 52.jpg | Top, before the text | The same two-row motif, in the same visible orientation. |
| 36 | 53.jpg | No motif visible | Numbered item 1 begins. |
| 37 | 54.jpg | No motif visible | Items 2, 3 and 4 appear in that order. |
| 38 | 55.jpg | Top, before the text | Text continues from the preceding page, then item 5 begins. |
| 39 | 56.jpg | Top, before the text | Text continues, followed by a separated block; the marked section ends after this page. |

The two rows consistently support the conventional integer readings **1033 above 3301**, as discussed in Session 9. The copies do not alternate their row order, mirror their wedge shapes, or visibly rotate to tell the reader to change direction.

View the four source pages:

<div class="page-grid">
<figure><a href="/files/50.jpg"><img src="/files/50.jpg" alt="LP2 page 33 with numeral motif below the text" loading="lazy"></a><figcaption>Page 33 · bottom</figcaption></figure>
<figure><a href="/files/52.jpg"><img src="/files/52.jpg" alt="LP2 page 35 with numeral motif above the text" loading="lazy"></a><figcaption>Page 35 · top</figcaption></figure>
<figure><a href="/files/55.jpg"><img src="/files/55.jpg" alt="LP2 page 38 with numeral motif above the text" loading="lazy"></a><figcaption>Page 38 · top</figcaption></figure>
<figure><a href="/files/56.jpg"><img src="/files/56.jpg" alt="LP2 page 39 with numeral motif above the text" loading="lazy"></a><figcaption>Page 39 · top</figcaption></figure>
</div>

All seven source images were inspected. They contain runic text, colored initials/punctuation, the numeral motifs and list labels, rather than a sequence of illustrations or arrows selecting a reading direction. The original images are 2400 × 3600 pixels. Binary-ink bounds were measured separately without changing the originals. The three top motifs have matching measured bounding dimensions; the bottom motif differs by one pixel in each dimension at the chosen threshold. This measurement does not establish exact pixel identity or a hidden graphical signal.

Sources: [community page grouping](https://www.cicadasolvers.com/wiki/liber-primus/lp2/cuneiform-section/), [source image repository](https://github.com/iBotPeaches/cicada_3301/tree/master/liber_primus), [new image URLs and hashes](/files/reversal_image_sources.json), [measured image bounds](/files/reversal_layout_results.json).

## Exact text boundaries

The seven-page block contains 1,680 runes. The marked section under investigation contains **1,589 runes**: it begins 91 runes into page 33 and runs to the end of page 39. These scopes must not be conflated.

Within the marked section, the four motifs partition the normal reading stream as follows:

| Portion | Runes |
|---|---:|
| Section start on page 33, before the first motif | 123 |
| After page-33 motif, before page-35 motif: page 34 | 261 |
| After page-35 motif, before page-38 motif: pages 35–37 | 737 |
| After page-38 motif, before page-39 motif: page 38 | 228 |
| After page-39 motif, to section end | 240 |

The cumulative motif offsets are 123, 384, 1121 and 1349, counted in runes from the marked section's start. They are not equally spaced. No correspondence to the literal values 1033 or 3301 is established by these positions.

In the preserved transcription, the first marker interrupts the three-rune word **ᛒᛗᛒ** after its first rune, across the page-33/page-34 break. The visible page edges agree with that transcription: page 33 ends in ᛒ and page 34 starts with ᛗᛒ before the separator. The other three motif positions fall between words. Thus the motif is not consistently a word delimiter. It also does not coincide with the marked section's beginning or ending.

## 2. What reading instruction is supported?

The reversed decimal numbers motivate considering reversal, but the layout does **not** select a unique operation. The additional copies weaken our simple two-bookend hypothesis.

The visible list labels run 1, 2, 3, 4, 5 in ordinary page order. Reversing page order would disrupt that sequence, so page-order permutations were not promoted into this test. A layered cipher could still reverse some content while leaving labels alone, but that would need additional evidence.

Repeated motifs could in principle trigger local changes in direction. We therefore tested a small exploratory family tied to the actual marker positions, alongside word and line reversal. These are explicitly hypotheses; none is established as the intended instruction.

## 3. Fixed reading-order probe

The rules below were fixed in code before inspecting their outputs. All use the same 1,589-rune marked section.

| Reading rule | Exact operation |
|---|---|
| Forward baseline | Keep the original reading stream. |
| Reverse each word | Reverse rune order within each word, joining words across physical line/page wraps first. Preserve word positions and separators. |
| Reverse each line | Reverse the character sequence within each transcribed physical line. Keep line order. |
| Reverse after every marker | Keep the initial 123-rune fragment forward; reverse the complete character sequence in each of the four following marker-defined blocks, preserving the blocks' original positions. |
| Alternate at markers | Start forward and switch direction after each motif: forward, reverse, forward, reverse, forward across the five blocks. |

Character-sequence reversal carries internal punctuation and layout markers with the reversed material. It does not reflect or reinterpret glyph shapes. In modes that move list digits, those digits travel with the text but are excluded from rune arithmetic. The word-reversal mode preserves cross-page words; marker modes may split them by construction. These details are model assumptions, not silently repaired exceptions.

For each order, test exactly three mappings of the zero-based rune index c: **c**, **28 − c**, and **2 − c**, modulo 29. These are direct reading and two fixed substitutions already found in solved LP1 material. No guessed numerical keys, additional shifts, fitted keys, selective skips, prime streams or further layers were added.

That gives **15 candidate settings: 3 forward baselines and 12 reversal variants**. Full rune streams and rendered outputs are retained in [reversal_results.json](/files/reversal_results.json).

Validation:

- Three hand-specified fixtures check word reversal across line/page boundaries, line reversal, and alternating block direction.
- Fifteen constructed controls use five copies of the known 95-rune LP2 page-57 plaintext, one for every order/mapping combination. All recover the exact 475-rune original. They validate those implemented transformations, not the real manuscript's unknown mechanism.
- Applying the known Atbash mapping to the actual LP1 warning reproduces its expected opening as a sanity check. This is not a fresh independent authentication of the source.

## Results and limits

Opening, middle and ending excerpts of all fifteen candidates were inspected. None provided a coherent passage. All full outputs remain available for audit; every character of every output was not manually reviewed.

Whole-output language scores ranged from approximately −5.388 to −5.272. The forward direct baseline was approximately −5.288. The slight ranking improvement from word reversal is not evidence of decryption. These are descriptive scores from the small existing LP1 English model, with no calibrated significance test or held-out validation in this run.

This result does not support simple reversal plus direct reading or the two tested fixed substitutions. It does not rule out reversal inside a more complex cipher, other languages, alternative block conventions, or a non-cipher purpose for the motif.

Reproduction: run [reversal_pass.py](/files/reversal_pass.py) with Python and NumPy beside the existing lp_core.py, word_core.py and master transcription. The optional [image-bound measurement](/files/reversal_layout_audit.py) additionally requires Pillow and the four original images. No network access is needed for either script.

## Decision

The reliable outcome is a corrected map of **four repeated motifs**, not a demonstrated reversal instruction. A signature or recurring section decoration remains possible; this audit cannot establish intent. The forward numbered list is a more concrete structural constraint than the reversal speculation and should be preserved in subsequent hypotheses.

Session 9's two inspected copies and the initial naming page were incomplete accounts of occurrence count. They are superseded by this seven-page audit. No verified new plaintext was found.