# Session 61: Does the neighboring glyph explain the three-dot marks?

**The same AE glyph occurs beside both three-dot and four-dot punctuation.** The two compared binary glyph masks are identical, including every one of their 1,617 black pixels. Both leave five blank pixel columns before the first visible dot. AE identity alone therefore does not explain why one group has three dots.

The wider solved-page screen also finds three-dot groups after several other runes. Six of seven text triples correspond to semicolons in the saved master; the seventh exposes another physical-class discrepancy. This supports keeping three-dot punctuation distinct in our working representation. **No new plaintext was recovered.**

## Fixed scope

The [plan](typography_plan.json) covers all **17 solved text images in our retained corpus**: archive 01, 03–16, 73 and 74. The last two are LP2 pages 56 and 57. This is not a claim about whether additional solutions have appeared elsewhere.

The screen reused Session 60's small black component rules at thresholds 80, 100 and 120. Every location with three or four isolated components at any threshold was retained for manual source review. Image hashes matched the original manifest.

| Reviewed class | Locations |
|---|---:|
| Four-dot groups in text | 59 |
| Three-dot groups in text | 7 |
| Illustration fragments | 4 |
| **Total screened candidates** | **70** |

All 66 text candidates have stable counts, group boxes and individual dot boxes across the three thresholds. The four rejected candidates are within the illustration on archive 74. All candidates remain in the [review ledger](typography_review_results.json) and the contact sheets: [1](typography_screen_contact_01.png), [2](typography_screen_contact_02.png), [3](typography_screen_contact_03.png).

This is a screen for isolated components within fixed size limits. It does not count every separator or exclude marks merged with runes. The manual review is not independent OCR or a new full transcription.

## Three-dot marks are not confined to A or AE

The selected line and gap alignments were checked against the source images and the immutable master. A gap counts the runes preceding a mark on its page. Cipher-rune names here describe the printed glyphs, not necessarily their decoded values.

| Archive image | Line | Rune gap | Preceding cipher rune | Literal master mark |
|---|---:|---:|---|---|
| 11 | 2 | 25 | K | `;` |
| 12 | 10 | 171 | N | `;` |
| 12 | 10 | 182 | N | `;` |
| 13 | 2 | 27 | Y | `;` |
| 14 | 6 | 110 | T | `;` |
| 15 | 3 | 60 | O | `-` |
| 73, LP2 56 | 1 | 19 | AE | `;` |

All seven are inline groups with the same broad right-pointing arrangement. The master already represents six as semicolons. Their recurrence after K, N, Y, T, O and AE defeats a rule that requires an A/AE predecessor.

The archive-15 mark is an additional discrepancy in our retained transcription: the image shows three isolated dots where the master records an ordinary word separator. The ledger records the physical group at gap 60 without changing the master, the runes, or the known decoded words. This is a source-quality observation within our project, not a claim of priority over the solving community.

## Matched AE comparison

Two complete four-dot groups follow AE in the screened solved images: archive 03 at gap 55 and archive 04 at gap 56. The archive-04 example gives a particularly close comparison with the three-dot mark on archive 73 at gap 19.

The comparison was selected after the screen. A [separate follow-up plan](typography_followup_plan.json) records that selection: initial component inspection showed matching dimensions and black-pixel totals, after which the full aligned masks were compared.

| Measurement | Archive 04, four dots | Archive 73, three dots |
|---|---:|---:|
| AE glyph width × height | 34 × 114 pixels | 34 × 114 pixels |
| Glyph black pixels | 1,617 | 1,617 |
| Blank columns before first visible dot | 5 | 5 |
| Visible dot-group width | 40 pixels | 24 pixels |

**The aligned AE masks differ at zero pixels**, at all three thresholds. The punctuation forms differ while the preceding glyph remains identical.

![Identical AE masks beside four-dot and three-dot groups](typography_matched_AE.png)

That rules out AE shape alone as the cause of the three-dot appearance. Together with the other predecessor runes and the established semicolon transcription, the evidence favors retaining a distinct punctuation class. It does not prove the author's intended meaning or rule out every spacing-dependent rendering process.

## What remains possible about overpainting?

For the three original inline LP2 examples, on pages 38, 46 and 56, the predicted missing fourth-square position lies entirely within the neighboring rune's rectangular bounding box. A hypothetical white-backed glyph rectangle could therefore erase a square drawn earlier at that position.

That is only **geometric compatibility**. We have no source layout file or rendering history showing that the author used such a rectangle or drew elements in that order. The solved examples demonstrate that three-dot punctuation is neither confined to A/AE nor forced by the AE glyph itself. We should not turn a possible reconstruction into an asserted cause.

This preserves the earlier pixel finding: a filled fourth square is absent at the tested position. It distinguishes that observation from the unproven explanation of how the page was constructed.

## Consequence for the cipher work

Three-dot groups should remain separate from ordinary four-dot groups and from the line-break fragment pairs found in Session 60. The physical-class overlays now include two inline LP2 discrepancies, pages 38 and 46, and the solved archive-15 discrepancy.

Session 60's known-page control still applies: the prime sequence on LP2 page 56 continues across its three-dot mark. Punctuation shape alone does not justify restarting a key.

The next useful checkpoint is to apply the explicit overlays to a derived boundary table, preserving each source event and its uncertainty, then make any proposed boundary rule pass known solved controls before using it on unsolved text. There is no reason yet to launch another expanded keyword search.

## Reproduction

- [Fixed screening plan](typography_plan.json)
- [Screening script](typography_screen.py)
- [Full component results](typography_screen_results.json)
- [Post-screen matched-comparison plan](typography_followup_plan.json)
- [Review and comparison script](typography_review.py)
- [Labels, source hashes, alignments and mask results](typography_review_results.json)
- Original sources: [03.jpg](03.jpg), [04.jpg](04.jpg), [15.jpg](15.jpg), [73.jpg](73.jpg)

Run `typography_screen.py`, then `typography_review.py`, beside the retained JPEGs, `three_dot_scan.py`, `lp_core.py`, and earlier result files. Requires Python, NumPy, SciPy and Pillow. The scripts reproduce encoded manual classifications; they do not independently decide punctuation meaning.

Plan SHA-256: `b71c62fdc96e266ae49e46dbe6067974bd6298e41d8403e7342fecfa406a6356`.