# Page 23 final-step audit

## Result

The strongest reproducible semantic payload of the signed 11-by-14 matrix is:

```text
BUNU YAZAN TOSUN, OKUYANA KOSUN
```

With Turkish diacritics restored, the last word is normally read as
`koşsun`. The line is an attested Turkish toilet/wall-writing rhyme. A literal
rendering is approximately: **“May the young bull who wrote this run at the
one who reads it.”** Contemporary Turkish commentary also records a vulgar
double-entendre reading; the matrix itself does not encode the implied vulgar
verb.

## Reproduction

Take the first two or three letters of the eleven forward rows, from top to
bottom, with cuts:

```text
2 2 2 3 2 3 3 2 2 2 3
```

| Row | Signed forward row | Prefix |
|---:|---|---|
| 1 | `BURUMUTREFAMTU` | `BU` |
| 2 | `NURESUTREGUMFA` | `NU` |
| 3 | `YAPSUAZBEHIMLA` | `YA` |
| 4 | `ZANRUAZBENOMBA` | `ZAN` |
| 5 | `TOBIKOTLUBUMYO` | `TO` |
| 6 | `SUNOKURGANOZYI` | `SUN` |
| 7 | `OKUZIKUFAUSIHE` | `OKU` |
| 8 | `YABEKANSABERHO` | `YA` |
| 9 | `NAFERANSAHOTFE` | `NA` |
| 10 | `KOREMORBIZUMRO` | `KO` |
| 11 | `SUNAKIRFANEMBA` | `SUN` |

Concatenation:

```text
BU NU YA ZAN TO SUN OKU YA NA KO SUN
BUNU YAZAN TOSUN OKUYANA KOSUN
```

The independently signed mirror edition gives the same result by taking the
rightmost two or three letters of each row and reading each selected fragment
right-to-left. The known `M`/`N` mirror-publication error lies outside every
selected fragment.

Run the public-checkout reproduction:

```text
python scripts/audit_page23_final_step.py
```

## Why this is the solution rather than a lucky segmentation

- The extraction was publicly described on 2 August 2017, long before the
  account's 2025 hint.
- The five resulting words form a known Turkish rhyme attested online by
  25 May 2000, sixteen years before the book.
- Exhausting all `2^11 = 2,048` two-or-three-letter prefix choices ranks this
  segmentation first under the retained dictionary-aware model and makes it
  the only exact concatenation of the five attested words.
- The account wrote on 27 July 2025 that Page 23 contains an animal name “but
  not the fox.” `TOSUN` is the animal-bearing noun in the recovered sentence:
  a young bull or steer.
- The complete extraction survives both author-signed matrix orientations.

This establishes the semantic payload with high confidence. It does not prove
that the anonymous author regards the rhyme as every layer or as the final
reward: an author-linked 2021 post still described Page 23 as unsolved, and no
signed message explicitly declares the extraction final.

## The magic-number clue: `URUABZ`, then a later `E`

The signed source prints:

```text
2, 8, 20, 28, 50, 82, ...
```

It does **not** print `126`. The latter is a community continuation of the
nuclear magic-number sequence and also follows the proposed reading
`666666 mod (7 × 6 × 5) = 126` from the historical onion prefix.

If the forward matrix is flattened row-major and indexed one-based at the six
author-printed positions, the proposed extraction is reproducibly:

```text
URUABZ
```

Only if the later community value `126` is appended does it become:

```text
URUABZE
```

Neither output is source-selected or self-authenticating plaintext. The
MP3Stego test used only the community-extended `URUABZE` candidate and failed
without producing a file; it is not a negative test of an author-supplied
password. Neither string can replace the Turkish sentence as the solved
semantic payload or be reported as a verified next-stage key.

## Verdict

**Page 23's recoverable message is solved to a high evidentiary standard:**
`BUNU YAZAN TOSUN, OKUYANA KOŞSUN`.

**Still unproven:** whether the filler letters contain a second payload,
whether the author intended another operation after the rhyme, and whether the
six-letter `URUABZ` or its community-extended `URUABZE` ever had a role beyond
being attractive indexing artifacts.