# Vumarii 2019: Turkish-completion control

## Scope

The image's red legend directly fixes eleven letter classes. This audit provisionally accepts seven further 2019 community readings and then enumerates every injective assignment of the five remaining classes to the eight unused Latin letters: 8P5 = 6,720 completions. Each resulting 154-letter body is scored by the repository's calibrated, held-out Turkish A23 character 5-gram model.

The test does not claim that Turkish is source-selected, or that the seven community readings are authoritative. It is a bounded test of the historically proposed Turkish surface, including tempting local matches such as `ANLAMAK` / `ETMEK`.

## Result

- Exactly `6720` completions were evaluated.
- The best candidate scores `-5.4116` bits/character, with held-out Turkish z-score `-8.70` and percentile `0.00`.
- It has no full Turkish dictionary segmentation (`None`) and is not readable Turkish.
- Among `512` glyph-position-shuffled controls, `0` maxima are at least as high as the observed optimum; empirical p = `0.0019`. This says the Layer-3 surface is more Turkish-like than a random placement of its glyph classes.
- The crucial selection-bias control keeps glyph positions and the red legend fixed but randomises all seven community readings: `3` / `512` maxima are at least as high as observed; empirical p = `0.0078`.
- Under the accepted Layer-3 key, C07 is `D`. Setting C20=K produces `ANDAMAK`, not `ANLAMAK`; the latter also requires replacing C07=D with L. The same conditional completion does produce local `ETMEK`, but it does not select the other four classes or a whole sentence.

The search does not turn the image into Turkish plaintext. A local word-looking completion is insufficient when the full 154-character text remains strongly unlike held-out Turkish and has no full Turkish segmentation. The two null families separate a real-looking phonotactic surface from the selection freedom already present in the seven community readings; neither replaces a source-selected key or readable translation.

## Highest-scoring completions

| Rank | C16 C17 C18 C20 C21 | Turkish 5-gram | z | Candidate |
| ---: | --- | ---: | ---: | --- |
| 1 | J Y S H K | -5.4116 | -8.70 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGJA GENENNE NYOP SODANE DE QUINNA LANG HLANA CAUKU ABILTO DAUK CYAHAT ANDAMAH TENGRI ITHYIN MEP GENENNE TLYOH HAPIST ETMEH DE CYED JEKARU YEDOFER` |
| 2 | K Y S H Z | -5.4324 | -8.77 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGKA GENENNE NYOP SODANE DE QUINNA LANG HLANA CAUZU ABILTO DAUZ CYAHAT ANDAMAH TENGRI ITHYIN MEP GENENNE TLYOH HAPIST ETMEH DE CYED KEZARU YEDOFER` |
| 3 | J Y S K Z | -5.4329 | -8.77 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGJA GENENNE NYOP SODANE DE QUINNA LANG KLANA CAUZU ABILTO DAUZ CYAKAT ANDAMAK TENGRI ITKYIN MEP GENENNE TLYOK KAPIST ETMEK DE CYED JEZARU YEDOFER` |
| 4 | W Y S H K | -5.4397 | -8.80 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGWA GENENNE NYOP SODANE DE QUINNA LANG HLANA CAUKU ABILTO DAUK CYAHAT ANDAMAH TENGRI ITHYIN MEP GENENNE TLYOH HAPIST ETMEH DE CYED WEKARU YEDOFER` |
| 5 | H Y S K Z | -5.4419 | -8.81 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGHA GENENNE NYOP SODANE DE QUINNA LANG KLANA CAUZU ABILTO DAUZ CYAKAT ANDAMAK TENGRI ITKYIN MEP GENENNE TLYOK KAPIST ETMEK DE CYED HEZARU YEDOFER` |
| 6 | J Y S H Z | -5.4454 | -8.82 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGJA GENENNE NYOP SODANE DE QUINNA LANG HLANA CAUZU ABILTO DAUZ CYAHAT ANDAMAH TENGRI ITHYIN MEP GENENNE TLYOH HAPIST ETMEH DE CYED JEZARU YEDOFER` |
| 7 | Z Y S H K | -5.4503 | -8.84 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGZA GENENNE NYOP SODANE DE QUINNA LANG HLANA CAUKU ABILTO DAUK CYAHAT ANDAMAH TENGRI ITHYIN MEP GENENNE TLYOH HAPIST ETMEH DE CYED ZEKARU YEDOFER` |
| 8 | W Y S K X | -5.4580 | -8.86 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGWA GENENNE NYOP SODANE DE QUINNA LANG KLANA CAUXU ABILTO DAUX CYAKAT ANDAMAK TENGRI ITKYIN MEP GENENNE TLYOK KAPIST ETMEK DE CYED WEXARU YEDOFER` |
| 9 | H Y S K X | -5.4671 | -8.90 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGHA GENENNE NYOP SODANE DE QUINNA LANG KLANA CAUXU ABILTO DAUX CYAKAT ANDAMAK TENGRI ITKYIN MEP GENENNE TLYOK KAPIST ETMEK DE CYED HEXARU YEDOFER` |
| 10 | W Y S K Z | -5.4680 | -8.90 | `TVI ULTO DA VUMARII DO CERPO TA TENGRI IGWA GENENNE NYOP SODANE DE QUINNA LANG KLANA CAUZU ABILTO DAUZ CYAKAT ANDAMAK TENGRI ITKYIN MEP GENENNE TLYOK KAPIST ETMEK DE CYED WEZARU YEDOFER` |

## Limits

- The author fixes only the red legend; the seven additional readings are contemporary secondary claims, not a signed key.
- The control tests a Turkish monoalphabetic completion only. It does not exclude another language, a non-Latin transliteration, a keyed/polygraphic layer, or a reading of the upper diagram.
- The external language model is calibrated for Turkish-like text, not proof of literal word boundaries or author intent.

## Reproduce

```powershell
& '<bundled-python>' scripts\audit_vumarii_2019_turkish_completion.py
```

Word-class source SHA-256: `81262c1ca605c686da3acba6b4ac1ea2f11196cf74e530a9c71f607d0c5c76f8`
Transcription SHA-256: `2c6eb4492a3975c034e9fb4ba75ae676acad0fc979b76e7a0463bd77170ad6b3`