# Vumarii later-white-script source-alignment audit

## Verdict

**Not fully decoded.** Two later white-script images have a strong, independently
checkable source-text match to Dani Leviss, [*Anybody out there?*](https://scienceline.org/2019/01/anybody-out-there/),
published by Scienceline on 15 January 2019.  The match identifies the likely
English wording of those two images; it does **not** yield a reproducible
glyph/ligature alphabet, and it does not translate the separate signed 2019
Vumarii star-map image.

## What the raster supports

| sample | supplied line layout | candidate words | thresholds that reproduce every line count |
| --- | --- | ---: | --- |
| short | `8 / 7 / 9 / 8` | 32 | `8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32` |
| long | `14 / 15 / 13 / 10 / 11 / 6 / 13 / 12 / 9 / 13 / 13 / 15 / 7 / 13 / 11 / 15 / 10` | 200 | `4, 5, 6, 7, 8, 9` |

The short image has a wide blank-gap plateau, while the long image's matching
threshold range is narrower.  Both facts are preserved rather than collapsed
into an unqualified claim that every component is a letter.

## Why this is not a font-table decode

- The known Discord crib fixes ten Latin values in the common visual family:
  `V U M A R I T E N G`.  It does not identify every later white component or
  its context-sensitive marks/ligatures.
- The existing lightweight template diagnostic has only
  `7.4480%` self-classification on
  `913` labelled letter samples.
- The existing forced-alignment record has `1095`
  edit errors and `0.0%` exact complete
  words on its calibration run.
- Therefore the article match is source recovery, not an independently
  reproducible character-by-character transcription.

## Separate 2019 image

The PGP-signed 2019 Vumarii image remains only partially transliterated: its
154-slot body contains five unresolved glyph classes and has no verified
plaintext.  Its upper panel has a strong visual attribution to the Dunhuang
Star Map, but no author-selected traversal or extraction rule.  The later
Scienceline excerpts must not be used to backfill it as solved.

## Reproduction

Run:

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

The JSON companion records image hashes, fixed raster bands, every checked
gap threshold, candidate line counts, and the limitations of the existing
glyph-model probes.