| --- |
| pretty_name: Vesuvius August 2026 Evidence |
| license: mit |
| task_categories: |
| - other |
| tags: |
| - vesuvius-challenge |
| - tifxyz |
| - geometry-processing |
| --- |
| |
| # Vesuvius August 2026 evidence |
|
|
| Reproducible evidence for two narrow Villa/TIFXYZ contributions by Aviad |
| Cohen. The source data are public Vesuvius Challenge artifacts. This dataset |
| contains metadata, hashes, measurements, scripts, and tests—not scroll imagery |
| or inferred text. |
|
|
| ## Boundary corpus study |
|
|
| A frozen scan of all 450 TIFXYZ roots in the 2026-08-02 public registry |
| measures the corpus impact of the final-quad bounds fix merged as Villa PR |
| #1264. All roots completed. The fixed domain admits 24,757 fully finite XYZ |
| boundary quads across 96 roots (49,514 exporter triangles), 0.01206% of valid |
| quads corpus-wide. This is systematic, sparse geometric recovery; no claim is |
| made that every recovered point is correct papyrus or changes downstream |
| reading. |
|
|
| ## UUID producer audit |
|
|
| All 450 current public `meta.json` objects were fetched and hashed. 277 reuse |
| the literal UUID `output_tifxyz`, including 262/262 normalized roots. A public |
| Villa writer has a matching default fingerprint, but the private production |
| command is not observable, so producer attribution is explicitly an inference. |
| The proposed remediation adds an opt-in `--uuid=<id>` while preserving the |
| historical directory-name default. |
|
|
| Each directory includes its methodology, executable script, tests where |
| applicable, and machine-readable frozen results. |
|
|
