# TIFXYZ UUID producer audit

TIFXYZ Doctor found that most registered normalized roots reuse the literal
UUID `output_tifxyz`. This audit distinguishes three questions:

1. **Corpus fact:** how many current registered artifacts reuse each UUID?
2. **Producer fingerprint:** which public Villa writer has defaults matching
   the emitted metadata?
3. **Safe remediation:** can the writer accept an explicit identity without
   changing its backward-compatible directory-name default?

The corpus result alone does not prove which private production command was
run. The source fingerprint is reported as an inference and kept separate from
the directly observed metadata.

## Frozen result (2026-08-02)

- 450 registered TIFXYZ roots were fetched from the current public registry.
- They contain 174 distinct metadata UUID values.
- 277/450 roots use the literal UUID `output_tifxyz`.
- The reuse is systematic in normalized artifacts: 262/262 normalized roots
  use `output_tifxyz`, versus 15/188 original roots.
- All 262 normalized roots also use scale `[0.05, 0.05]` and the same five-key
  metadata shape (`bbox`, `format`, `scale`, `type`, `uuid`).
- Among 140 segments with both variants, all 140 normalized roots use
  `output_tifxyz`; 127 of their original counterparts have a source-specific
  identity.

Every downloaded `meta.json` is represented in `uuid-audit-2026-08-02.json`
with its public URL and SHA-256. These are collection-identity observations,
not a claim that geometry or text content is duplicated.

## Producer inference and remediation

The normalized fingerprint is consistent with public
`vc_obj2tifxyz_legacy` defaults: step size 20 produces scale 0.05, while the
writer derives `uuid` from the output-directory basename. A production job
that always writes to a staging directory named `output_tifxyz` would therefore
create exactly this metadata. The public corpus does not expose its private
command line, so this remains an inference.

The isolated Villa patch in `../villa` adds `--uuid=<id>` to both current and
legacy OBJ-to-TIFXYZ converters. An explicit value wins; omitting the option
retains the historical output-directory behavior. The shared resolver has four
standalone C++ regression tests covering override, compatibility, fallback,
and empty-input rejection; all four pass under `clang++ -std=c++20 -Werror`.

A read-only GitHub issue/PR search on 2026-08-02 found no existing Villa item
matching `uuid`, `tifxyz`, or `output_tifxyz` together.

```bash
python audit_uuid_metadata.py \
  --manifest ../boundary-corpus-study/manifest-2026-08-02.json \
  --output uuid-audit-2026-08-02.json
```