# Gazetteer vocabulary files

These `*.txt` files are loaded once at server startup and used to annotate
Whisper transcripts with correction hints of the form `Token [?Canonical]`
before the analysis LLM runs. Design: `~/.claude/plans/happy-mapping-snail.md`.

## File format

- One entry per line, UTF-8, no surrounding whitespace required.
- Lines that are blank or start with `#` are skipped.
- Case-insensitive. Duplicates across files are collapsed.
- No header row, no columns — this is a plain word list.

## Files

| File              | Domain                                | Populated?    |
| ----------------- | ------------------------------------- | ------------- |
| `anatomy.txt`     | Latin/Greek anatomical terms          | build script  |
| `substances.txt`  | Exotic active substances (INNs)       | TBD           |
| `medications.txt` | Exotic medication brand names         | TBD           |

## Build scripts

The `anatomy.txt` is produced by shell pipeline under `../scripts/`:

```
./scripts/build-anatomy.sh
```

It crawls the German Wikipedia "Kategorie:Anatomie", filters the result
against `hunspell-de-de` to remove everyday German words, and writes the
remainder here. The result is checked into git — we rebuild only when the
source vocabulary materially changes, not on every server build.

Intermediate files land under `vocab/raw/`, which is gitignored.