Open Wine Map

A reference wiki + map of European wine appellations, generated mechanically from public regulator data. France (INAO + JORF) is the canonical pipeline; the other 20 countries, the United Kingdom included, each have a sibling pipeline under scripts/<cc>/ sourced from the eAmbrosia EU register, the national regulator, the Swiss federal repertoire or the UK GI register. The per-country list, coverage and mechanics live in CLAUDE.md. Every per-record fact traces back to a public-source document — nothing here is hand-written narrative.

Status

The FR pipeline runs end-to-end across the full AOC/AOP/IGP corpus, emitting per-denomination markdown pages (one per appellation plus one per DGC — Muscadet sub-crus, Côtes du Rhône Villages, Alsace grands crus, Chablis premier-cru climats, etc.). The other 20 country pipelines follow the same stage layout (00 fetch → 01 fetch documents → 02 extract → 02d/02e terroir facts → 03 wiki), each with its own document source and geometry chain; coverage per country is a function of which wines have a fetchable specification, and CLAUDE.md documents every country's sources, coverage and curator workflow. Stage 04 merges all 21 streams into a single four-locale interactive map (EN / FR / ES / NL). The site is deployed at https://www.openwinemap.com.

Setup

Requires Python 3.12 and uv.

uv sync

For LLM-driven stages (machine translation in 02c/02e, terroir-fact extraction in 02d), set:

export ANTHROPIC_API_KEY=...

Each of those stages also supports --provider=ollama (local Ollama HTTP API) and --provider=manual (round-trip flow — see below).

Running the pipeline

Stages are independent and re-runnable. Each writes a manifest, so reruns are no-ops when nothing upstream changed. Run them in order from a clean checkout to rebuild wiki/ from scratch.

For the unattended happy path, scripts/run_pipeline.py drives FR + ES end-to-end (then stage 04). It forwards --provider / --workers / --model / --ollama-url to the LLM stages (02c / 02d / 02e), and supports slicing with --fr / --es / --from=STAGE / --to=STAGE. Stage names are the script path under scripts/ minus .py (e.g. 02_extract_cahiers, es/02_extract_pliegos). Use --list to preview the resolved plan. The PT pipeline is run directly (see below) — it's not yet wired into the driver.

.venv/bin/python scripts/run_pipeline.py --provider=ollama --workers=2
.venv/bin/python scripts/run_pipeline.py --fr --from=02_extract_cahiers --provider=ollama
.venv/bin/python scripts/run_pipeline.py --es --from=02_extract_pliegos --list

Caveat for stage 01: PDF downloads are content-addressed and skip when the sha is already on disk, but the INAO product-page → show_texte → BO Agri resolution walk runs on every invocation (thousands of HTTP requests at --delay 0.8). This is intentional — it's how newly-published modifying arrêtés get discovered — so a rerun against a fully-populated manifest is still chatty, just bandwidth-light.

FR pipeline:

uv run scripts/00_fetch_data.py               # public datasets       → raw/
uv run scripts/01_scrape_cahiers.py           # cahier PDFs           → raw/inao/cahiers/
uv run scripts/01b_solve_legifrance.py        # Legifrance-only AOCs via headless Chromium (optional)
uv run scripts/02_extract_cahiers.py          # PDF → JSON            → raw/inao/cahier-extracted/
uv run scripts/02b_fetch_grape_lexicon.py     # Wikipedia (grapes)    → raw/wikipedia/grapes/
uv run scripts/02b_fetch_aoc_lexicon.py       # Wikipedia (AOCs, fr)  → raw/wikipedia/aocs/fr/
uv run scripts/02b_fetch_style_lexicon.py     # Wikipedia (styles)    → raw/wikipedia/styles/
uv run scripts/02b_translate_grapes.py        # grape-tooltip i18n    → raw/translations/grapes/
uv run scripts/02b_translate_styles.py        # style-tooltip i18n    → raw/translations/styles/
uv run scripts/02g_fetch_vivc.py              # VIVC IDs + prime names → raw/vivc/
uv run scripts/02d_extract_terroir_facts.py   # cahier+wiki bullets   → raw/terroir-facts/
uv run scripts/02c_translate_summaries.py     # FR → en/es/nl         → raw/translations/summaries/
uv run scripts/02e_translate_terroir_facts.py # FR → en/es/nl         → raw/translations/terroir-facts/
uv run scripts/03_generate_wiki.py            # markdown pages        → wiki/*.md

ES pipeline (run before stage 04 so its records merge into the map):

uv run scripts/es/00_fetch_data.py             # eAmbrosia + GISCO + SIGPAC → raw/es/
uv run scripts/es/01_fetch_pliegos.py          # EU-OJ HTML pliegos → raw/es/oj-pages/
uv run scripts/es/01b_solve_waf.py             # WAF-blocked subset via headless Chromium
uv run scripts/es/02_extract_pliegos.py        # HTML → JSON        → raw/es/pliegos-extracted/
uv run scripts/es/02f_extract_national_pliegos.py --all  # national-pliego variety augmentation
uv run scripts/02b_fetch_aoc_lexicon.py --lang es --source raw/es/pliegos-extracted/
uv run scripts/es/02d_extract_terroir_facts.py
uv run scripts/es/02e_translate_terroir_facts.py
uv run scripts/es/03_generate_wiki.py          # markdown pages     → wiki/*.md

PT pipeline (run before stage 04 so its records merge into the map):

uv run scripts/pt/00_fetch_data.py             # eAmbrosia + IVV + CAOP 2025 → raw/pt/
uv run scripts/pt/01_fetch_cadernos.py         # IVV caderno PDFs   → raw/pt/ivv/cadernos/
uv run scripts/pt/02_extract_cadernos.py       # PDF → JSON         → raw/pt/cadernos-extracted/
uv run scripts/02b_fetch_aoc_lexicon.py --lang pt --source raw/pt/cadernos-extracted/
uv run scripts/pt/02d_extract_terroir_facts.py
uv run scripts/pt/02e_translate_terroir_facts.py
uv run scripts/pt/03_generate_wiki.py          # markdown pages     → wiki/*.md

Then build the combined map:

uv run scripts/04_build_maps.py             # map + tiles (FR + ES + PT merged) → wiki/index.html, wiki/{fr,es,nl}/, wiki/map-data/

To preview the built site locally:

This is the beginning of the README. Read the rest at the original. Read the full README ↗