Corpus verification

Periodic reconciliations of each country's extracted appellation set against an independent, public, licence-clear authority (i.e. not the same upstream we use as the pipeline spine). One section per country.

The goal is to catch silent drift: stale eAmbrosia cache, missing pliegos, a parser branch that dropped a record, or upstream re-numberings we'd never notice from inside the pipeline.

Spain

2026-05-17 — MAPA register reconciliation ✅

Independent authority: Ministerio de Agricultura, Pesca y Alimentación (MAPA), Listado de DOP/IGP de vinos registradas en la Unión Europea, PDF updated 2025-12-01. URL: https://www.mapa.gob.es/es/dam/jcr:f9643333-ef75-4a2f-8864-afd1ade63fd1/02_vinos.pdf Cached locally: /tmp/mapa_vinos.pdf (re-fetchable).

Check	MAPA	Pipeline	Match
Total wines	149	149	✓
DOPs	106	106	✓
IGPs	43	43	✓
`Nº expediente UE` identity	149	149	148/149
First-synonym name (diacritic-normalized)	149	149	149/149

Discrepancies — both benign:

DOP Tharsys (Valencia, Vino de Pago). MAPA's PDF carries Nº expediente = PDO-ES-02086; eAmbrosia carries PDO-ES-02980. Same wine (name=Tharsys, slug=tharsys). Likely a transposed-digit typo in MAPA's listing — eAmbrosia is the EU-authoritative source for the expediente number, so the pipeline carries the correct value.
Manzanilla-Sanlúcar de Barrameda / Manzanilla — synonym order differs between MAPA and our name field. Both forms are official.

Recipe (re-run as the pipeline ages):

# 1. Download fresh MAPA PDF + convert
curl -L 'https://www.mapa.gob.es/es/dam/jcr:f9643333-ef75-4a2f-8864-afd1ade63fd1/02_vinos.pdf' -o /tmp/mapa_vinos.pdf
pdftotext -layout /tmp/mapa_vinos.pdf /tmp/mapa_vinos.txt

# 2. Diff file-numbers
.venv/bin/python <<'PY'
import json, re
mapa = set(re.findall(r'(?:PDO|PGI)-ES-[A-Z0-9]+', open('/tmp/mapa_vinos.txt').read()))
idx = json.load(open('raw/es/pliegos-extracted/_index.json'))
ours = {v['file_number'] for v in idx.values() if not v.get('is_sub_denomination')}
print(f'MAPA: {len(mapa)}  Ours: {len(ours)}')
print(f'In MAPA only: {sorted(mapa - ours)}')
print(f'In ours only: {sorted(ours - mapa)}')
PY

Pipeline-internal spine ↔ extraction check (always run first): cardinality of raw/es/eambrosia/index.json wines[] (status=registered) should equal parent count in raw/es/pliegos-extracted/_index.json. Current: 149 = 149.

France

2026-05-17 — eAmbrosia FR-wine reconciliation ✅

Independent authority: EU eAmbrosia GI register (separate administrative pipeline from INAO's SIQO referentiel; same wines must appear in both). API: https://webgate.ec.europa.eu/eambrosia-api/api/v1/geographical-indications (full register, ~6.5 MB / ~4000 GIs, JSON). Cached locally: /tmp/eambrosia_all.json.

Check	eAmbrosia (FR + WINE + registered)	Pipeline (FR wine parents)	Match
Total wines	442	440	Δ = 6 (after parent-detection fix; was 7 before)

The Δ decomposes cleanly after manual review. No silent dropouts — every divergence is either a known modeling choice or a known data upstream gap. Findings:

#	Pattern	Wines	Action
1	Synonym / composite-name fold	Bourg ↔ "Côtes de Bourg, Bourg et Bourgeais"; Corse ↔ "Vin de Corse ou Corse"	None — same wines, INAO uses long composite, eAmbrosia uses short form. Improve matcher to compare across all `protectedNames` not just `protectedNames[0]`.
2	Bordeaux umbrella rollup	Blaye, Sainte-Foy-Bordeaux	None — eAmbrosia tracks Blaye / Sainte-Foy-Bordeaux as separate PDOs (PDO-FR-A0712, A0407); INAO rolls them under id_appellation=685 "Côtes de Bordeaux" as DGCs. Same wines, modeling difference.
3	Parent-detection bug (4 AOCs) ⚠️ Alsace (id=1), Blagny (id=136), Comté Tolosan (id=861), Fiefs Vendéens (id=1028)	Resolved 2026-05-17	Stage 02's parent-detection used strict `denomination == appellation` equality and fell back to `denoms[0]` when it failed — the fallback row was then re-emitted as a DGC, clobbering the parent's index entry. Replaced with a `_is_parent_denom` helper that folds synonym order ("Alsace" vs "Alsace ou Vin d'Alsace"), case ("Comté Tolosan" vs "Comté tolosan"), and composite forms ("Blagny" vs "Blagny ou Blagny Côte de Beaune") via the existing `candidate_keys()` normaliser. When no SIQO row matches at all (Fiefs Vendéens — all 5 rows are DGCs), a parent is synthesized from the cahier header. DGC loop now skips by chosen parent's `id_denomination_geo`, not by strict equality. The bug also hit 3 cider/spirit AOCs (id=335 Calvados Domfontais, id=553 Cidre de Bretagne, id=1268 Euskal Sagardoa) — 7 orphans total, now 0. Also fixed a latent key-collision bug where `index[id_app]` fallback could clash with a same-numeric `id_denomination_geo` from another appellation (Fiefs Vendéens id_app=1028 vs Pommard DGC "Clos de la Commaraine" id_denom=1028) — synthetic-parent index entries now use `f"app:{id_app}"` keys. See `scripts/02_extract_cahiers.py:_is_parent_denom`.
4	Missing from INAO SIQO entirely	Cabernet de Saumur (PDO-FR-A0257), Côtes de Blaye (PDO-FR-A0271)	✅ Resolved 2026-08-26 (web-research pass): both RETIRED — intentionally absent, no pinning. Cabernet de Saumur: suppressed July 2016, folded into AOC Saumur as Saumur rosé — Arrêté du 19 juillet 2016 (JORF 29/07/2016), art. 2 abrogates Décret n° 2011-1360 (abrogation banner at legifrance.gouv.fr/loda/id/JORFTEXT000024717966); INAO product page 8125 is 404. Côtes de Blaye: not claimed since 2015, removed from INAO's AOC list (national retirement by cessation, no abrogation décret found); EU cancellation application PDO-FR-A0271-CANCEL filed 13/01/2026, status "Applied" in eAmbrosia as of 2026-08-26 (likely OJ notice C/2026/2994 — EUR-Lex WAF-blocked, confirm in a browser). Both eAmbrosia `registered` rows are stale-register artifacts (the Austrian-PDO precedent).

Two wines exist in our pipeline as status=registered parents but in eAmbrosia as status=applied (not yet registered): Grés de Montpellier (PDO-FR-03288), Laudun (PDO-FR-03408). Correct behaviour — INAO publishes the cahier as soon as the French recognition lands, ahead of the EU registration. They'll flip to registered upstream when the EU register catches up.

Out-of-scope by design: eAmbrosia carries 53 FR SPIRIT GIs vs our 15. The gap (Marc d'Auvergne, Cassis de Bourgogne, Absinthe de Pontarlier, …) is fruit/herb eaux-de-vie + cassis IGPs that don't have a SIQO cahier PDF — the pipeline scope only covers spirits that flow through INAO's SIQO referentiel (Cognac, Armagnac, Calvados, Cognac sub-types, …). This is documented behaviour, not a bug.

Recipe (re-run as the pipeline ages):

# 1. Fetch full eAmbrosia register
curl -sS -H 'Accept: application/json' \
  -H 'User-Agent: open-wine-map/0.0.1 (mailto:code@devloed.com)' \
  'https://webgate.ec.europa.eu/eambrosia-api/api/v1/geographical-indications' \
  -o /tmp/eambrosia_all.json

# 2. Diff FR wines (eAmbrosia status=registered vs our parents)
.venv/bin/python <<'PY'
import json, re, unicodedata
def norm(s):
    s = unicodedata.normalize('NFKD', s).encode('ascii','ignore').decode().lower()
    return re.sub(r'[^a-z0-9]+', ' ', s).strip()
d = json.load(open('/tmp/eambrosia_all.json'))
cc = lambda r: [c if isinstance(c,str) else c.get('code') for c in (r.get('countries') or [])]
ea = {norm((r.get('protectedNames') or [''])[0].split(' ou ')[0]): r.get('fileNumber')
      for r in d if 'FR' in cc(r) and r.get('productType')=='WINE' and r.get('status')=='registered'}
idx = json.load(open('raw/inao/cahier-extracted/_index.json'))
WINE = lambda r: any(c.startswith(('Vin ','Crémant')) for c in (r.get('categories') or []))
ours = {norm(r['name'].split(' ou ')[0]): r.get('id_appellation')
        for r in idx.values() if not r.get('is_sub_denomination') and WINE(r)}
print(f'eAmbrosia: {len(ea)}  ours: {len(ours)}')
print('EA-only:', sorted(set(ea) - set(ours)))
print('Ours-only:', sorted(set(ours) - set(ea)))
PY

Portugal

2026-05-17 — eAmbrosia ↔ IVV spine cross-check ✅

No separate Portuguese national-level register exists. Unlike Spain (MAPA maintains a parallel national register that we cross-checked against eAmbrosia for the ES verification), Portugal's DGADR — Direção-Geral de Agricultura e Desenvolvimento Rural explicitly defers to the EU eAmbrosia register as the authoritative source (page last updated 2025-12-02: "Os produtos que estão a ser considerados ou que foram reconhecidos como IG constam dos registos das indicações geográficas" → links out to eAmbrosia + GIview).

So the strongest available verification is the eAmbrosia ↔ IVV spine cross-check: two distinct administrative surfaces (the EU register vs the Portuguese institute's caderno publication index) which the pipeline uses jointly but which are independently maintained at different administrative levels. They must agree by definition — any divergence flags either a stale IVV crawl, an EU register update we haven't refreshed, or a wine that's registered at the EU level but hasn't had its caderno published nationally yet.

Check	eAmbrosia (PT + WINE + registered)	IVV cadernos master indexes	Pipeline (parents)	Match
Total wines	44	44	44	✓
DOPs	30	30	30	✓
IGPs	14	14	14	✓
First-name identity (diacritic-normalized)	44	—	44	44/44

Zero discrepancies. Every PT wine in eAmbrosia has a caderno PDF on IVV, and vice versa. Sub-regiões count separately: 32 sub-regiões attached to the 44 parents (one IVV caderno per parent).

Recipe (re-run as the pipeline ages):

# 1. Refresh both spines
.venv/bin/python scripts/pt/00_fetch_data.py
.venv/bin/python scripts/pt/01_fetch_cadernos.py
.venv/bin/python scripts/pt/02_extract_cadernos.py  # rewrites _index.json

# 2. Cross-check
.venv/bin/python <<'PY'
import json, re, unicodedata
def norm(s):
    s = unicodedata.normalize('NFKD', s).encode('ascii','ignore').decode().lower()
    return re.sub(r'[^a-z0-9]+', ' ', s).strip()
ea = json.load(open('raw/pt/eambrosia/index.json'))
ea_names = {norm(w['name']): w.get('fileNumber') for w in ea['wines']}
idx = json.load(open('raw/pt/cadernos-extracted/_index.json'))['by_slug']
ours = {norm(v['name']): v for v in idx.values() if not v.get('is_sub_denomination')}
ivv = json.load(open('raw/pt/ivv/cadernos-index.json'))['entries']
print(f'eAmbrosia: {len(ea_names)}  IVV: {len(ivv)}  ours: {len(ours)}')
print(f'EA-only:   {sorted(set(ea_names) - set(ours))}')
print(f'Ours-only: {sorted(set(ours) - set(ea_names))}')
PY

Trap to watch for: scripts/pt/02_extract_cadernos.py --only <slug> rewrites _index.json from scratch and drops every record outside the --only filter from the index (per-AOC JSONs on disk are left intact, but the index becomes a partial view). Always re-run stage 02 without --only after a --only invocation, otherwise the index reports a misleadingly small parent count. Same pattern hit the FR pipeline before the 2026-05-17 fix; the PT script has the same destructive-write design and would benefit from the same hardening (out of scope here — tracked as a code follow-up).

Candidate stronger sources for future runs

Diário da República — Portuguese national gazette listing of approved cadernos. Per-publication search is possible but doesn't yield a flat count of currently-in-force designations.
IVDP (Instituto dos Vinhos do Douro e Porto) — separate institute for the Douro/Port cluster (subset of ~6 of our 44, including Douro, Porto, Moscatel do Douro, Trás-os-Montes, Beira Interior, Távora- Varosa). Useful for spot-checking that cluster only.
Wines of Portugal (trade body) — high-level marketing surface; doesn't enumerate all 44 cleanly.

Italy

2026-05-19 — initial pipeline drop ⏳ pending independent cross-check

eAmbrosia count (pipeline spine, 2026-05-19 fetch):

Bucket	Count
Total IT wine GIs (registered)	531
DOPs	412
IGPs	119

Independent authority candidates:

MASAF DOP-IGP portal — https://dopigp.politicheagricole.gov.it/en/vino Reports ~524 wines per the project plan's research. Reconcile by: scraping the portal index (HTML, no API), name-matching against eAmbrosia. Expected delta: ~7 (eAmbrosia's status=registered vs. MASAF's in-force list often diverge by a handful around recent amendments).
Qualivita (Fondazione Qualivita, a private foundation) — publishes an annual Atlante Qualivita with the canonical IT wine GI count per region. Useful for per-regione cross-check.
Federdoc — coordinating body of Italian consorzi di tutela. Member directory could surface DOP-equivalent wines.

Re-run recipe:

.venv/bin/python scripts/it/00_fetch_data.py
.venv/bin/python scripts/audit_it_coverage.py
# Then manually fetch MASAF portal HTML and count rows.

Expected drift: MASAF + eAmbrosia track each other within ~5 wines month-to-month. A larger delta would indicate either an eAmbrosia cache miss, a MASAF batch update we haven't seen, or a parser drop.

Note on coverage:

408 of 531 (76.8 %) wines have a Bétard 2022 Figshare polygon (all DOPs that existed pre-2022; newer DOPs and all 119 IGPs miss it).
129 of 531 (24.3 %) wines have a fully extracted documento unico (139 had an EUR-Lex URL; 9 failed looks_like_documento_unico; 1 missing the DOCUMENTO UNICO anchor).
392 wines (74 %) have no eAmbrosia publication URL at all — they need MASAF / Gazzetta Ufficiale fallback via stage 02f or the curator manual-overrides path.

2026-05-30 — complete-coverage pass + 7 cancelled IGTs removed ✅

Corpus 531 → 524 after filtering the 7 Abruzzo IGTs cancelled by Commission Implementing Regulations (EU) 2026/558–708 (consolidated into IGP Terre Abruzzesi). This corroborates the MASAF in-force count of ~524 noted in the 2026-05-19 entry above — the 7-wine delta between eAmbrosia (status=registered, not retroactively cleaned) and MASAF was exactly these cancellations. Registry: CANCELLED_GIS in scripts/it/00_fetch_data.py, surfaced by audit_it_coverage.py.

Map coverage after this pass: 523 / 524 IT wines carry a polygon (only Salemi — no parseable source, pending its own cancellation — stays stub-no-geometry), up from 442 / 531. The new gisco-comune/provincia/regione-union fallback resolves ~81 IGTs; spot-checked areas land within ~1 % of the true administrative km² (Umbria 8 455 vs 8 464 km², Lazio 17 214 vs 17 232, Pavia 2 972 vs 2 968). Sottozone: 0 → 38 sub-denomination records (Chianti's 7 etc.). Re-run recipe:

.venv/bin/python scripts/it/00_fetch_data.py        # applies CANCELLED_GIS
.venv/bin/python scripts/it/02f_extract_masaf.py --all --include-nonstub
.venv/bin/python scripts/it/02h_extract_regional_registers.py
.venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_it_coverage.py

2026-05-20 — completeness audit + MASAF grape-extraction fix ✅

Independent re-audit of the IT pipeline. Spine intact: 531 eAmbrosia wine GIs → 531 extracted records → 531 entries in wiki/_index.json. Geometry: 408 / 531 Figshare polygons (4 newer DOPs + all 119 IGPs miss — known v1 limitation). 15 wines carry neither a documento unico nor a MASAF sidecar (2 DOPs keep a Figshare polygon; 13 IGPs are stub-no-geometry).

Audit found the MASAF grape extraction badly under-performing — 108 of 388 sidecars had grapes=0, and audit_it_coverage.py reported only "123 of 531 with grapes" because it reads the on-disk doc-unico extractions and ignores the stage-04 MASAF augmentation. Root cause was a parser defect, not (mostly) a vocab gap — see CURATOR_TODO.md "MASAF grape-extraction fix". After the fix: MASAF sidecars with grapes 280 → 354 of 387; residual 33 are 4 empty-Article-2, ~24 genuinely-generic IGTs, ~5 layout misses. No false-positive grape matches remain (verified: no spurious nielluccio / listan).

Re-run recipe:

.venv/bin/python scripts/it/02f_extract_masaf.py --all
.venv/bin/python - <<'PY'
import json, pathlib
sc = {p.stem: json.loads(p.read_text())
      for p in pathlib.Path("raw/it/masaf-disciplinari-extracted").glob("*.json")
      if p.stem != "_index"}
ng = lambda d: len((d.get("grapes") or {}).get("principal", [])) \
             + len((d.get("grapes") or {}).get("accessory", []))
print("with grapes:", sum(1 for d in sc.values() if ng(d) > 0), "/", len(sc))
PY
.venv/bin/python scripts/04_build_maps.py

Caveat: audit_it_coverage.py's "Wines with grapes" line still reads doc-unico extractions only — it understates true coverage by the whole MASAF-augmented set. Reading the MASAF sidecars (as the recipe above does) gives the real figure.

Austria

2026-05-21 — eAmbrosia ↔ de.wikipedia DAC-list cross-check ✅

Independent authority: German Wikipedia, Weinbau in Österreich (CC BY-SA 4.0) — independent of the eAmbrosia EU-register spine. URL: https://de.wikipedia.org/wiki/Weinbau_in_Österreich

The pipeline corpus is 32 wine GIs (29 DOP + 3 IGP), sourced from eAmbrosia (country=AT + productType=WINE + status=registered). Wikipedia independently enumerates 18 DAC appellations and 3 Weinbauregionen (Landwein / PGI: Bergland, Steirerland, Weinland).

Check	Wikipedia	Pipeline	Match
DAC appellations	18	18	✓
Landwein regions (IGP)	3	3	✓
Total DOP	—	29	—

Reconciliation of the 29 DOP (Wikipedia lists DACs, not the full g.U. set): 18 DAC + 9 generic Bundesland g.U. (Niederösterreich, Burgenland, Steiermark, Wien, Kärnten, Oberösterreich, Salzburg, Tirol, Vorarlberg) + 2 superseded names still EU-registered (Neusiedlersee-Hügelland → Leithaberg + Rosalia; Südburgenland → Eisenberg) = 29. The 2 superseded names extract as content-stubs — their only OJ-C publication is a Löschungsantrag (see CURATOR_TODO).

No discrepancies. The 18 Wikipedia DACs all appear in the corpus by name.

Re-run recipe:

.venv/bin/python scripts/at/00_fetch_data.py
.venv/bin/python - <<'PY'
import json
w = json.load(open("raw/at/eambrosia/index.json"))["wines"]
print("DOP:", sum(1 for x in w if x["kind"] == "DOP"),
      "IGP:", sum(1 for x in w if x["kind"] == "IGP"),
      "total:", len(w))
PY

Compare against the DAC list at the Wikipedia URL above.

2026-05-22 — commune-precise geometry: DAC disjointness ✅

After switching AT geometry off Bétard 2022 (whole-municipality polygons that overlap on shared communes) to commune-precise resolution from each Einziges Dokument (scripts/_lib/at/gemeinde.py), the 16 proper DACs must be mutually disjoint — Austrian wine law assigns each commune to exactly one DAC.

Check	Before (Bétard)	After (commune-union)
Südsteiermark ∩ Vulkanland Steiermark	22.4 %	0 %
Vulkanland Steiermark ∩ Weststeiermark	6.7 %	0 %
Leithaberg ∩ Neusiedlersee	15.9 %	0 %
Wagram ∩ Weinviertel	4.5 %	0 %
any proper-DAC pair	up to 22 %	0 % (all 120 pairs)
AT wines mapped	27 / 32	30 / 32

The region-wide g.U.s (Steiermark, Niederösterreich, …) and the 3 Landwein IGPs (Steirerland, Weinland, Bergland) still contain their DACs at 100 % — that is correct regulatory nesting, not an overlap. The 2 unmapped wines are the Löschungsantrag content-stubs.

Re-run recipe:

.venv/bin/python scripts/04_build_maps.py --no-tippecanoe --no-translations
.venv/bin/python - <<'PY'
import json, itertools
from shapely.geometry import shape
fc = json.load(open("wiki/map-data/appellations.geojson"))
g = {f["properties"]["slug"]: shape(f["geometry"])
     for f in fc["features"] if f["properties"].get("country") == "at"}
regional = {"burgenland", "niederosterreich", "steiermark", "wien", "karnten",
            "oberosterreich", "salzburg", "tirol", "vorarlberg",
            "weinland", "steirerland", "bergland",
            "ruster-ausbruch", "wiener-gemischter-satz"}
dacs = [s for s in g if s not in regional]
bad = [(a, b) for a, b in itertools.combinations(dacs, 2)
       if g[a].intersects(g[b])
       and g[a].intersection(g[b]).area / min(g[a].area, g[b].area) > 0.005]
print("overlapping proper-DAC pairs:", bad or "none — all disjoint")
PY

Slovenia

2026-05-22 — eAmbrosia ↔ vinorodne-dežele cross-check ✅

Independent authority: the Slovenian wine-law region/district structure — 3 vinorodne dežele (wine regions) partitioned into 9 vinorodni okoliši (districts) — independent of the eAmbrosia EU-register spine. Reference: English Wikipedia, Slovenian wine (CC BY-SA 4.0), https://en.wikipedia.org/wiki/Slovenian_wine.

The pipeline corpus is 17 wine GIs (14 DOP + 3 IGP), sourced from eAmbrosia (country=SI + productType=WINE + status=registered).

Check	Expected	Pipeline	Match
Total wines	17	17	✓
DOP	14	14	✓
IGP (= the 3 vinorodne dežele)	3	3	✓
Figshare 2022 DOP polygons	14	14	✓

Reconciliation of the 14 DOP: 9 vinorodni okoliši (Štajerska Slovenija, Prekmurje, Bizeljsko Sremič, Dolenjska, Bela krajina, Goriška Brda, Vipavska dolina, Kras, Slovenska Istra) + 5 traditional-name DOPs (Cviček, Belokranjec, Bizeljčan, Metliška črnina, Teran) = 14. The 3 IGPs are the regions themselves (Podravje, Posavje, Primorska).

The curated file_number → region map in scripts/_lib/si/region.py partitions all 17 wines cleanly across the 3 regions (7 + 5 + 2 DOP,

1 IGP each). 16 of 17 are content-stubs (no-publication) — only Cviček carries a fetchable EU single document; see CURATOR_TODO.

Re-run recipe:

.venv/bin/python scripts/si/00_fetch_data.py
.venv/bin/python scripts/audit_si_coverage.py

Compare the kind counts + region distribution against the region/ district structure at the Wikipedia URL above.

Romania

2026-05-30 — eAmbrosia ↔ ONVPV national DOC/IG register cross-check ✅

Independent authority: Oficiul Național al Viei și Produselor Vitivinicole (ONVPV), the Romanian national wine regulator — its published denumiri de origine controlată (DOC) and indicații geografice (IG) register pages (a different bureaucratic list from the EU eAmbrosia register the pipeline is built on). URLs: https://www.onvpv.ro/ro/content/caiete-de-sarcini-pentru-obtinerea-vinurilor-cu-denumire-de-origine-controlata-doc-0 (DOC) and https://www.onvpv.ro/ro/content/indicatii-geografice (IG).

Check	ONVPV	Pipeline	Match
IG (IGP)	12	12	✓ (exact, name-for-name)
DOC (DOP)	35	34	34/35 — delta explained

Explained delta: the one DOC in the ONVPV national register absent from our corpus is Strunga (a Moldavian DOC). It is not present in eAmbrosia at all (verified by scanning the full EU register, all statuses) — i.e. it is a Romania-national-only DOC never registered / protected at EU level. The pipeline's spine is the EU eAmbrosia register, so excluding Strunga is correct by scope, not a dropped record. All other 34 DOC names reconcile name-for-name.

Note on the "54" figure: earlier docs cited 54 RO wine GIs. That counted eAmbrosia administrative re-registrations of the same wine (Murfatlar, Dealu Mare, Panciu, … each carry a 2007-protected entry + later modification entries, all status=registered). Stage 00 de-duplicates by slug to 46 unique GIs (34 DOP + 12 IGP), which is what ONVPV's register also reflects (35 DOC − 1 national-only + 12 IG).

Re-runnable recipe:

.venv/bin/python scripts/audit_ro_coverage.py        # 46 wines, 0 stubs
# Independent counts: fetch the two ONVPV register pages above and
# count DOC / IG names; compare to `by_kind` in the audit header.

Bulgaria

2026-05-23 — eAmbrosia ↔ Закон за виното (5-region structure) cross-check ✅

Independent authority: the Bulgarian Wine Act (Закон за виното и спиртните напитки) partitions the country into 5 traditional wine regions (винарски район). The 2 EU PGIs are themselves named regions covering the two macro halves of the country (north / south of Stara Planina). Reference: English Wikipedia, Bulgarian wine (CC BY-SA 4.0), https://en.wikipedia.org/wiki/Bulgarian_wine.

The pipeline corpus is 54 wine GIs (52 DOP + 2 IGP), sourced from eAmbrosia (country=BG + productType=WINE + status=registered).

Check	Expected	Pipeline	Match
Total wines	54	54	✓
DOP	52	52	✓
IGP (= the 2 macro PGIs)	2	2	✓
Figshare 2022 PDO polygons	52	52	✓

Reconciliation of the 52 DOP across 5 wine regions (via the curated file_number → region map in scripts/_lib/bg/region.py):

Wine region	PDOs	PGI
Дунавска равнина (North)	21	+ 1 PGI
Черноморски район (Black Sea coast)	8	—
Розова долина (Sub-Balkan / Rose Valley)	2	—
Тракийска низина (South-Central)	17	+ 1 PGI
Долината на Струма (Southwestern)	4	—
Total	52	+ 2

The 2 PGIs map cleanly: Дунавска равнина PGI = the entire Northern wine region (21 PDOs); Тракийска низина PGI = the four southern wine regions combined (31 PDOs). Hand-verified mapping in scripts/_lib/bg/geometry.py BG_PGI_MEMBER_PDOS.

Only 3 of 54 wines (Мелник, Нова Загора, Дунавска равнина) carry a fetchable EUR-Lex ЕДИНЕН ДОКУМЕНТ; the other 51 are content-stubs (no-publication) — see CURATOR_TODO. Bétard 2022 nonetheless gives every BG PDO a polygon, so 100 % of records render on the map.

Re-run recipe:

.venv/bin/python scripts/bg/00_fetch_data.py
.venv/bin/python scripts/audit_bg_coverage.py

Compare the kind counts + region distribution against the wine-law 5-region structure at the Wikipedia URL above. Note the asymmetric PGI mapping — the south PGI subsumes 4 sub-regions while the north PGI is coextensive with one.

2026-05-30 — ИАЛВ national-spec layer: grape + terroir coverage ✅

Independent authority: the ИАЛВ / IAVV per-wine продуктова спецификация (Изпълнителна агенция по лозата и виното), published as PDFs and listed at https://eavw.com/wps/portal/executive-ialv/legislation/wines.with.pdo.and.pgi/specifications.of.wines.with.pdo.and.pgi ("Спецификации на вина със ЗНП и ЗГУ", upd. 14 Jun 2024 — 52 ЗНП + 2 ЗГУ). Official act of a state administration body → ЗАПСП Art. 4 copyright exemption. Resolved 2026-05-30 via /research-gaps; EUR-Lex cross-check independently confirmed 0 of the 51 grandfathered names have a published EU-OJ single document (only Мелник / Нова Загора / Дунавска равнина do).

The 51 grandfathered stubs were augmented from this source (stages 01c

02f → stage-04 augment → 02d/02e). Each spec is the same numbered 1–8 template: section 5 = винени сортове грозде, section 6 = Връзка с географския район (terroir).

Check	Expected	Pipeline	Match
Specs fetched (51 grandfathered)	51	51	✓
Sidecars with grapes	51	51	✓
Sidecars with terroir text ≥200 chars	51	51	✓
All 54 BG wines with ≥1 grape	54	54	✓
All 54 BG wines with terroir bullets (02d)	54	54	✓
Total principal grape slugs	—	418	—
Total terroir bullets (02d)	—	544	—

Spot-checks (smoke wines): Поморие (Шардоне, Мускат отонел, Каберне совиньон, Мерло; 2.4 KB terroir), Сухиндол (Каберне совиньон, Мерло, Гъмза→kadarka), Долината на Струма (20 varieties incl. Широка мелнишка лоза / Мелник 82). 18 BG-native crossings + 9 international synonyms added to grape_lexicon.py with researched colours; 1 residual source-typo (Шардоне Димят) handled by the parser's whitespace-split fallback.

Re-run recipe:

.venv/bin/python scripts/bg/01c_fetch_specifikacije.py
.venv/bin/python scripts/bg/02f_extract_national_specs.py --all
.venv/bin/python scripts/audit_bg_coverage.py   # see the national-spec + terroir sections

Hungary

2026-09-20 — Mura / Murai added (eAmbrosia 41 → 42) ✅

Live eAmbrosia (country=HU, productType=WINE, status=registered) lists 42 wine GIs (36 DOP + 6 PGI); the cached index had 41. The delta is Mura / Murai (PDO-HU-02817), registered by Commission Implementing Regulation (EU) 2026/1792 of 15 July 2026 (single document OJ C/2026/1833).

Check	reference	pipeline	Match
eAmbrosia wine GIs	42	42 (`raw/hu/eambrosia/index.json`)	✓
Mura single document	EUR-Lex C/2026/1833	fetched (stage 01), 12 varieties, 8 települések, §8 link text (stage 02)	✓
Region	Zalai borvidék → Balaton borrégió (the document's own wording)	`Balaton` (`_lib/hu/region.py`)	✓
Geometry	post-Bétard; §6 lists Becsehely, Csörnyeföld, Dobri, Letenye, Murarátka, Muraszemenye, Tormafölde, Zajk	`gisco-commune-union`, 8 / 8 communes matched	✓
On the map	—	42 / 42 (`audit_hu_coverage.py`)	✓
Terroir facts	—	12 (gate: 10 supported, 2 rewritten, 0 dropped), translated en/fr/es/nl + back-checked	✓

Recipe:

.venv/bin/python scripts/hu/00_fetch_data.py
.venv/bin/python scripts/hu/01_fetch_pliegos.py --only mura
.venv/bin/python scripts/hu/02_extract_pliegos.py --only mura
.venv/bin/python scripts/02b_fetch_aoc_lexicon.py --lang hu --source raw/hu/dokumentumok-extracted
.venv/bin/python scripts/hu/02d_extract_terroir_facts.py --only mura --provider anthropic
.venv/bin/python scripts/02d_verify_terroir_facts.py --only mura
.venv/bin/python scripts/hu/02e_translate_terroir_facts.py --only mura --provider anthropic
.venv/bin/python scripts/02e_verify_terroir_facts.py --only mura
.venv/bin/python scripts/hu/03_generate_wiki.py && .venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_hu_coverage.py

2026-05-30 — eAmbrosia ↔ Agrárminisztérium termékleírás register cross-check ✅

Independent authority: the Agrárminisztérium (Hungarian Ministry of Agriculture) wine-sector portal boraszat.kormany.hu/termekleirasok2 publishes the national termékleírás (product specification) for every Hungarian OEM/OFJ — the regulator's own per-appellation register, a different bureaucratic list from the EU eAmbrosia register the pipeline is built on. HNT (Hegyközségek Nemzeti Tanácsa, hnt.hu) delegates to this index. Public official act (Szjt. 1999. évi LXXVI. tv. §1(4) — úrhivatalos exemption).

Check	termékleírás register	Pipeline	Match
15 grandfathered flagships have a published spec	15	15	✓ (each fetched + sha-pinned in `raw/hu/national-specs/manifest.json`)
eAmbrosia wine GIs (corpus spine)	—	41 (35 DOP + 6 PGI)	—
On the map	—	41 / 41 (100 %)	—
Wines with grapes	—	41 / 41 (1292 slugs; 16 via termékleírás)	—
Wines with terroir facts (02d)	—	41 / 41 (368 bullets)	—

(The 16th national spec is Badacsony — a non-stub whose EU document has an awkward structure that left its grape section unrouted; it is backfilled from its termékleírás via the fill-if-empty augment.)

What was independently confirmed: the 15 grandfathered wines whose eAmbrosia entry carries only a non-fetchable Ares(...) reference (Tokaj, Villány, Sopron, Szekszárd, Pannonhalma, Pécs, Bükk, Somlói, Nagy-Somló, Balatonfüred-Csopak, Csongrád, Balatonboglár, Káli, + the Balatonmelléki and Zemplén PGIs) each resolve to a published termékleírás PDF on the Agrárminisztérium register — independently confirming they are genuine Hungarian OEM/OFJ, not pipeline artefacts. Each spec is the EU single-document template in Hungarian (Roman-numeral outline) and parses cleanly (hu-termekleiras-v1): 15/15 yield grapes + commune list + terroir text. The other 26 wines carry a fetchable EUR-Lex EGYSÉGES DOKUMENTUM (the pipeline's primary path) and also appear on the register.

Caveat / not-yet-done: a full machine-countable cross-check of the register total (to mirror the RO 35-DOC + 12-IG name-for-name table) is blocked because the per-appellation leaf pages (…/termekleirasok2/<slug>) are JavaScript shells — the document links are JS-injected, not in the raw HTML. A curator browser pass (or the /research-gaps Claude-extension prompt path) can enumerate the index to confirm the 41-wine total against the register name-for-name.

Spot-checks (smoke wines): Tokaj (Furmint, Hárslevelű, Kabar, Kövérszőlő, Sárgamuskotály→muscat-blanc, Zéta; 27 communes; UNESCO-világörökség terroir text), Villány (29 varieties incl. Kékfrankos→blaufränkisch, Kékoportó→blauer-portugieser; 16 communes), Soltvadkerti (single-variety Ezerjó; section-8 terroir recovered after the older-template keyword fix — meszes lepelhomok, 75 %+ kvarc → filoxéra-immunitás).

Re-run recipe:

.venv/bin/python scripts/hu/01c_fetch_specifikacije.py        # fetch 15 termékleírás (curl; incomplete TLS chain)
.venv/bin/python scripts/hu/02f_extract_national_specs.py --all
.venv/bin/python scripts/audit_hu_coverage.py                 # 41/41 mapped, 0 stubs needing input
# Independent check: open boraszat.kormany.hu/termekleirasok2 in a browser
# and confirm each corpus name has a termékleírás leaf page.

Slovakia — national-spec layer (ÚPV SR špecifikácia výrobku)

Date: 2026-05-30. Source: Úrad priemyselného vlastníctva SR / Slovak Industrial Property Office (indprop.gov.sk), register of designations of origin / geographical indications — listing https://www.indprop.gov.sk/oznacenia-povodu-vyrobkov-a-zemepisne-oznacenia-vyrobkov/OPVAZOV/specifikacie-op-zo/vina-a-liehoviny, one text-layer PDF per protected name at /swift_data/source/pdf/specifikacie_op_oz/<slug>.pdf. Official act (úradné dielo, §3 Autorský zákon). Resolved via /research-gaps national-spec sk.

What was independently confirmed:

EUR-Lex negative-check — 0 of the 6 grandfathered SK wines (Východoslovenská, Južnoslovenská, Nitrianska, Malokarpatská, Karpatská perla, Slovenská PGI) have a fetchable EU-OJ JEDNOTNÝ DOKUMENT. Positive control passed: Skalický rubín (a non-stub) surfaced CELEX 32017D0713(01), OJ C 224/2017 — the search finds single documents when they exist. So the national source is the correct path.
National-source scout + fetch verification — 5 of the 6 resolve to an ÚPV PDF (HTTP 200, real sizes 100–198 KB), each carrying §f variety table + §g terroir narrative + §d commune list. The 6th, Karpatská perla (PDO-SK-A1598), is absent from the ÚPV register (404) but its canonical spec is public on the mpsr.sk mirror (download.php?fID=15089): the 1996 ÚPV Prihláška označenia pôvodu (application 0005-96), an OCR-scanned PDF with the old numbered 03.N template + flat §03.5 variety list — parsed by a second branch upv-sr-prihlaska-v1. Completed from its own spec, not aliased to Malokarpatská (which the public-source rule would forbid).

Outcome scorecard (scripts/audit_sk_coverage.py, 2026-05-31):

National-spec sidecars: 6 / 6, all with grapes + terroir ≥ 200 chars; 237 principal grape-slugs total (41–42 per modern-spec wine; 31 for Karpatská perla — the SK regions permit the full Listina registrovaných odrôd).
Terroir-fact bullets (02d, country=sk): 10 / 10 wines, 91 bullets (4 prior non-stubs + 5 modern-spec + Karpatská perla's 4 from the prihláška §03.2/03.4 granite-soil narrative), translated en/fr/es/nl.
Geometry: 10 / 10 mapped (unchanged — figshare-pdo + alias + PGI union).
Curator queue: 0 (empty — every Slovak wine extracted).
Parser upv-sr-specifikacia-v1 takes only the left Odroda column of the §f Odroda | Synonymum table, so the Pesecká leánka ↔ Feteasca regala synonym confusion never reaches the matcher. 6 VÚVV/Pospíšilová crossings folded into grape_lexicon.py (VIVC-anchored): Breslava #1671, Mília #22818, Noria #22819 (blanc); Nitranka #17282, Rudava #17283, Torysa #22419 (noir).

Re-run recipe:

.venv/bin/python scripts/sk/01c_fetch_specifikacije.py
.venv/bin/python scripts/sk/02f_extract_national_specs.py --all
.venv/bin/python scripts/sk/02d_extract_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/sk/02e_translate_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_sk_coverage.py    # 6/6 augmented, 10/10 terroir, queue=0
# Independent check: open the ÚPV listing above (5 specs) + the mpsr.sk
# specifications listing (Karpatská perla = download.php?fID=15089).

Malta (MT) — 2026-05-31

Claim: 3 Maltese wine GIs (2 DOP + 1 IGP), all 3 on the map.

eAmbrosia (country=MT + productType=WINE + status=registered): 3 wines — Malta PDO-MT-A1630, Gozo PDO-MT-A1629, Maltese Islands PGI-MT-A1631. Matches Malta's wine law (the only two PDOs are the two inhabited islands; the PGI is the archipelago umbrella). No cancelled/duplicate registrations.
Extraction: 2 / 2 PDOs parsed from the EU-OJ English SINGLE DOCUMENT (sections 1–9, ~30 varieties each incl. the indigenous Ġellewża + Girgentina); the PGI is a no-publication content-stub.
Geometry: 3 / 3 mapped — 2 figshare-pdo (Bétard 2022 carries both MT PDOs) + 1 region-pdo-union (Maltese Islands = Malta ∪ Gozo).
Styles: red / rosé / sparkling / white (from grape-colour + the oenological-practice block).
Terroir-fact bullets (02d, country=mt, Wikipedia-primary — the amendment documents carry no link-to-terroir section): 2 each for Malta / Gozo / Maltese-Islands, grounded on the en.wikipedia "Maltese wine" article, translated fr/es/nl.
Curator queue: 1 (the PGI, geometrically resolved; no action needed).

Re-run recipe:

.venv/bin/python scripts/mt/00_fetch_data.py        # 3 wines (DOP=2, IGP=1)
.venv/bin/python scripts/mt/01_fetch_pliegos.py     # ok=2, no-pub=1
.venv/bin/python scripts/mt/02_extract_pliegos.py   # extracted=2 stubs=1
.venv/bin/python scripts/02b_fetch_aoc_lexicon.py --lang en --source raw/mt/dokumente-extracted
.venv/bin/python scripts/mt/02d_extract_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/mt/02e_translate_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/mt/03_generate_wiki.py
.venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_mt_coverage.py       # 3 wines, 2 figshare + 1 union, 0 open stubs
# Independent check: eAmbrosia register UI, filter country=Malta + Wine.

Cyprus (CY) — 2026-05-31

Claim: 11 Cypriot wine GIs (7 PDO + 4 PGI), all 11 on the map with grapes; 8 with terroir facts.

eAmbrosia (country=CY + productType=WINE + status=registered): 11 wines — 7 PDO (PDO-CY-A1622 Κουμανδαρία/Commandaria, A1623 Κρασοχώρια Λεμεσού - Αφάμης, A1624 … - Λαόνα, A1625 Βουνί Παναγιάς – Αμπελίτης, A1626 Λαόνα Ακάμα, A1627 Πιτσιλιά, A1628 Κρασοχώρια Λεμεσού) + 4 PGI (PGI-CY-A1618 Πάφος, A1619 Λεμεσός, A1620 Λάρνακα, A1621 Λευκωσία). The 4 PGIs are the island's wine districts; matches Cyprus wine law.
Geometry: all 7 PDOs in Bétard 2022 EU_PDO.gpkg (figshare-pdo); the 4 PGIs union the GISCO LAU CY communities by district GISCO_ID digit (gisco-district-union). 11/11 on the map, 0 stubs.
National-spec source: moa.gov.cy Department of Agriculture τεχνικός φάκελος (Greek EU single-document PDF), one per wine. Stage 01c scrapes the «Αμπελουργία / Οινολογία» listing page (WAF-free) and name-matches all 11; stage 02f parses (pdftotext, OCR-ell fallback for the 3 image-only scans). 11/11 with grapes (226 slugs); 11/11 with a terroir-source section (the 3 image-only scans were swapped for the text-layer eAmbrosia attachment PDFs — see CURATOR_TODO 2026-05-31).
Grapes spot-check: Commandaria = Mavro + Xynisteri (the two canonical Commandaria varieties); Pitsilia = Giannoudi/Maratheftiko/Mavro/ Xynisteri/Ofthalmo. Native Cypriot varieties folded into the lexicon with VIVC/wein.plus-researched colours.
Terroir-fact bullets (02d, country=cy, Greek source → en/fr/es/nl): 4–10 each for all 11 wines (73 bullets total), Anthropic batch.

Re-run recipe:

.venv/bin/python scripts/cy/00_fetch_data.py        # 11 wines (DOP=7, IGP=4)
.venv/bin/python scripts/cy/01_fetch_pliegos.py     # all no-pub (EU-OJ stubs)
.venv/bin/python scripts/cy/02_extract_pliegos.py   # 11 stubs
.venv/bin/python scripts/cy/01c_fetch_specifikacije.py   # scrape moa.gov.cy → fetch 11 PDFs
.venv/bin/python scripts/cy/02f_extract_national_specs.py --all   # grapes/terroir (OCR fallback)
.venv/bin/python scripts/02b_fetch_aoc_lexicon.py --lang el --source raw/cy/dokumenti-extracted
.venv/bin/python scripts/cy/02d_extract_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/cy/02e_translate_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/cy/03_generate_wiki.py
.venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_cy_coverage.py       # 11 wines, 7 figshare + 4 district-union, 11/11 grapes
# Independent check: eAmbrosia register UI, filter country=Cyprus + Wine.

United Kingdom (GB) — 2026-09-06

Independent cross-check of the corpus count against the regulator's own register, which for the UK is not eAmbrosia: since Brexit the authority is the GOV.UK "protected food and drink names" register run by DEFRA.

Source of truth queried directly (not via the cached index): https://www.gov.uk/api/search.json?filter_format=protected_food_drink_name&filter_register=wines&filter_country_of_origin=united-kingdom → 7 UK wine entries, of which 6 status=registered and 1 status=applied-for (The Crouch Valley, PDO, applied 2023-03-06 — still in assessment, correctly excluded from the corpus).
Corpus: 6 wines — 4 PDO (PDO-GB-A1585 English, PDO-GB-A1587 Welsh, PDO-GB-02365 Sussex, PDO-GB-N1636 Darnibole) + 2 PGI (PGI-GB-A1589 English Regional, PGI-GB-A1590 Welsh Regional). Matches WineGB's public PDO/PGI page.
File numbers cross-checked against raw/eambrosia-register/gi-index.json (countryId=gb, qualityProductType=Wine): all 6 present there too — the five pre-Brexit names retained under the Withdrawal Agreement, Sussex added 2025-01-31 under the UK-EU agreement. The GOV.UK register itself publishes no file number, hence the bridge in stage 00.
Specification coverage: 6/6. Every registered wine ships a public product specification from assets.publishing.service.gov.uk (5 PDF + Sussex's .docx). No stub tier, no national-spec fallback layer, no WAF.
Grapes: 6/6 with a resolved roster — English/Welsh PDO 81 each, the two Regional PGIs 85 each, Sussex 28, Darnibole 1 (100% Bacchus, as its specification states). Spot-check: the Sussex sparkling roster (Chardonnay, Pinot Noir, Pinot Meunier, Arbanne, Pinot Gris, Pinot Blanc, Petit Meslier, Pinot Noir Précoce) matches the specification's §7.2 verbatim.

Geometry: 6/6 on the map, all from ONS Open Geography (OGL v3.0) — Bétard 2022 is an EU PDO layer and has no PDO-GB-* rows. Areas cross-checked against published administrative figures:

record	resolved km²	published km²
English / English Regional	130 522	130 279 (England)
Welsh / Welsh Regional	20 790	20 779 (Wales)
Sussex	3 788	3 782 (E. Sussex 1 709 + W. Sussex 1 991 + Brighton & Hove 82)
Darnibole	0.061	"whole 5 hectare area" (its specification)

Darnibole placement check (its boundary is reconstructed — see CLAUDE.md): the resolved centroid sits 502 m from the ONS centroid of Camel Valley's postcode PL30 5LG and 918 m from Nanstallon village, and falls inside the English PDO polygon. An EU-DEM 25 m transect confirms the parcels sit at 47–80 m on a ~13 % south-facing slope with the valley floor at 11–14 m immediately south — matching the specification's "steep south facing slope" bounded by the River Camel's old bed.
Terroir-fact bullets (02d, country=gb, English source → fr/es/nl): 51 bullets across 6/6 wines (5–10 each), Anthropic batch; 49 of 51 grounded in the specification itself (provenance=cahier), which is the expected shape for a cahier-primary country.

Re-run recipe:

.venv/bin/python scripts/gb/00_fetch_data.py        # 6 registered (+1 pending), ONS boundaries
.venv/bin/python scripts/gb/01_fetch_specs.py       # 6 specs (5 pdf + 1 docx)
.venv/bin/python scripts/gb/02_extract_specs.py     # 6/6 extracted, 0 stubs
.venv/bin/python scripts/02b_fetch_aoc_lexicon.py --lang en --source raw/gb/specs-extracted
.venv/bin/python scripts/gb/02d_extract_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/gb/02e_translate_terroir_facts.py --batch --provider anthropic
.venv/bin/python scripts/gb/03_generate_wiki.py
.venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_gb_coverage.py       # 6 wines, 0 problems
# Independent check: https://www.gov.uk/protected-food-drink-names
#   → filter Register = "Wines", Country of origin = "United Kingdom".

2026-09-07 — traditional terms vs the national rosters (IT MASAF elenco, ES MAPA listado)

Independent cross-check of the national_term axis introduced with the two-axis naming layer (scripts/_lib/gi_terms.py, scripts/_lib/traditional_terms.json).

check	corpus (parents, wines)	reference	result
IT DOCG	79	MASAF Elenco alfabetico dei vini DOP agg. 18.03.2026: 79 DOCG rows	✅ equal
IT DOC	333	elenco 332 + Valtènesi (registered 2026-03-18, Reg. (EU) 2026/572, pinned)	✅ equal
IT IGT	111	eAmbrosia IT PGI records (112 incl. the Salemi stub not on the map)	✅
ES DOCa/DOQ	1 + 1	MAPA listado 2 July 2026: DOCa 2 (Rioja, Priorat; Priorat shown as DOQ)	✅
ES Vino de Pago	28	listado VP 27 + Urbezo (VP per MAPA 2024-10-25, listado still DO; pinned)	✅
ES Vino de Calidad	7	listado VC 7	✅
ES DO	69	listado DO 70 − Urbezo	✅
ES Vino de la Tierra	43	listado VT 43 (all 43 ES PGIs)	✅
AT DAC	18	BML DAC-Verordnungen page 17 + Wagram (RIS)	✅
sub == parent	1,242 subs	every sub-denomination's (eu_scheme, national_term) equals its parent's	✅ 0 mismatches
kind invariant	AOC 1404 / AOP 1 / DOP 944 / EDV 28 / IGP 537	same counts in `appellations{,-villages}.geojson` before and after	✅ unchanged

Recipe:

.venv/bin/python scripts/it/00_fetch_data.py        # refreshes raw/it/masaf-elenchi/ (dated ServeAttachment link)
.venv/bin/python scripts/es/00_fetch_data.py        # refreshes raw/es/mapa/listado-dop-igp-vinos.pdf
.venv/bin/python scripts/04_build_maps.py
.venv/bin/python scripts/audit_gi_terms.py --strict
grep -o '"kind": *"[A-Z]*"' wiki/map-data/appellations.geojson | sort | uniq -c

FilesExpand file tree

VERIFICATION.md

Latest commit

History

VERIFICATION.md

File metadata and controls

Corpus verification

Spain

2026-05-17 — MAPA register reconciliation ✅

France

2026-05-17 — eAmbrosia FR-wine reconciliation ✅

Portugal

2026-05-17 — eAmbrosia ↔ IVV spine cross-check ✅

Candidate stronger sources for future runs

Italy

2026-05-19 — initial pipeline drop ⏳ pending independent cross-check

2026-05-30 — complete-coverage pass + 7 cancelled IGTs removed ✅

2026-05-20 — completeness audit + MASAF grape-extraction fix ✅

Austria

2026-05-21 — eAmbrosia ↔ de.wikipedia DAC-list cross-check ✅

2026-05-22 — commune-precise geometry: DAC disjointness ✅

Slovenia

2026-05-22 — eAmbrosia ↔ vinorodne-dežele cross-check ✅

Romania

2026-05-30 — eAmbrosia ↔ ONVPV national DOC/IG register cross-check ✅

Bulgaria

2026-05-23 — eAmbrosia ↔ Закон за виното (5-region structure) cross-check ✅

2026-05-30 — ИАЛВ national-spec layer: grape + terroir coverage ✅

Hungary

2026-09-20 — Mura / Murai added (eAmbrosia 41 → 42) ✅

2026-05-30 — eAmbrosia ↔ Agrárminisztérium termékleírás register cross-check ✅

Slovakia — national-spec layer (ÚPV SR špecifikácia výrobku)

Malta (MT) — 2026-05-31

Cyprus (CY) — 2026-05-31

United Kingdom (GB) — 2026-09-06

2026-09-07 — traditional terms vs the national rosters (IT MASAF elenco, ES MAPA listado)