eac-ledger

An open, queryable ledger of every EAEU certificate and declaration of conformity, built to show which Western-made goods are still being certified for sale in Russia, by whom, and through which loopholes.

Before a product can be sold anywhere in the Eurasian Economic Union (Russia, Belarus, Kazakhstan, Kyrgyzstan, Armenia) someone inside the Union has to obtain a certificate of conformity or register a declaration of conformity for it. Each document names the manufacturer and its country, the applicant (usually the importer), the products and their HS customs codes, the certification body, the test laboratory whose report it rests on, and its status over time.

Put together, that makes an official, state-maintained record of parallel imports. Journalists have used single documents from it before: The Insider and NZZ traced Swiss sniper ammunition to Russian importers through three certificates. But nobody has published the register as a dataset, tracked its changes, or linked it to the export-control lists. eac-ledger does that:

  • loads the complete register: 8.9 million documents from 2010 to today, from all five member states;
  • applies the daily change files and keeps every status change (suspended, terminated, resumed) as an event;
  • detects disappearances, meaning documents that drop out of a later full snapshot, so records removed after an investigation are noted rather than lost;
  • flags documents whose HS codes are on the EU/US/UK/Japan Common High Priority List of battlefield goods (50 codes, tiers 1–4B);
  • flags coalition manufacturers (EU, US, UK, Japan, Korea, Switzerland, Taiwan and other states with export restrictions on Russia) and matches 236 brand owners across semiconductors, machine tools, test equipment, bearings, optics, firearms and consumer goods;
  • normalises applicants, places Russian ones in their federal subject, and marks occupied Ukrainian territory;
  • links nominally separate firms through shared phone numbers, e-mail addresses and websites, stored only as keyed hashes.

It is a single static Go binary with one SQLite file, a CLI and a JSON API. It needs no API keys or accounts.

A research aid, not a legal determination. A conformity document shows that someone declared a product compliant for the EAEU market. It does not prove a shipment took place, that the brand owner was involved, or that any law was broken. Brand and person detection are heuristic. Verify anything that matters against the original record (the document number is in every row).

Why this source

Russia's own registry (pub.fsa.gov.ru) Geoblocked outside Russia since 2022. Its open-data download stopped updating in 2021–22, and after investigations it hid declarations for some customs codes.
The EAEU register (this tool's source) Published openly by the Eurasian Economic Commission at opendata.eaeunion.org, with no key and no geoblock. It includes the Russian national part and documents issued in Kyrgyzstan, Armenia and Kazakhstan, which is where the loophole is.
The loophole Certificates issued in Kyrgyzstan rose about sixfold between 2020 and 2024. In February 2026 Rosakkreditatsiya stopped recognising documents from certification bodies in Kyrgyzstan, Armenia and Kazakhstan, affecting Apple, ASUS, Dell, HP, Bosch, Nike and others (CNews). By October 2026, criminal cases had been opened over 15,000 certificates from 11 laboratories in Kyrgyzstan and Armenia (Izvestia). The register shows each of those documents.

The publisher keeps only about six days of daily change files, so history that is not captured is lost. eacl sync is meant to run daily.

Quick start

make build
./eacl fetch --full                 # ~10 GB, parallel and resumable
./eacl ingest data/raw/OP35.1_v1.0.1_full.zip data/raw/OP35.1_v1.0.1_delta_*.zip
./eacl stats
./eacl serve                        # http://127.0.0.1:8088/api/stats

# every day, from cron:
./eacl sync

The first load streams 64 GB of JSON in about 45–60 minutes on 8 cores and produces a database of about 40 GB. Later full snapshots can be ingested the same way. Each one updates last_seen and records any documents that have disappeared.

Commands

Command
eacl fetch [--full] [--day YYYY-MM-DD] Download the full archive and/or daily deltas. Parallel range requests, resumable, re-downloads when the publisher's ETag changes.
eacl ingest FILE.zip... Load archives. Full archives first, then deltas in date order. Re-ingesting the same file and snapshot day is a no-op; an older version never overwrites a newer one.
eacl sync Fetch and apply every delta still online.
eacl stats Headline counts.
eacl serve [--addr] JSON API.

Environment: EACL_DB, EACL_DATA, EACL_ADDR, EACL_HASH_KEY.

API

Endpoint
GET /api/stats Headline counts.
GET /api/docs?brand=&hs6=&tier=&chpl=1&coalition=1&mfr_country=&issuing=&status=&from=&to=&region=&applicant=&vanished=1&doc_id= Search documents (newest first; limit ≤ 1000, offset).
GET /api/docs/{doc_key} One document with products, HS codes, evidence and its event history.
GET /api/applicants/{applicant_id} An applicant's documents by year, brands, manufacturer countries, issuing countries, and other applicants sharing a contact.
GET /api/brands?from=2022-02-24 Brands by documents, applicants and issuing country.
GET /api/events?since=&event=status|vanished|reappeared|deleted The change feed.

Data model

docs holds one row per document (keyed by the register's stable _id). Child tables are doc_manufacturers, doc_products, doc_hs, doc_brands, doc_labs (test reports and other evidence), and contacts (keyed hashes). Then there are applicants, manufacturers, events and ingests, plus the reference tables chpl, brands, countries and regions. The schema is in internal/store/migrations/001_init.sql.

Status codes come from classifier P.TS.01.CLS.058: 01 valid, 02 suspended, 03 terminated, 04 extended, 05 resumed, 09 expired.

Privacy

The register lists individual entrepreneurs, heads of farm households and their staff by name. eac-ledger never stores officers' names, and it never stores raw phone numbers or e-mail addresses: contacts are kept as HMAC-SHA256 hashes under a secret key (data/hash.key, or EACL_HASH_KEY), so they can be compared but not reversed by enumerating phone numbers. Natural persons are detected by registration type, ОГРНИП, an "ИП"/"КФХ" prefix, or a surname–name–patronymic or initials pattern. The published dataset and the API withhold their names, IDs and addresses. Personal names in free text (product descriptions, status notes) are replaced with [лицо] ("[person]").

Published dataset

scripts/export.sh writes privacy-safe Parquet (zstd) for publication. It's published as a Hugging Face dataset at abr00kx/eac-ledger.

Reference data

  • ref/chpl.csv: Common High Priority Items List, version of February 2024 (EU/US/UK/Japan), 50 HS-6 codes. Source: European Commission.
  • ref/countries.csv: EAEU members and the coalition of states with export restrictions against Russia.
  • ref/brands.csv: 236 brand owners with match patterns. mfr scope means a brand is matched only in manufacturer names, which keeps short names such as HP or LG out of free text.

Limitations

  • Documents are not shipments. One declaration can cover a single batch or years of serial production.
  • Declared manufacturers are not verified. A Western brand on a document may be genuine, counterfeit, or a misdescription. The brand owner is usually not a party.
  • No Russian tax IDs. The EAEU copy rarely carries the applicant's INN/OGRN, so applicants are keyed by normalised name and region, and linked by shared contacts.
  • Pre-June-2019 records are informational according to the publisher. Older documents may be incomplete.
  • History starts when you start. Status changes and disappearances before the first sync are not recoverable from this source.

Development

make test     # unit tests and an end-to-end ledger lifecycle test
make bench    # transform throughput on a real delta archive

License

MIT. The register data is published by the Eurasian Economic Commission.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support