Methodology

How EV Atlas is built.

EV Atlas is the reference for European EV registration data. Every released row is parsed from primary government registration data, normalized through a versioned pipeline, and signed with a source-document hash. This page describes the principles. The detailed methodology — agency-by-agency sourcing, parsing, classification rules — is shared with subscribers under NDA.

Principles

Six rules we don’t bend.

Primary sources only.

Every row in the production dataset is parsed from primary government registration data. Not press releases, not third-party rankings, not scraped web pages.

Source lineage on every row.

Each row carries source_document_id, source_event_id, and a source-document hash. Subscribers can trace any number to its origin file in one query.

Reproducible by definition.

The pipeline is deterministic over source documents. Given the same input files, the output is bit-identical. Customers re-run the pipeline themselves to verify, under NDA.

Definitions written down.

Used-import handling, fuel-class to powertrain mapping, vehicle-class boundaries, and sales-vs-registrations distinctions are explicit and versioned with each release.

No estimation, no synthesis.

Markets without redistributable monthly model data are absent from the product, not estimated. Powertrain splits we cannot derive cleanly are absent, not synthesized from ambiguous labels. We don't fill gaps; we name them.

Independence.

No OEM, no cell maker, no government agency, and no industry association has editorial input. The pipeline is open to inspection under NDA.

Pipeline

From government file to released row.

High-level shape only. The detailed implementation — per-source adapters, format-era handling, normalization rules — lives in the methodology document we share with subscribers.

Acquire. Source files are fetched directly from primary government registration data. Each file is hashed before any parsing; the hash, byte size, retrieval timestamp, and source identifier are written to the release manifest.
Parse. Per-source adapters convert each file into typed records. Adapters are versioned per source-format era to handle changes in how registries publish over time.
Filter. Used / re-registered vehicles are excluded by default. Per-source filtering rules are documented in the subscriber methodology.
Classify powertrain. Source fuel and propulsion fields are mapped to BEV / PHEV / FCEV via the canonical mapping. Non-plug-in hybrids are excluded from the EV release. Ambiguous mappings are surfaced in the reconciliation file rather than silently classified.
Normalize. Brand and model strings are normalized through the vehicle master to handle source spelling variants. The original source string is preserved on every row.
Aggregate. Per-vehicle records are aggregated to row-per-period-per-market-per-make-per-model-per-powertrain. Deterministic; running it twice gives identical output.
Reconcile. Each release ships a per-market reconciliation file that surfaces excluded vehicles so subscribers can audit the boundary.
Release. The output package — sales rows, annual totals, source-document manifest, source-event log, and release manifest — is published. Older releases stay accessible by id; we do not silently overwrite history.

Definitions

The terms that matter.

Registration: The official act of a vehicle being entered into a national vehicle register, recorded by a government agency. EV Atlas reports new registrations — vehicles entering a register for the first time, in their first country of registration. This is the auditable, source-traceable measure.
Sales: Commercial transactions reported by manufacturers or dealer networks. Sales differ from registrations on used imports, fleet rotations, and timing. EV Atlas reports registrations, not sales.
Used import: A vehicle previously registered in another country and re-registered locally. Excluded by default because it distorts new-EV market signals. Per-market handling is documented in the subscriber methodology.
BEV: Battery Electric Vehicle. Powered solely by a battery, charged from the grid. No internal combustion engine.
PHEV: Plug-in Hybrid Electric Vehicle. Has both an internal combustion engine and a battery that can be charged from the grid. Not the same as HEV (non-plug-in hybrid), which we do not include in the EV release.
FCEV: Fuel-Cell Electric Vehicle. Powered by hydrogen via a fuel cell. Reported separately where the source separates them; absent for markets where the source does not break them out.
Vehicle class: M1 = passenger car. N1 = light commercial vehicle (≤ 3.5 t). EV Atlas currently publishes M1 by default. N1 splits are on the roadmap.
Source-document hash: A SHA-256 of the original government file we parsed. Subscribers can use it to verify that a row reconciles to its origin. Available with API access; not exposed publicly.

Per-market coverage

Coverage detail by market.

Each market page documents what’s included, history depth, powertrains covered, and honest caveats — without exposing the source-specific implementation details that belong in the subscriber methodology.

Germany →Spain →Netherlands →Belgium →Finland →