Skip to main content

Data Quality

Data Provenance & Traceability - China EV Mart Docs

China EV Mart field-level data provenance conventions — source types, confidence levels, and verification dates for every vehicle record.

Every vehicle record on China EV Mart carries provenance metadata so readers can tell where a number came from and how much to trust it. This page is the convention document for that system (task book §7).

Machine-readable convention

The canonical field convention lives in src/data/source-master.json under the _conventions block (schema version 1, adopted 2026-09-24). It defines:

Source type enum

Every source that feeds a vehicle field is classified as exactly one of:

source_typeMeaningExample
manufacturerOfficial manufacturer datasheet / websiteHongqi official spec sheet
governmentGovernment-issued data (tax, customs, statistics)National VAT legislation
regulationRegulatory textEU Reg 2024/2754 (CVD)
customsCustoms tariff schedulesEU TARIC, WTO Tariff Facility
industryIndustry media / industry benchmarksCnEVPost, RoRo freight benchmarks
primaryPrimary source (generic alias for official channels)—
secondarySecondary / aggregated catalog dataAutohome / 懂车帝
chinaevmart_calculationComputed by China EV Mart from a documented formulalanded_cost, VAT compounding
chinaevmart_estimateChina EV Mart benchmark estimateRoRo freight, certification fees

Confidence levels

confidenceMeaning
highOfficial primary source, cross-verified
mediumOfficial + estimate mix, or single official source not yet reviewed
lowTier-3/4 media source, or awaiting verification

Verification dates

verified_date (YYYY-MM-DD) records the last authoritative verification, either per field or per record. Vehicle-level data_updated serves this role today.

Vehicle frontmatter fields (current state)

The vehicle markdown frontmatter already carries record-level provenance:

  • data_source — text description of where the record’s data comes from
  • data_updated — last update/verification date (enforced ≤ 90 days by scripts/qa_data.py DATE QA)
  • data_reviewed — human review flag
  • data_tier — tier 1–4 (1 = official primary, 4 = tertiary cross-check)
  • confidence — optional record-level confidence (high / medium / low); when absent the template derives a display value from the source mix (landed-cost estimates render as Medium)

The template on every vehicle page renders data_source, data_updated, data_reviewed and the confidence level in the Data Sources & Verification block.

Per-field provenance (future)

field_sources — an optional object mapping each frontmatter field to {source_id, source_type, confidence, verified_date} — is specified in the convention but not yet rolled out. Batch-migrating 517 vehicle files is a separate decision; until then record-level metadata remains authoritative.

Migration status

  • 8 source records in source-master.json keep their legacy source_type values; the convention block maps them to the new enum (customs tariff → customs, spec → manufacturer, catalog → secondary, tax → government, industry media → industry, tertiary → secondary).
  • No vehicle file was batch-modified as part of this step.
  • New or hand-fixed records (e.g. Hongqi Guoya powertrain, verified 2026-09-24) record the verification date in data_source / data_updated.