Every series in EDS carries explicit provenance — source tier, transform note, refresh window, and confidence score. Where official data is missing or late, the portal labels the series with a caveat and exposes the fallback logic. We publish what we know and how we know it.
Data flows through four stages. Each stage emits a versioned artifact with full lineage. If a downstream stage rejects an upstream artifact, we publish the rejection — not a silent retry. Transparency is the architecture, not a feature on top.
Tier-1 official sources first — statistics offices, regulators, exchange feeds. Secondary sources and OSINT explicitly tagged with their tier and confidence ceiling.
Proprietary entity-resolution engine disambiguates duplicates and unifies identifiers across jurisdictions. Confidence scored per identifier, never per record.
Schema, null, dup, and freshness guards. Structural breaks reviewed by a regional analyst before release. Audit trail attached.
API, portal, and bulk export — each return source, last-updated, unit, and caveat badges per series. No black-box magic.
Not all sources are equal. We rank every feed on a four-tier scale, and the tier is exposed on every series. Tier-1 sources earn an unconditional confidence ceiling of 1.00; tier-4 sources are capped at 0.70 by policy regardless of volume.
| Tier | Source class | Examples | Confidence ceiling | Refresh |
|---|---|---|---|---|
| Tier I | Official primary | Statistics offices, regulators, exchange feeds, central banks | 1.00 | ≤ 4h |
| Tier II | Curated secondary | Multilateral institutions, audited industry bodies, vetted commercial | 0.95 | ≤ 12h |
| Tier III | Verified OSINT | Leak archives (cross-confirmed), trade data, satellite-derived series | 0.85 | ≤ 24h |
| Tier IV | Unverified OSINT | Single-source field reports, investigative tips, single-feed signals | 0.70 (cap) | on event |
Confidence is a function of source tier, recency, cross-confirmation count, and historical accuracy of the series. It is computed at ingestion, recomputed on every transform, and decays with staleness.
We publish a public corrections log and a per-series caveat badge. Coverage is uneven across the world by design — we go deep where institutional decisions are hardest to make, and we say so plainly where we don't.
Series held back when source feed is past its expected refresh by 2× the median interval. Confidence decays automatically until refreshed.
Series with only one Tier I/II source carry a thin badge and a confidence cap of 0.85 regardless of accuracy.
When two Tier I sources disagree by > 10%, both are surfaced with a contested badge. We do not silently pick a winner.
The full methodology document, including resolver pseudocode and confidence math, is available under NDA. Briefings include a walkthrough with a senior analyst.