The Data Layer

Architecture

1. Source registry

   647 authoritative publisher/dataset families — statistical agencies, central banks, treasuries, exchanges — each with its canonical URL, machine endpoint, licence and machine-readability score. The primary publisher is always canonical; vendor aliases (FRED, Bloomberg) are stored as aliases, never as the source of truth.

2. Series registry

   1,664 stable semantic identifiers of the form <ISO>.<DOMAIN>.<METRIC>.<FREQ>.<SA> — e.g. AUS.MACRO.REAL_GDP_QOQ.Q.SA — independent of vendor naming, so research never breaks when a vendor renames a field.

3. Binding layer

   BOUND the exact publisher code is stored (5) · DISCOVERABLE an official API exists and metadata discovery binds the exact key (498) · NEEDS BINDING manual or licensed mapping required (1161). A publisher series code is never fabricated.

4. Observations & vintages

   History is append-only. A revised figure becomes a new (series, observation date, vintage timestamp) row — the record of what was believed, and when, is preserved rather than overwritten. The same discipline as the filing chain, applied to data.

5. Transforms

   Derived views — YoY, QoQ annualised, real terms, z-scores — are computed from vintaged observations, never hand-keyed.

6. Signals

   Surprise, momentum, breadth and regime measures computed across the registry, with 73 semantic relationships (policy transmission, labour leading, credit impulse…) linking series into a research graph.

7. Thesis & decision layer

   Signals feed the research published on this site and the calls filed on the Ledger — so every conclusion can be traced back through a transform to a vintaged observation from a primary source.

Coverage

Series registry

Every canonical series in the registry. Search and filter, or click a column heading to sort.