How the numbers get here

1. Geography is never substituted. Each series has one registry geography. Provinces, metropolitan areas and countries are separate levels; metropolitan labour rates (three-month averages) are never plotted with provincial monthly rates. Where Statistics Canada publishes only part of a metropolitan area (Ottawa–Gatineau CPI and housing prices), the part is labelled as such.
2. Units come from the agency’s unit of measure. Scale codes and unit text are reconciled, never multiplied together; implausible magnitudes block publication.
3. Observations, forecasts and scenarios are different kinds and never share a series. No forecast is published.
4. Failures keep the last good data. A failed refresh marks the series stale and visible; revisions are logged.
5. Licence decides publication. Data whose terms do not allow commercial redistribution are withheld. Some published series still carry conditions or are left out of paid work, so being shown here does not mean a series may be resold: see data attribution.
6. Retired source tables are blocked by the pipeline.
7. Relationships carry their evidence. 162 edges in six classes: an established mechanism or estimated effect needs a source that was read (its quote is recorded); a documented exposure or observed correlation needs a data computation; a hypothesis may not carry a number. Correlations always state that they are not causal estimates, show four pre-specified lags, and report no significance test because overlapping windows violate its assumptions.
8. Maps use authentic boundaries and one measured indicator. Natural Earth (public domain) boundaries; colour shows the latest value of a single indicator, never the effect of a relationship. Metropolitan areas are shown as points, not boundaries.

When a source revises a value that was already published, the change is logged with the date it was detected. A revision larger than the tolerance for its unit is listed for editorial attention and the documents that cite the series are flagged; the text is not rewritten automatically.

Also used: Natural Earth boundaries (public domain); fonts Newsreader, Source Sans 3 and IBM Plex Mono (SIL Open Font License 1.1), self-hosted; Federal Reserve, Statistics Canada and Bank of Canada schedule pages for the calendar (dates only).

• IMF WEO real GDP growth: estimates and projections
• IMF WEO inflation, average consumer prices: estimates and projections
• OECD composite leading indicators and other OECD series
• Bank of England Bank Rate and gilt yields
• Bank of Japan policy rate and JGB yields
• BIS credit-to-GDP gaps and debt service ratios
• US Treasury Monthly Statement and auction data

Statistics Canada gaps (a table does not publish the geography): none. Job vacancy rates for Prince Edward Island and Nunavut are suppressed in the latest release and are not published.

• Source-linked numbers. Prose never contains a typed statistic. Each number is a reference to a registered series (or a registered constant) resolved when the site is built, with its period one hover away and its source listed with the text. A typed statistic fails the build for any text beyond a draft.
• Claim types. Descriptive statements restate what a reference resolves to. Interpretation, historical context, opinion and hypotheses are labelled and carry their evidence (a relationship edge or a cited document). Forecasts are not published.
• Lifecycle. draft → data verified → editorial review → approved → published. “Approved” and “published” exist only if a human editor recorded them for the exact current text; any edit afterwards demotes the text. Published writing prepared with AI assistance says so.
• When data change. If a referenced value changes after verification, the text is flagged for review; its interpretation is never rewritten automatically.
• Corrections. Substantive corrections are logged publicly at /corrections/.