macro_get_series

Observations for one macro series, newest first, with interpretive context.

The result is the shared envelope. data holds the series metadata you need to read the numbers — series_key, source, title, units (e.g. "percent", "USD", "index 1982-84=100"), currency, frequency, seasonal_adjustment, geography, license_tag, and the release (name + schedule_url, or null for continuous datasets) — plus the echoed start/end and the bounded observations list. Each observation has period (the first day of the represented span), period_type (date/month/quarter/year/fiscal_year/fiscal_quarter), value (an exact decimal serialized as a JSON string), and as_of (when we ingested it). meta holds source (the originating agency), as_of (the freshest value on this page), and the point-in-time mode fields; pagination holds the cursor paging state.

Set as_known_at to answer as if asked at that instant: each observation comes from the newest vintage ingested at or before it. Omit it for today's latest view. Point-in-time metadata echoes the timestamp and names the series' vintage_coverage_start.

To page, pass pagination.next_cursor back as cursor; when pagination.has_more is false, next_cursor is null and you have reached the oldest observation in the window. For a long daily series prefer narrowing start/end (or a coarser-frequency series key from macro_search_series) over paging through years of rows. U.S. agency data is public domain; Eurostat data (license_tag: cc-by-4.0) requires attribution when republished — cite "Source: Eurostat, © European Union, CC BY 4.0". Both are publicly redistributable, unlike the price tools.

Errors carry a machine code: unknown_entity for a series_key we do not hold (with Did you mean: …? suggestions when a close key exists — or call macro_search_series), out_of_coverage when as_known_at predates the series' vintage capture, and bad_parameter for input the schema can't reject (a malformed cursor or naive as_known_at).