Developers

Observations are served as compact series: one file per metric, subject and period resolution, carrying provenance once and points as [periodStart, value] pairs. A 190-year monthly series is one file with 2,300 pairs, not 2,300 near-identical objects. The catalogue endpoints say what exists and how to address it, so you can enumerate the whole dataset without downloading it.

{
  "metricId": "precipitation-annual",
  "subjectId": "uk-england",
  "unit": "mm",
  "periodKind": "year",
  "basis": "reported",
  "sourceId": "uk-metoffice-haduk",
  "count": 190,
  "from": "1836-01-01",
  "to": "2025-01-01",
  "points": [["1836-01-01", 856.4], ["1837-01-01", 782.1]]
}

Identifiers are stable and are the contract. A metric id, geography id, entity id or dataset id does not change once published; if a record is withdrawn, its id is retired rather than reused.

Observation ids are deterministic, which is what makes re-ingestion idempotent: the same fact always gets the same id, so a re-run supersedes rather than duplicates.

Provenance survives API delivery. Every observation carries the fields below, and a derived value carries its calculation record as well.

Access classes describe what Hydrionis is willing to offer. They never widen what a publisher permitted: a paid tier is subject to exactly the same upstream licences as an anonymous visitor.

Only the public class exists today, and it is the whole dataset. The others are modelled in the access-policy engine so that the data layer already knows which outputs each class is eligible for — see licensing. There is no pricing, because nothing is being sold yet.

There are none, and that is a design consequence rather than an omission. Every endpoint is a static file: it either exists and returns 200, or it does not and your host returns 404. There is nothing to paginate because a series file is already the unit of retrieval, and nothing to rate-limit because serving a file costs Hydrionis the same whether it happens once or a million times.

If a future authenticated tier introduces keys, quotas and error envelopes, it will be documented here as a separate delivery channel over the same data — not as a replacement for this one.

All 23 current datasets carry licences that permit redistribution and commercial reuse, but that is a fact about today’s sources rather than a guarantee about future ones — check the dataset.

The API path carries a major version. Datasets carry their own semantic versions and changelogs, which is where meaningful change is recorded: a recomputation with a new method changes values without changing any endpoint. Cite the dataset version, not the date you fetched it.

Dataset catalogue · Data dictionary · Ingestion diagnostics · Data enquiries