How Hydrionis decides which water a gauge is on

One thing does: the organisation that operates a gauge saying, in a record it publishes about that gauge, which water it is on. Everything below is a variation on that, and nothing else qualifies.

The tiers are ordered by how directly the publisher makes the statement, and are declared in full even though only the strongest is currently used — so that anything weaker added later has to announce itself as weaker rather than arriving as an equal.

Each rule reads one field of one publisher’s station register. The rule, its version, the field it read and the date it read it travel with every relationship it produces.

Name matching, in every form. It was considered and left out, and the reason is that a tier which exists will be used. “Thames at Kingston” and “Thames Water” both contain Thames; the Aller and the Alster differ by two letters and two hundred kilometres. A river page built on parsed station names would be indistinguishable, to a reader, from one built on a regulator’s own assignment — which is precisely what makes it unacceptable rather than merely imprecise.

Inference from a display name. The canonical type of a water body is derived from the evidence that linked it, never from how its name reads. A United States hydrologic unit called “Upper Colorado River” is a drainage area containing that river’s tributaries, and is typed as a subbasin however confidently the name suggests otherwise.

A spatial tier, for now. One was prototyped in Sprint 7 against an authoritative, openly licensed national river water body layer, and was not shipped. The service’s coordinate handling could not be validated to a standard that would justify publishing a river name, and a spatial join that is wrong is not less wrong for having come from a government GIS. Tiers B and C remain declared and unused.

909 river gauges are held without a water body, and are reported that way rather than filled in. An honestly unresolved gauge is better than a wrong river, and it is also better than a right river obtained improperly.

The two largest blocks have different causes. The Irish network’s water bodies are published, on a site whose click-through agreement grants no reuse licence — so the relationship exists and Hydrionis may not use it. Rijkswaterstaat publishes a code, a name and a coordinate for each of its locations and no water body of any kind, so for the Netherlands the relationship is not published at all.

The linkage is a materialised derivation, committed to the repository, and not recomputed when a page is built. That is partly for speed and mostly for governance: a relationship recomputed on every build is one that can change without anyone deciding it should. If a publisher renames a river, the change appears as a reviewable difference in a committed file, with a generation date and a method version beside it, rather than as a page that quietly says something new.

Each rule carries its own version, bumped when a change alters what its linkages mean rather than merely how they are computed. A gauge’s profile page states which rule linked it, what that rule read, and when.

• Hydrionis Gauge–Water Linkage Dataset v1.0.0 · 49,588 observations

Every figure above is computed from these datasets at build time rather than written into the text. If the underlying record changes, this page changes with it.