Trust
Coverage-aware answers
A register that only covers Aalsmeer cannot tell you anything about Amstelveen. GeoAI knows, from measurement, where every dataset actually has data, and restricts its answers to that area, saying so in the scope note.
The principle
Judge only the data we have, and be exactly right on it. Missing registers and missing coverage are for the data owner to fix, not for the system to paper over. An answer is restricted to where evidence exists, and says so.
Where it applies
| Situation | What GeoAI does |
|---|---|
| Finding datasets for a named place | Searches only datasets with data in that municipality, and compares only similar registers that have data there. |
| “Not near”, “not inside”, “no record in” questions | Restricts the answer to municipalities where the other register has data, and states it. |
| “Areas with none of X” | Only considers areas where X is recorded at all, and only in the municipalities this deployment covers — a register’s few records across the border do not make that side “covered”. |
| A register that is heavily lopsided | Warns when your place is not where the register mostly records: “records N here against M there; not a complete inventory”. |
| No place in the question | Answers for the municipalities this deployment covers. National data is restricted to them, and so is a register loaded as a wider extract: the records it holds across the border are not counted, and the answer names where they were. |
| Ratios between datasets | Keeps the denominator to the area the counted dataset actually covers. |
| A place outside the covered area | Says it has no data there, unless the chosen dataset genuinely covers it. |
How coverage is known
Coverage is measured, not read from a dataset’s name. The catalogue overlays every dataset’s shapes on the official municipality and neighbourhood boundaries and stores how many features fall in each. It separates meaningful coverage from a few features spilling over a border. A dataset whose name suggests one municipality but whose data covers both is treated by what it holds. For each municipality it also stores what share of the dataset lies there: area for areas, length for lines, the count for points. A single village outline that only brushes two neighbouring municipalities therefore reads as all in its own municipality, not as spread over three.
Technical noteCoverage is re-measured on every catalogue sync. “Not measured” is never read as “covers nothing”; every consumer of coverage has a safe fallback. On the Stage, coverage is drawn as solid where a dataset meaningfully covers a municipality and dashed where only a few features spill over.
Further reading on the GeoAI blog: A layer name is a promise the data doesn’t have to keep.