TL;DR: Every value on a ShowMeOnMap map comes from a live upstream source, fetched at query time, with the citation printed on the layer. When a source has nothing to say about the area you asked about, the map says so rather than filling the gap. The full list of sources and their licences is public at /data-sources.
That rule is stricter than it sounds. A model can propose what to look for — which datasets might be relevant, which places match "the Pacific Northwest" — but it never supplies the answer. Coordinates come from gazetteer verification; values come from fetched sources with a citation attached. Anything that cannot be verified is dropped rather than guessed.
What the rule costs
Features, mostly. A tool that is allowed to estimate can answer anything; a tool that must fetch can only answer where a real source exists. There are questions ShowMeOnMap declines that a general-purpose chatbot will happily answer with plausible-looking numbers. We think the trade is right: a map is usually the input to a decision, and a confident wrong number on a map is worse than no map.
It also costs speed. Fetching at query time means the map is only as fast as the slowest source it needs. Caching softens this, but the floor is set by the upstream, not by us.
What it buys
Two things a generated map normally cannot offer.
A distinction between "zero" and "no data". A source whose footprint does not reach the area you asked about returns nothing — and reporting that as "nothing found" would invent an all-clear. ShowMeOnMap keeps those as different statements: an empty area either means the source says it is empty, or it means the source does not cover it, and the map tells you which.
A citation someone else can check. Every layer names its source, and the data-sources page records each source's licence. If you put one of these maps in a report or an article, the provenance travels with it — including through the embed code, which carries the citation in the caption.
Common questions
Does the AI ever fill in missing values? No. The model's role is selection and planning — which source, which area, which rendering. Values come only from fetched payloads. If the fetch fails, the map reports the failure instead of a substitute number.
Where can I see every source you use? /data-sources — generated from the same registry the app itself reads, so it cannot drift out of date independently of the code. Sources whose licence review is still pending are listed as such rather than hidden.
What happens when a source is down? The layer renders as unavailable, with the source named. Live pages like the BC wildfire tracker do the same: an upstream outage shows as an outage, never as stale numbers presented as current.