Essay · February 9, 2026
The map is a decision record
Ask a GIS analyst what they make and they will say maps. Watch what their institution actually does with the output and you'll see something else: the map is consulted at the moment of decision, then filed as the justification for it. Ten years later, when the decision is challenged, nobody asks whether the map was beautiful. They ask what it showed, when it showed it, and who knew.
A map, in an institution, is a decision record. Most GIS practice (and nearly all spatial AI practice) is built as if this weren't true.
Cartography for the deposition
This sounds cynical and is the opposite. Designing a spatial product as a decision record raises the standard. Every layer needs a lineage: where the boundary came from, which ordinance or survey it encodes, when it was last reconciled against its source. Symbology becomes a claim about certainty, not just a style choice: a dashed line should mean something legally different from a solid one.
What this means for spatial AI
The current generation of AI tooling is very good at producing confident spatial statements and very bad at producing accountable ones. If the map is a decision record, then an AI layer over the map inherits the obligation: answers must carry their provenance, disagreement between sources must surface rather than average away, and "the system doesn't know" must be an expressible, first-class answer.
We think the institutions that get spatial AI right will be the ones that never stopped treating their maps as evidence. The technology is new. The obligation is very old.