The map only drew what it was sure of

The map used to draw the streets it had proved you could walk, and nothing else at all.

That sounds like the careful choice. It is the opposite of one. A map that silently omits whatever it is unsure of is not a cautious map, it is a confident wrong one, because the reader has no way to tell there is nothing here from I did not check here. Both are drawn as blank paper.

Drive past the last certified street and your marker sailed off the edge into that blank. In some places it happened a few seconds after arriving.

Lourmarin was the clearest case. The village is terraced, and the route-finding only believed in a step it could also climb back up, so everything below a terrace edge was left off. That came to five sixths of the walkable village. You could walk down there perfectly well. You could step off a ledge you could not climb, and the game would play the fall out and put you on the street below. You simply could not be shown to be there.

Ludlow had the strangest version of it. Its map was a single bent strip of a square you can drive end to end, because the map was reading the scanned surface bump by bump while the car drives a road that ignores those bumps. The square is also ringed with arches, and to a reading taken from directly above, an arch is a wall rather than a street passing underneath one.

Maps now draw the whole captured place: the terrain as a quiet silhouette, the buildings a shade brighter, and the streets you actually follow on top of both. The parts that are only context look like context, which is the thing a blank space could never do.

I do not have a tidy way to say the general lesson. Something like: the confidence of a drawing should match the confidence of the data under it, and the usual failure is not overclaiming. It is dropping everything below a threshold and presenting what is left as the whole truth.