Lourmarin is a village in the Luberon. The capture runs about two hundred metres of it one way and three hundred the other, with stone going up thirty metres on either side of lanes that mostly leave a car a few metres of room. It went in as the fourth place you could drive, and for two days it was the fourth place you could not.

The car arrived, settled, and would not move. Not wedged against a wall, not sunk into the ground, not upside down. It sat level on clean paving with an open lane in front of it and refused to go anywhere.

Everything I had said it was fine. The patch under the spawn was flat to a few centimetres. The height reading agreed with the visible street. The car had room in front of it for tens of metres. The physics built without a complaint.

What was actually there was two triangles. Left over from the scan, hanging in the air about a metre and a half ahead of where the car starts, roughly the size of a postcard, at exactly the height of the bumper.

This is the failure mode I keep meeting with captured places, and it is worth naming because it is not the one you brace for. Authored geometry goes wrong in large obvious ways: a wall in the wrong place, a floor missing, a door that opens into rock. A scan goes wrong in specks.

And a speck is invisible to every check that averages, which is nearly every check worth having. How flat is this patch. How high is the ground here. How far can the car get from this spot. All three are the right questions and all three are answered by looking at a region, so something the size of a postcard contributes almost nothing to any of them and stops a car dead.

The fix was not clever. Colliders now throw away fragments below a size, because a fragment that small is never architecture. It is reconstruction noise that happens to be solid.

What I took from it: when a place will not behave, the first question is not what is wrong with the physics. It is what is in the scan that I cannot see. Twice now the answer has been something too small to look at.