Here is a habit worth being suspicious of: a survey that grades its own homework. If the accuracy of a map is tested only against the same ground marks that were used to build it, the test mostly confirms that the software did what software does.
It says very little about the ground between the marks — which is exactly where your engineer is going to take measurements.
What a blind check actually measures
We place the ground marks as usual, spread across the whole site. Then we split them: most go into the processing as control, and the rest are measured to exactly the same standard but never shown to the software.
After processing, we compare the finished map against those withheld points. That comparison captures everything the standard check misses — the small errors that build up between marks, and any systematic warp in the model.
The two numbers differ more than people expect
On a well-run project the difference is modest but real. Where the gap becomes large, it is a warning that something in the survey needs attention — usually the distribution of marks, occasionally a setup error — and it is far better to catch that before delivery than after.
The cost
Withholding marks means measuring a few more of them, which adds roughly an hour of ground work on a typical site. It is the cheapest hour on the whole job, because it is the one that turns a delivery into a defensible delivery.