Buy the beta
Accuracy check

Does Scan Bench add any error?

A scan is only as accurate as the scanner that made it. What we can test is whether Scan Bench keeps that accuracy when it levels, turns and measures the scan. Here is how we checked, and what we found.

The test

We made a scan of a room where every size is known exactly: 6.000 m by 4.500 m, ceiling at 2.700 m, a door opening, and a cone shaped heap on the floor (1.5 m radius, 0.8 m high, so 1.885 m³). Then we tilted it 2.5° one way and 1.2° the other, turned it 31° and moved it, the way a scanner leaves a scan. We made two copies: one with 2 mm of noise (a good laser scanner) and one with 10 mm (a handheld or phone scanner). Each copy was opened in Scan Bench. The height and the volume were read with its own tools; the lengths were read between the wall surfaces of the levelled scan.

What was measuredTrue2 mm noise10 mm noise
Room length (wall to wall)6.000 m6.000 m6.000 m (0.4 mm short)
Room width (wall to wall)4.500 m4.500 m4.500 m (0.2 mm short)
Ceiling height (Quantities, auto)2.700 m2.699 m2.698 m
Floor left after levelling0° tilt0.04°0.02°
Walls turned square to the plan31°30.95°31.00°
Heap volume (Volume, stockpile)1.885 m³1.881 m³ (0.2% under)1.882 m³ (0.2% under)

What this shows

  • Levelling, turning and measuring do not add error you would notice on site.
  • Lengths stay within half a millimetre.
  • Volumes come out within 0.2% on a clean heap.

What it does not show

  • How accurate your scanner is. Check the maker's published figures.
  • A real site is messier than a test room: plants, glass, people and missing patches all cost accuracy.
  • Splat scans from a phone app have no real size until you set the scale, so their accuracy depends on the length you give it.
  • Clicking a point by hand adds your own error, so zoom in and use snapping.

We fixed two things on the way

The test found that a volume measured indoors counted the ceiling, so heaps under a roof read high; the ceiling is now left out. It also found the automatic ceiling height read about 15 mm low on noisy scans; it now uses the middle of the floor and ceiling readings. Both fixes are in build r510.

Want to try it yourself? Ask for the test scan and open it in Scan Bench. Test run 20 October 2026, build r510.