5 thousandths, or .12 mm - is enough to cause trouble. Depth of cut is .12mm for circuits.
See Shapeoko discussion - [1]
Latest Shapeoko uses GT2 9 mm
Most RepRaps use GT2 6 mm

Calculations

Positioning accuracy of GT2 belt - Say 1/2" pulley. Circumference about 40 mm. 200 steps per revolution = .2 mm positioning - but with 16 microstepping RAMPS standard - we get down to 0.01 mm positioning accuracy, or 0.5 thousandths. Positioning Accuracy does not stand in the way of a successful build - therefore only backlash and belt stretching need to be resolved. For belt stretching, the solution is proper tensioning. So this looks optimistic, gien a solid frame."
Chart at [2] shows positioning accuracy of GT2 belt to be 0.0003 - about half that of the RAMPS 16 microstepping calculation above - meaning that GT2s can accept up to 1/32 microstepping.
The question is - how much does the belt stretch upon acceleration (force)?
Why is there no resource that states belt stretch as a function of force?

SMEs (Subject Matter Experts

Contact Application Engineering Department - of SDP-SI - this belt bible - [3]- Tel: 516-328-3300

Ebay - 15 mm belt - $1/foot at 100M quantity - [4]. 500 lb per inch breaking strength. Operate at 10% of breaking strength?
SDP-SI has up to 9 mm belts - [5]
$40 for 10m of 9 mm - [6]
$32 Amazon prime for 9mm, 10m - [7]