Two coin presses, A and B, operate side-by-side, each producing
100,000 coins per shift. Both presses are nominally calibrated to the same design weight, but a quality manager wants to detect drift between them by comparing total batch weights: she will estimate
T^A−T^B from independent random samples drawn from each press, and check whether the result differs from zero by more than
0.1% of one batch's nominal total. Each coin from either press has an independent weight uncertainty of
1% of the design weight; assume no sources of systematic uncertainty. About how many coins must be sampled and weighed
from each press to reach this
0.1% uncertainty on the difference?