If that is a problem, then let prescribe the following process flow by the app: first sound the empty tube, then the tube filled with water, and only then the tube filled with the test liquid. The first two checks should catch any miscalibrations, I think.

Hence why it wasn't proposed to take one reading, but instead one of a known good control substance and one of the test substance. It doesn't matter if it's tuned properly or not, as you aren't looking for a specific frequency, but instead a difference in the two frequencies.